Printing process optimization decision-making system based on big data

Abstract

The invention belongs to the technical field of printing process optimization, and discloses a printing process optimization decision-making system based on big data. Comprising an image acquisition module used for acquiring a printed product image; the feature extraction module is used for analyzing the printed product image and extracting product feature data; the quality detection module is used for analyzing the product feature data and identifying quality defect problems; the reason tracing module traces the reason of the quality defect according to the quality defect problem; the process optimization module is used for dynamically optimizing printing process parameters based on quality defect reasons; according to the method, the advantages of advanced technologies such as computer vision, machine learning and intelligent optimization can be fully played, intelligent monitoring and dynamic optimization of the whole printing production process are achieved, the method can rapidly adapt to the changing production environment, the quality defect problem in printed products is reduced, and the printing quality and the production efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of printing process optimization, and more specifically, to a printing process optimization decision-making system based on big data. Background Art

[0002] With the rapid development of modern manufacturing, especially in the printing industry, the market competition is becoming increasingly fierce, and enterprises' requirements for defect coefficients and efficiency are constantly increasing; traditional printing process optimization mainly relies on experience and manual adjustment, which is often limited by the skills and experience of operators, and has low efficiency and cannot quickly adapt to changing production environments and complex process parameters; therefore, how to improve production efficiency and reduce costs while ensuring defect coefficients has become an urgent problem to be solved in the printing industry.

[0003] With the rapid development of technologies such as big data, artificial intelligence, and the Internet of Things, the production process in the printing industry is gradually transforming towards intelligence and digitization; through the collection and analysis of a large amount of production data, more accurate decision-making support for process optimization can be obtained; for example, the patent with the publication number CN118228551 B discloses a method and system for optimizing printing process parameters; including: screening non-uniformity features of the same clustering for historical printing image data to obtain non-uniformity clustering data; performing three-dimensional correlation merging on the historical printing process data according to the non-uniformity clustering data to obtain three-dimensional merged printing data; constructing an information gain ratio decision for the three-dimensional merged printing data and wireless radio frequency historical data according to radio frequency index data to obtain a radio frequency printing decision model; constructing a secondary information gain ratio decision for the three-dimensional merged printing data according to printing process index data to obtain an energy consumption reduction decision model; performing gradient boosting integration on the energy consumption reduction decision model and the radio frequency printing decision model to obtain a printing process parameter optimization decision model; realizing intelligent and rapid decision-making optimization of printing process parameters.

[0004] However, although the above technology constructs a printing process parameter optimization model through clustering and information gain ratio decision-making to achieve intelligent and rapid decision-making optimization, it mainly relies on historical data and feature screening, lacking a dynamic analysis and feedback mechanism for real-time printing process data; in the actual printing process, the relationship between printing process parameters and quality defects will change in a complex and ever-changing printing environment, and traditional models often cannot respond to these changes in a timely manner; therefore, the above technology has significant deficiencies in terms of real-time performance and adaptability, especially when sudden changes occur during the printing process, it is difficult to quickly adjust and optimize parameters.

[0005] In view of this, the present invention proposes a printing process optimization decision-making system based on big data to solve the above problems. Summary of the Invention

[0006] To overcome the above defects of the prior art and to achieve the above object, the present invention provides the following technical solution: A printing process optimization decision-making system based on big data, comprising:

[0007] An image acquisition module for acquiring printing product images;

[0008] A feature extraction module for analyzing the printing product images and extracting product feature data;

[0009] A quality inspection module for analyzing the product feature data and identifying quality defect problems;

[0010] A cause tracing module for tracing the cause of quality defects according to the quality defect problems;

[0011] A process optimization module for dynamically optimizing printing process parameters based on the cause of quality defects.

[0012] Furthermore, the product feature data includes color values, sharpness values, the number of scratches, the number of stains, grayscale values, and printing areas; the color value is the color representation of each pixel point in the printing product image; the sharpness value is the sharpness degree of each pixel point in the printing product image; the number of scratches is the number of scratches in the printing product image; the number of stains is the number of stains in the printing product image; the grayscale value is the brightness value of each pixel point in the printing product image; the printing area is the area of the actually printed region in the printing product image;

[0013] The method for obtaining color values is as follows: Using a color space conversion method, convert the printing product image from the RGB color space to the Lab color space; According to the corresponding Lab color space of the printing product image, obtain the color value corresponding to each pixel point in the printing product image; The method for obtaining sharpness values is as follows: Define a convolution kernel, and the convolution kernel includes a horizontal convolution kernel and a vertical convolution kernel; Perform convolution operations on each pixel point in the printing product image according to the convolution kernel to obtain the horizontal gradient value and the vertical gradient value corresponding to each pixel point; According to the horizontal gradient value and the vertical gradient value, calculate the sharpness value corresponding to each pixel point;

[0014] The method for obtaining the number of scratches is as follows: Using a trained scratch detection model, identify the printing product image and output the number of scratches; The method for obtaining the number of stains is as follows: Using a trained stain detection model, identify the printing product image and output the number of stains; The training process of the stain detection model is the same as that of the scratch detection model, and both are convolutional neural network models;

[0015] The method for obtaining the grayscale value is as follows: according to the RGB color space corresponding to the printed product image, obtain the color data corresponding to each pixel point in the printed product image; calculate the grayscale value corresponding to each pixel point according to the color data; the method for obtaining the printed area is as follows: preset a grayscale threshold, compare the grayscale value of each pixel point with the grayscale threshold respectively, mark the pixel points with grayscale values less than the grayscale threshold as printed points, and do not mark the pixel points with grayscale values greater than or equal to the grayscale threshold; count the number of printed points and mark it as the printed quantity; obtain the point area, where the point area is the area of a pixel point in the printed product image; multiply the point area by the printed quantity to obtain the printed area.

[0016] Further, the color value includes brightness, red-green axis, and blue-yellow axis;

[0017] The expression of the horizontal convolution kernel is: In the formula, G x is the horizontal convolution kernel;

[0018] The expression of the vertical convolution kernel is: In the formula, G y is the vertical convolution kernel;

[0019] The expression of the sharpness value is: In the formula, G is the sharpness value, is the horizontal gradient value, is the vertical gradient value;

[0020] The training process of the scratch detection model includes:

[0021] Pre-collect a printed product images, where a is an integer greater than 1, label each printed product image, and label it as the number of scratches; divide the labeled printed product images into a training set and a test set; use the training set to train the scratch detection model, and use the test set to test the scratch detection model; preset an error threshold, and when the mean value of the prediction errors of all printed product images in the test set is less than the error threshold, output the scratch detection model;

[0022] The color data includes red channel value, green channel value, and blue channel value;

[0023] The expression of the grayscale value is: gray = 0.2989×R + 0.578×G + 0.114×B; in the formula, gray is the grayscale value, R is the red channel value, G is the green channel value, and B is the blue channel value.

[0024] Further, the method for identifying quality defect problems includes:

[0025] Obtain a standard product image, and obtain the color value, grayscale value, and printing area corresponding to the standard product image; mark the color value corresponding to the standard product image as the first color value, and mark the color value corresponding to the printed product image as the second color value; subtract the first color value corresponding to each pixel point from the second color value to obtain the color difference of each pixel point; mark the grayscale value corresponding to the standard product image as the first grayscale value, and mark the grayscale value corresponding to the printed product image as the second grayscale value; subtract the first grayscale value corresponding to each pixel point from the second grayscale value to obtain the grayscale difference of each pixel point; mark the printing area corresponding to the standard product image as the first area, and mark the printing area corresponding to the printed product image as the second area; subtract the first area from the second area to obtain the area difference;

[0026] Take the sharpness value, the number of scratches, the number of stains, the color difference, the grayscale difference, and the area difference as analysis data, and input the analysis data into the trained quality detection model to predict the corresponding detection result; the detection result is a set label, and the set label is the digital label corresponding to the pre-constructed problem set, and the digital labels corresponding to different problem sets are different; the method for pre-constructing the problem set is: obtain all quality defect problems; count the number of quality defect problems and mark it as the number of problems; combine all quality defect problems, and randomly select h quality defect problems each time during the combination process, h ∈ [0, H], where H is the number of problems; take the quality defect problems combined each time as a group of problem sets, and a total of D groups of problem sets are obtained, D = 2 H ; The D groups of problem sets are all different, and the D groups of problem sets cover all combination situations of quality defect problems; according to the predicted set label, obtain the quality defect problems in the corresponding problem set.

[0027] Further, the steps of tracing the cause of quality defects include:

[0028] Step S101: Convert all quality defect problems into vector form and mark them as problem vectors;

[0029] Step S102: Take each problem vector as a sample point, and the sample point corresponds to the problem vector one by one; preset the number of categories g, randomly select g sample points as the center points, sequentially increment the digital label for each center point, and mark it as ωk, k ∈ [1, g];

[0030] Step S103: Mark the sample points that are not used as center points as classification points, sequentially increment the digital label for each center point, and mark it as ψc, c ∈ [1, H - g];

[0031] Step S104: Calculate the point distance from each classification point to each center point in turn;

[0032] Step S105: Establish g corresponding quality defect causes based on g center points;

[0033] Step S106: Assign the classification point ψc to the corresponding quality defect cause;

[0034] Step S107: Let c = c + 1, and jump back to Step S106;

[0035] Step S108: Loop Steps S106 to S107 until c = H - g, then the loop ends and enter Step S109;

[0036] Step S109: Recalculate the new center point corresponding to each quality defect cause;

[0037] Step S110: Loop Steps S104 to S109 until the new center points of each quality defect cause recalculated in Step S109 are the same as the new center points calculated in the previous loop, then the loop ends, and obtain the quality defect problems corresponding to the sample points of k quality defect causes;

[0038] Step S111: Mark the identified quality defect problem as a real-time problem, compare the real-time problem with the quality defect problems corresponding to each quality defect cause, obtain the quality defect cause corresponding to the quality defect problem that is the same as the real-time problem, and use it as the traced quality defect cause.

[0039] Further, in Step S104, the expression of the point distance is: In the formula, D ck is the point distance from the classification point ψc to the center point ωk, ω k is the problem vector corresponding to the center point ωk, ψ c is the problem vector corresponding to the classification point ψc;

[0040] In Step S106, the method of assigning the classification point ψc to the corresponding quality defect cause is: preset a distance threshold, compare the point distance from the classification point ψc to each center point with the distance threshold respectively, mark the center point with a point distance less than the distance threshold as the assigned point, and do not mark the center point with a point distance greater than or equal to the distance threshold; assign the classification point ψc to the quality defect cause corresponding to the assigned point;

[0041] In Step S109, the calculation method of the new center point of each quality defect cause includes:

[0042]

[0043] In the formula, ω′ k is the new center point corresponding to the kth quality defect cause, ψ kris the r-th classification point among the k-th quality defect causes, R k is the number of classification points among the k-th quality defect causes, r ∈ [1, R k .

[0044] Further, the steps of dynamically optimizing the printing process parameters include:

[0045] Step S201: Determine the printing process parameters based on the quality defect causes;

[0046] Step S202: Collect the printing process parameters in real time and mark them as real-time process parameters;

[0047] Step S203: Perform anomaly detection on the real-time process parameters to obtain the abnormal process parameters;

[0048] Step S204: Obtain the process parameter range corresponding to the abnormal process parameters;

[0049] Step S205: Based on the process parameter range, construct M groups of parameter sets and set different parameter labels for each group of parameter sets. The parameter labels are digital labels, and the range of the parameter labels is [1, M];

[0050] Step S206: Randomly select a parameter label as the valid solution;

[0051] Step S207: Update the valid solution and determine whether the valid solution applies the updated value;

[0052] Step S208: Adjust the preset step factor and update the valid solution according to the adjusted step factor, and determine whether the valid solution applies the updated value;

[0053] Step S209: Determine the search stage and adjust the preset perturbation factor;

[0054] Step S210: Update the valid solution according to the adjusted perturbation factor and determine whether the valid solution applies the updated value;

[0055] Step S211: Calculate the defect coefficients of the valid solution in the current iteration process and the valid solution in the previous iteration process respectively, and calculate the defect difference;

[0056] Step S212: Loop steps S207 to S211 until the defect difference is less than the preset difference threshold, then the loop ends and enter step S213;

[0057] Step S213: Calculate the mean difference;

[0058] Step S214: Loop through steps S206 to S213 until the mean difference is less than the preset difference threshold or the defect coefficient corresponding to the valid solution is 0, at which point the loop ends; obtain the parameter tags corresponding to the valid solution and mark them as valid tags; dynamically optimize the printing process parameters according to the parameter set corresponding to the valid tags.

[0059] Further, in the step S201, the method for determining the printing process parameters includes:

[0060] Obtain the historical adjustment set, which includes the reasons for quality defects identified at historical times and the printing process parameters corresponding to the dynamic optimization; use the association rule learning method to analyze the historical adjustment set to obtain the parameter corresponding set, which includes different reasons for quality defects and the printing process parameters corresponding to each reason for quality defect; according to the traced reason for quality defect, obtain the corresponding printing process parameters from the parameter corresponding set;

[0061] In the step S203, the method for obtaining the abnormal process parameters includes:

[0062] Obtain the historical process parameters, which are the printing process parameters collected at historical times; take each same parameter in the historical process parameters as a calculation set, and the calculation set corresponds one by one to the parameters in the historical process parameters; calculate the mean and sample standard deviation corresponding to each calculation set; subtract the mean corresponding to each calculation set from each parameter in the real-time process parameters, and then divide by the standard deviation corresponding to the corresponding calculation set to obtain the abnormal coefficient corresponding to each parameter in the real-time process parameters; preset the coefficient range [-u, u]; if -u ≤ YC v ≤ u, then do not mark the corresponding parameter; if -u > YC v ∪ YC v > u, then mark the corresponding parameter as an abnormal parameter; where YC v is the abnormal coefficient corresponding to the v-th parameter in the real-time process parameters, and ∪ represents or; take all the abnormal parameters as the abnormal process parameters;

[0063] In the step S204, the method for obtaining the process parameter range corresponding to the abnormal process parameters includes:

[0064] Add u times the corresponding standard deviation to the mean corresponding to each abnormal parameter to obtain the maximum value of each abnormal parameter; subtract u times the corresponding standard deviation from the mean corresponding to each abnormal parameter to obtain the minimum value of each abnormal parameter; construct the parameter range corresponding to each abnormal parameter according to the maximum and minimum values of each abnormal parameter, and take all the parameter ranges as the process parameter range.

[0065] Further, in the step S205, the method for constructing M groups of parameter sets includes:

[0066] Randomly select a value from each parameter range within the process parameter range, and construct a set of parameter sets. A total of M sets of parameter sets are constructed, and the M sets of parameter sets are all different;

[0067] The method for updating the effective solution in step S207 includes:

[0068]

[0069] In the formula, is the updated effective solution, C1 ∈ [0, 1], N(0, 1) is a random number in the standard normal distribution, and x is the effective solution;

[0070] The method for determining whether the effective solution applies the updated value includes:

[0071] Mark the updated effective solution as the updated solution, calculate the defect coefficients corresponding to the updated solution and the effective solution respectively, mark the defect coefficient corresponding to the updated solution as the first coefficient, and mark the defect coefficient corresponding to the effective solution as the second coefficient; compare the first coefficient with the second coefficient; if the first coefficient is greater than the second coefficient, the effective solution remains the value before the update; if the first coefficient is less than or equal to the second coefficient, the effective solution applies the updated value; the calculation method of the defect coefficient is: collect all parameters in the printing process in real time and mark them as process comprehensive parameters; obtain the parameter tags corresponding to the effective solution, replace the abnormal process parameters in the process comprehensive parameters with the parameter sets corresponding to the parameter tags, use the replaced process comprehensive parameters as test data, input the test data into the trained defect analysis model, and predict the corresponding defect coefficient; the training process of the defect analysis model is the same as that of the quality inspection model, and both are deep neural network models;

[0072] In step S208, the method for adjusting the preset step factor includes:

[0073] If the effective solution in step S207 applies the updated value, then In the formula, is the adjusted step factor, p is the preset step factor, γ is the adjustment factor, and sig(f) is the improvement factor, f old is the second coefficient, f new is the first coefficient;

[0074] If the effective solution in step S207 does not apply the updated value, then In the formula, pen(f) is the penalty factor,

[0075] The method for updating the effective solution according to the adjusted step factor includes:

[0076]

[0077] wherein, is the effective solution updated according to the adjusted step factor, x' is the effective solution after performing step S207, and x best is the effective solution with the smallest defect coefficient in the iterative process, and C2 ∈ [0, 1];

[0078] The method for determining whether to apply the updated value to the effective solution is the same as the method in step S207.

[0079] Furthermore, in step S209, the method for determining the search stage includes:

[0080] The search stage includes a pre-stage and a post-stage;

[0081] Count the number of times of loop steps S207 to S211 and mark it as the iteration number; preset a number threshold, and compare the iteration number with the number threshold; if the iteration number is greater than the number threshold, determine that the exploration stage is the pre-stage; if the iteration number is less than the number threshold, determine that the exploration stage is the post-stage;

[0082] The method for adjusting the preset perturbation factor includes:

[0083] If the exploration stage is the pre-stage, then wherein, is the adjusted perturbation factor, q is the preset perturbation factor, e is the natural constant, and t is the iteration number;

[0084] If the exploration stage is the post-stage, then wherein, T is the number threshold;

[0085] In step S210, the method for updating the effective solution according to the adjusted perturbation factor includes:

[0086]

[0087] wherein, is the effective solution updated according to the adjusted perturbation factor, x″ is the effective solution after performing step S208, and f best is the defect coefficient corresponding to the effective solution with the smallest defect coefficient in the iterative process, and f x″ is the defect coefficient corresponding to the effective solution after performing step S208;

[0088] The method for determining whether to apply the updated value to the effective solution is the same as the method in step S207;

[0089] In the step S211, the calculation method of the defect difference is as follows: subtract the defect coefficient corresponding to the effective solution in the previous iteration process from the defect coefficient corresponding to the effective solution in the current iteration process, and take the absolute value to obtain the defect difference;

[0090] In the step S213, the method for calculating the mean difference is as follows: add up the defect differences calculated in the current iteration process in sequence, then divide by the number of iterations to obtain the defect mean; calculate the defect mean in the previous iteration process, subtract the defect mean corresponding to the previous iteration process from the defect mean corresponding to the current iteration process, and take the absolute value to obtain the mean difference.

[0091] The technical effects and advantages of a printing process optimization decision-making system based on big data according to the present invention:

[0092] By collecting printing product images and using computer vision technology to extract various product feature data, the quality status of printing products can be comprehensively reflected; based on the product feature data, deep learning technology is used to quickly identify various quality defect problems, and the specific quality defect reasons can be traced through cluster analysis; an optimization strategy based on an optimization algorithm is adopted to dynamically optimize abnormal process parameters, thereby reducing quality defect problems in printing products and improving printing quality and production efficiency; giving full play to the advantages of cutting-edge technologies such as computer vision, machine learning, and intelligent optimization, realizing intelligent monitoring and dynamic optimization of the entire printing production process, so as to quickly adapt to changing production environments and improve the overall quality and market competitiveness of printing products. Brief Description of the Drawings

[0093] Figure 1 It is a schematic diagram of a printing process optimization decision-making system based on big data according to Embodiment 1 of the present invention;

[0094] Figure 2 It is a flowchart of a method for dynamically optimizing printing process parameters according to Embodiment 1 of the present invention. Detailed Embodiments

[0095] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0096] Embodiment 1

[0097] Please refer to Figure 1As shown in the figure, the optimization decision-making system for printing processes based on big data in this embodiment includes an image acquisition module, a feature extraction module, a quality detection module, a cause tracing module, and a process optimization module; each module is connected by wired and / or wireless means to achieve data transmission between modules.

[0098] The image acquisition module is used to acquire images of printed products.

[0099] The image of the printed product is the visual representation of the product generated after the printing process; the image of the printed product is obtained through an image sensor installed on the printing production line.

[0100] The feature extraction module is used to analyze the image of the printed product and extract product feature data.

[0101] The product feature data includes color values, sharpness values, the number of scratches, the number of stains, grayscale values, and printing areas.

[0102] The color value is the color representation of each pixel point in the image of the printed product, reflecting the color characteristics of the image of the printed product, so as to judge whether there is a color deviation in the printed product. The reasons for causing color deviation are, for example, inaccurate ink ratio, too fast printing speed, etc.; the method for obtaining the color value is: using a color space conversion method to convert the image of the printed product from the RGB color space to the Lab color space; among them, the color space conversion method is a prior art, so it will not be elaborated here; according to the Lab color space corresponding to the image of the printed product, obtain the color value corresponding to each pixel point in the image of the printed product; the color value includes brightness, red-green axis, and blue-yellow axis; among them, the red-green axis represents the chromaticity difference of the color corresponding to the pixel point between green and red, that is, the degree of green or red deviation of the color; the blue-yellow axis represents the chromaticity difference of the color corresponding to the pixel point between blue and yellow, that is, the degree of blue or yellow deviation of the color.

[0103] The sharpness value is the degree of sharpness of each pixel point in the image of the printed product, reflecting the clarity of the image of the printed product, so as to judge whether there is printing blurring in the printed product. The reasons for causing printing blurring are, for example, focusing problems, insufficient ink drying, etc.; the printed image is the image printed on the printed product; the method for obtaining the sharpness value is: defining a convolution kernel, and the convolution kernel includes a horizontal convolution kernel and a vertical convolution kernel; perform a convolution operation on each pixel point in the image of the printed product according to the convolution kernel to obtain the corresponding horizontal gradient value and vertical gradient value of each pixel point; among them, the convolution operation is a prior art, so it will not be elaborated here; calculate the sharpness value corresponding to each pixel point according to the horizontal gradient value and the vertical gradient value.

[0104] The expression of the horizontal convolution kernel is: In the formula, G x is the horizontal convolution kernel;

[0105] The expression of the vertical convolution kernel is as follows: In the formula, G y is the vertical convolution kernel;

[0106] The expression of the sharpness value is as follows: In the formula, G is the sharpness value, is the horizontal gradient value, is the vertical gradient value.

[0107] The number of scratches is the number of scratches in the printed product image, which reflects the surface damage of the printed product image, so as to judge whether there is mechanical damage to the printed product. The reasons for mechanical damage are, for example, machine failure, unevenness of the printed product, etc.; the method for obtaining the number of scratches is: using the trained scratch detection model to identify the printed product image and output the number of scratches.

[0108] The specific training process of the scratch detection model includes:

[0109] Pre-collect a printed product images, where a is an integer greater than 1. Annotate each printed product image with the number of scratches; divide the annotated printed product images into a training set and a test set. Take 70% of the printed product images as the training set and 30% of the printed product images as the test set; use the training set to train the scratch detection model and use the test set to test the scratch detection model; preset an error threshold. When the mean value of the prediction errors of all printed product images in the test set is less than the error threshold, output the scratch detection model; among them, the calculation formula for the mean value of the prediction error is where Z is the prediction error, b is the number of the printed product image, and α b is the predicted annotation corresponding to the b-th printed product image, and μ b is the actual annotation corresponding to the b-th printed product image, and A is the number of printed product images in the test set; the error threshold is preset according to the required accuracy of the scratch detection model; the scratch detection model is specifically a convolutional neural network model.

[0110] The number of stains is the number of stains in the printed product image, which reflects the surface stain situation of the printed product image, so as to judge whether there is surface contamination of the printed product. The reasons for surface contamination are, for example, ink overflow, dust, etc.; the method for obtaining the number of stains is: using the trained stain detection model to identify the printed product image and output the number of stains; the training process of the stain detection model is the same as that of the scratch detection model, and both are convolutional neural network models.

[0111] The gray value is the brightness value of each pixel in the printed product image, which reflects the brightness distribution of the printed product image, so as to judge whether there is ink shortage in the printed product. The reasons for ink shortage include, for example, insufficient ink concentration and uneven ink flow. The method for obtaining the gray value is as follows: According to the RGB color space corresponding to the printed product image, obtain the color data corresponding to each pixel in the printed product image. The color data includes the red channel value, the green channel value, and the blue channel value. Calculate the gray value corresponding to each pixel according to the color data. The expression of the gray value is: gray = 0.2989×R + 0.578×G + 0.114×B. In the formula, gray is the gray value, R is the red channel value, G is the green channel value, and B is the blue channel value.

[0112] The printed area is the area of the actually printed area in the printed product image, which reflects the printing coverage of the printed product image, so as to judge whether there is missing printing in the printed product. The reasons for missing printing include, for example, uneven printing pressure and insufficient ink concentration. The method for obtaining the printed area is as follows: Preset a gray threshold, which is preset by those skilled in the art according to the actual situation. Compare the gray value of each pixel with the gray threshold respectively, mark the pixel with a gray value less than the gray threshold as a printed point, and do not mark the pixel with a gray value greater than or equal to the gray threshold. Count the number of printed points and mark it as the printed quantity. Obtain the point area, which is the area of a pixel in the printed product image, and the point area is obtained by those skilled in the art according to the resolution of the printed product image. Multiply the point area by the printed quantity to obtain the printed area.

[0113] The quality inspection module is used to analyze the product feature data and identify quality defect problems.

[0114] The methods for identifying quality defect problems include:

[0115] Obtain a standard product image, which is the image of a standard printed product. The standard printed product is an ideal product that meets the requirements of predetermined design, quality, process, etc. during the printing process. Obtain the color value, gray value, and printed area corresponding to the standard product image. Mark the color value corresponding to the standard product image as the first color value, and mark the color value corresponding to the printed product image as the second color value. Subtract the first color value corresponding to each pixel from the second color value to obtain the color difference of each pixel. Mark the gray value corresponding to the standard product image as the first gray value, and mark the gray value corresponding to the printed product image as the second gray value. Subtract the second gray value corresponding to each pixel from the second gray value to obtain the gray difference of each pixel. Mark the printed area corresponding to the standard product image as the first area, and mark the printed area corresponding to the printed product image as the second area. Subtract the first area from the second area to obtain the area difference.

[0116] Take the sharpness value, the number of scratches, the number of stains, the color difference value, the grayscale difference value, and the area difference value as analysis data, and input the analysis data into the trained quality detection model to predict the corresponding detection result; the detection result is a set label, and the set label is the digital label corresponding to the pre-constructed problem set, and the digital labels corresponding to different problem sets are different; the method for pre-constructing the problem set is as follows: Obtain all quality defect problems, such as color deviation, printing blur, ink shortage, missing printing, etc.; count the number of quality defect problems and mark it as the number of problems; combine all quality defect problems, and randomly select h quality defect problems each time during the combination process, h ∈ [0, H], where H is the number of problems; take the quality defect problems combined each time as a group of problem sets, and a total of D groups of problem sets are obtained, D = 2 H ; the D groups of problem sets are all different, and the D groups of problem sets cover all combination situations of quality defect problems; according to the predicted set label, obtain the quality defect problems in the corresponding problem set.

[0117] The training process of the quality detection model includes:

[0118] Pre-collect d groups of analysis data, and set corresponding set labels for each of the d groups of analysis data. d is an integer greater than 1. Convert the analysis data and the corresponding set labels into a corresponding set of feature vectors; the set label corresponding to the analysis data is determined by those skilled in the art during the process of historically identifying quality defect problems. Collect d groups of analysis data, and under the conditions of each group of analysis data, analyze each group of analysis data in combination with the actual situation and practical experience, identify the quality defect problems corresponding to each group of analysis data, and set corresponding set labels for each group of analysis data, and sequentially set corresponding set labels for the d groups of analysis data;

[0119] Take each set of feature vectors as the input of the quality detection model. The quality detection model outputs a set of predicted set labels corresponding to each group of analysis data, and takes the actual set label corresponding to each group of analysis data as the prediction target. The actual set label is the set label pre-set corresponding to the analysis data; take minimizing the sum of prediction errors of all analysis data as the training target; among them, the calculation formula for the prediction error is η w =(β w -ε w ) 2 , where η w is the prediction error, w is the group number of the feature vector corresponding to the analysis data, β w is the predicted set label corresponding to the w-th group of analysis data, and ε w is the actual set label corresponding to the w-th group of analysis data; train the quality detection model until the sum of prediction errors reaches convergence and then stop training.

[0120] The above quality inspection model is specifically a deep neural network model, which includes an input layer, a hidden layer, and an output layer. Each hidden layer contains multiple neurons, and there are connections between each neuron and the neurons in the next layer. These connections contain weights that determine the importance and influence of data transmission in the neural network. An activation function is applied to each neuron between the hidden layer and the output layer. The activation function introduces non-linearity and allows the network to learn more complex patterns and features.

[0121] The cause tracing module traces the cause of the quality defect according to the quality defect problem.

[0122] The steps of tracing the cause of the quality defect include:

[0123] Step S101: Adopt word embedding technology (such as Word2Vec, GloVe, etc.) to convert all quality defect problems into vector forms and label them as problem vectors.

[0124] Step S102: Take each problem vector as a sample point, and the sample points correspond to the problem vectors one by one. Preset the number of categories g, which is preset by those skilled in the art according to the actual situation. The number of categories is the total number of quality defect causes. Randomly select g sample points as the center points, and sequentially incrementally set digital labels for each center point and label them as ωk, where k ∈ [1, g].

[0125] Step S103: Mark the sample points that are not used as center points as classification points, and sequentially incrementally set digital labels for each center point and label them as ψc, where c ∈ [1, H - g].

[0126] Step S104: Calculate the point distances from each classification point to each center point in turn. The expression for the point distance is: In the formula, D ck is the point distance from the classification point ψc to the center point ωk, ω k is the problem vector corresponding to the center point ωk, ψ c is the problem vector corresponding to the classification point ψc.

[0127] Step S105: Establish corresponding g quality defect causes based on the g center points.

[0128] Step S106: Assign the classification point ψc to the corresponding quality defect cause.

[0129] Preset a distance threshold, which is preset by those skilled in the art according to the actual situation. Compare the point distances from the classification point ψc to each center point with the distance threshold respectively. Mark the center points with point distances less than the distance threshold as the assigned points, and do not mark the center points with point distances greater than or equal to the distance threshold. Assign the classification point ψc to the quality defect cause corresponding to the assigned point.

[0130] Step S107: Let c = c + 1, and jump back to Step S106;

[0131] Step S108: Loop Steps S106 to S107 until c = H - g, then the loop ends and proceed to Step S109;

[0132] Step S109: Recalculate the new center point for each quality defect cause;

[0133] The calculation method for the new center point of each quality defect cause includes:

[0134]

[0135] In the formula, ω′ k is the new center point corresponding to the kth quality defect cause, ψ kr is the rth classification point in the kth quality defect cause, R k is the number of classification points in the kth quality defect cause, r ∈ [1, R k ;

[0136] Step S110: Loop Steps S104 to S109 until the new center points of each quality defect cause recalculated in Step S109 are the same as those calculated in the previous loop, then the loop ends, and obtain the quality defect problems corresponding to the sample points of the k quality defect causes;

[0137] Step S111: Mark the identified quality defect problems as real-time problems, compare the real-time problems with the quality defect problems corresponding to each quality defect cause, and obtain the quality defect causes corresponding to the quality defect problems that are the same as the real-time problems, and use them as the traced quality defect causes.

[0138] The process optimization module dynamically optimizes the printing process parameters based on the quality defect causes.

[0139] The printing process parameters are, for example: the printing speed corresponding to too fast printing speed, the drying time, drying temperature, etc. corresponding to insufficient ink drying, the ink concentration, ink flow rate, etc. corresponding to insufficient ink concentration.

[0140] As Figure 2 shown, the steps for dynamically optimizing the printing process parameters include:

[0141] Step S201: Determine the printing process parameters based on the quality defect causes;

[0142] Step S202: Real-time collect the printing process parameters and mark them as real-time process parameters;

[0143] Step S203: Perform anomaly detection on real-time process parameters to obtain abnormal process parameters;

[0144] Step S204: Obtain the process parameter range corresponding to the abnormal process parameters;

[0145] Step S205: Based on the process parameter range, construct M sets of parameter collections and set different parameter labels for each set of parameter collections. The parameter labels are digital labels, and the range of the parameter labels is [1, M];

[0146] Step S206: Randomly select a parameter label as the valid solution;

[0147] Step S207: Update the valid solution and determine whether the valid solution applies the updated value;

[0148] Step S208: Adjust the preset step factor and update the valid solution according to the adjusted step factor, and determine whether the valid solution applies the updated value;

[0149] Step S209: Determine the search stage and adjust the preset perturbation factor;

[0150] Step S210: Update the valid solution according to the adjusted perturbation factor and determine whether the valid solution applies the updated value;

[0151] Step S211: Calculate the defect coefficients of the valid solution in the current iteration process and the valid solution in the previous iteration process respectively, and calculate the defect difference;

[0152] Step S212: Loop steps S207 to S211 until the defect difference is less than the preset difference threshold, then the loop ends and enter step S213; The difference threshold is preset by those skilled in the art according to the actual situation;

[0153] Step S213: Calculate the mean difference;

[0154] Step S214: Loop steps S206 to S213 until the mean difference is less than the preset difference threshold or the defect coefficient corresponding to the valid solution is 0, then the loop ends; Obtain the parameter label corresponding to the valid solution and mark it as the valid label; Dynamically optimize the printing process parameters according to the parameter collection corresponding to the valid label; It should be understood that when the defect coefficient corresponding to the valid solution is 0, it means that there are no quality defect problems in the printed products.

[0155] In the above step S201, the method for determining the printing process parameters includes:

[0156] Obtain a historical adjustment set, which includes the reasons for quality defects identified at historical moments and the corresponding printing process parameters for dynamic optimization. The historical adjustment set is obtained through the production management system in the printing production line; use an association rule learning method (such as the Apriori algorithm, Eclat algorithm, etc.) to analyze the historical adjustment set to obtain a parameter corresponding set, which includes different reasons for quality defects and the corresponding printing process parameters for each reason for quality defect; according to the traced reason for quality defect, obtain the corresponding printing process parameters from the parameter corresponding set.

[0157] In the above step S203, the method for obtaining abnormal process parameters includes:

[0158] Obtain historical process parameters, which are the printing process parameters collected at historical moments; take each same parameter in the historical process parameters as a calculation set, and the calculation set corresponds one by one to the parameters in the historical process parameters; calculate the mean value and sample standard deviation corresponding to each calculation set; subtract the mean value corresponding to the corresponding calculation set from each parameter in the real-time process parameters, and then divide by the standard deviation corresponding to the corresponding calculation set to obtain the abnormal coefficient corresponding to each parameter in the real-time process parameters; preset a coefficient range [-u, u], and in this embodiment, it is preferably u = 3; if -u ≤ YC v ≤ u, then do not mark the corresponding parameter; if -u > YC v ∪ YC v > u, then mark the corresponding parameter as an abnormal parameter; where YC v is the abnormal coefficient corresponding to the v-th parameter in the real-time process parameters, and ∪ represents or; take all abnormal parameters as abnormal process parameters.

[0159] In the above step S204, the method for obtaining the process parameter range corresponding to the abnormal process parameters includes:

[0160] Add u times the corresponding standard deviation to the mean value corresponding to each abnormal parameter to obtain the maximum value of each abnormal parameter; subtract u times the corresponding standard deviation from the mean value corresponding to each abnormal parameter to obtain the minimum value of each abnormal parameter; according to the maximum and minimum values of each abnormal parameter, construct the parameter range corresponding to each abnormal parameter, and take all parameter ranges as the process parameter range.

[0161] In the above step S205, the method for constructing M groups of parameter sets includes:

[0162] Randomly select a value from each parameter range within the process parameter range and construct a group of parameter sets, and a total of M groups of parameter sets are constructed, and the M groups of parameter sets are all different.

[0163] In the above step S207: The method for updating the valid solution includes:

[0164]

[0165] Wherein, is the updated effective solution, C1 ∈ [0, 1], N(0, 1) is a random number in the standard normal distribution, and x is the effective solution.

[0166] The method for determining whether to apply the updated value to the effective solution includes:

[0167] Mark the updated effective solution as the updated solution, calculate the defect coefficients corresponding to the updated solution and the effective solution respectively, mark the defect coefficient corresponding to the updated solution as the first coefficient, and mark the defect coefficient corresponding to the effective solution as the second coefficient; compare the first coefficient with the second coefficient; if the first coefficient is greater than the second coefficient, the effective solution remains the value before the update; if the first coefficient is less than or equal to the second coefficient, the effective solution applies the updated value; the calculation method of the defect coefficient is: collect all parameters in the printing process in real time and mark them as process comprehensive parameters; obtain the parameter tags corresponding to the effective solution, replace the abnormal process parameters in the process comprehensive parameters with the parameter sets corresponding to the parameter tags, use the replaced process comprehensive parameters as test data, input the test data into the trained defect analysis model, and predict the corresponding defect coefficient; the training process of the defect analysis model is the same as that of the quality inspection model, and both are deep neural network models; it should be noted that during the training process of the defect analysis model, the defect coefficients corresponding to the test data are obtained by those skilled in the art during the historical printing process parameter optimization process. Collect multiple groups of test data, perform printing under the conditions of each group of test data, collect the analysis data after each printing, perform defect analysis on each printed product based on the analysis data and actual experience, obtain the corresponding defect coefficients, and set the corresponding defect coefficients for each group of test data in turn.

[0168] In the above step S208, the method for adjusting the preset step factor includes:

[0169] If the effective solution in step S207 applies the updated value, then Wherein, is the adjusted step factor, p is the preset step factor, γ is the adjustment factor, both the step factor and the adjustment factor are preset by those skilled in the art according to the actual situation, sig(f) is the improvement factor, f old is the second coefficient, f new is the first coefficient;

[0170] If the effective solution in step S207 does not apply the updated value, then Wherein, pen(f) is the penalty factor,

[0171] The method for updating the effective solution according to the adjusted step factor includes:

[0172]

[0173] In the formula, is the effective solution updated according to the adjusted step factor, x′ is the effective solution after executing step S207, and x best is the effective solution with the smallest defect coefficient in the iteration process, and C2 ∈ [0, 1].

[0174] The method for determining whether to apply the updated value to the effective solution is the same as the method in step S207.

[0175] In the above step S209, the method for determining the search phase includes:

[0176] The search phase includes a pre-stage and a post-stage;

[0177] Count the number of times of loop steps S207 to S211 and mark it as the iteration number; preset a number threshold, which is preset by those skilled in the art according to the actual situation; compare the iteration number with the number threshold; if the iteration number is greater than the number threshold, then determine that the exploration phase is the pre-stage; if the iteration number is less than the number threshold, then determine that the exploration phase is the post-stage.

[0178] The method for adjusting the preset perturbation factor includes:

[0179] If the exploration phase is the pre-stage, then In the formula, is the adjusted perturbation factor, q is the preset perturbation factor, e is the natural constant, and t is the iteration number; the perturbation factor is preset by those skilled in the art according to the actual situation;

[0180] If the exploration phase is the post-stage, then In the formula, T is the number threshold;

[0181] In the above step S210, the method for updating the effective solution according to the adjusted perturbation factor includes:

[0182]

[0183] In the formula, is the effective solution updated according to the adjusted perturbation factor, x″ is the effective solution after executing step S208, and f best is the defect coefficient corresponding to the effective solution with the smallest defect coefficient in the iteration process, and f x″ is the defect coefficient corresponding to the effective solution after executing step S208.

[0184] The method for determining whether to apply the updated value for the valid solution is the same as the method in step S207.

[0185] In the above step S211, the calculation method of the defect difference is as follows: subtract the defect coefficient corresponding to the valid solution in the previous iteration process from the defect coefficient corresponding to the valid solution in the current iteration process, and take the absolute value to obtain the defect difference.

[0186] In the above step S213, the method for calculating the mean difference is as follows: add up the defect differences calculated in the current iteration process in sequence, and then divide by the number of iterations to obtain the defect mean; calculate the defect mean in the previous iteration process, subtract the defect mean corresponding to the previous iteration process from the defect mean corresponding to the current iteration process, and take the absolute value to obtain the mean difference.

[0187] In this embodiment, by collecting printed product images and using computer vision technology to extract various product feature data, the quality status of printed products is comprehensively reflected; based on the product feature data, deep learning technology is used to quickly identify various quality defect problems, and the specific quality defect causes are traced through cluster analysis; an optimization strategy based on an optimization algorithm is adopted to dynamically optimize abnormal process parameters, thereby reducing quality defect problems in printed products, improving printing quality and production efficiency; giving full play to the advantages of cutting-edge technologies such as computer vision, machine learning, and intelligent optimization, realizing intelligent monitoring and dynamic optimization of the entire printing production process, quickly adapting to changing production environments, and improving the overall quality and market competitiveness of printed products.

[0188] Embodiment 2

[0189] This application also provides an electronic device. The electronic device may include one or more processors and one or more memories. Among them, computer-readable code is stored in the memory, and when the computer-readable code is run by one or more processors, it can execute a printing process optimization decision-making system based on big data as described above.

[0190] The method or system according to the embodiment of the present application can also be implemented by means of the architecture of the electronic device shown in the present application. The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, input / output, a hard disk, etc. The storage device in the electronic device, such as ROM or a hard disk, can store a printing process optimization decision-making system provided by the present application. Further, the electronic device may further include a user interface. Of course, the architecture shown in the present application is only exemplary, and when implementing different devices, one or more components shown in the electronic device of the present application can be omitted according to actual needs.

[0191] Embodiment 3

[0192] One embodiment of the present application discloses a computer-readable storage medium. Computer-readable instructions are stored on the computer-readable storage medium. When the computer-readable instructions are run by a processor, a big-data-based printing process optimization decision-making system according to an embodiment of the present application described with reference to the above drawings can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory, etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0193] In addition, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be run by a processor to execute instructions corresponding to the method steps provided by the present application, such as: a big-data-based printing process optimization decision-making system. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0194] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

[0195] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A printing process optimization decision system based on big data, characterized in that: include: An image acquisition module, used for acquiring images of printed products; A feature extraction module is used to analyze the printed product image and extract product feature data; The quality inspection module is used to analyze product feature data and identify quality defects; Cause tracing module: traces the causes of quality defects according to the quality defects; The process optimization module dynamically optimizes printing process parameters based on the causes of quality defects.

2. According to the big data-based printing process optimization decision system of claim 1, it is characterized in that: The product characteristic data includes color value, sharpness value, number of scratches, number of stains, gray value and printing area; the color value is the color representation of each pixel in the printed product image; the sharpness value is the sharpness degree of each pixel in the printed product image; the number of scratches is the number of scratches in the printed product image; the number of stains is the number of stains in the printed product image; the gray value is the brightness value of each pixel in the printed product image; the printing area is the area of ​​the area actually printed in the printed product image; The method for obtaining the color value is as follows: using a color space conversion method to convert the printed product image from the RGB color space to the Lab color space; according to the Lab color space corresponding to the printed product image, the color value corresponding to each pixel in the printed product image is obtained; the method for obtaining the sharpness value is as follows: defining a convolution kernel, the convolution kernel includes a horizontal convolution kernel and a vertical convolution kernel; performing a convolution operation on each pixel in the printed product image according to the convolution kernel, and obtaining a horizontal gradient value and a vertical gradient value corresponding to each pixel; according to the horizontal gradient value and the vertical gradient value, calculating the sharpness value corresponding to each pixel; The method for obtaining the number of scratches is: using a trained scratch detection model to identify the printed product image and output the number of scratches; the method for obtaining the number of stains is: using a trained stain detection model to identify the printed product image and output the number of stains; the training process of the stain detection model is consistent with the training process of the scratch detection model, and both are convolutional neural network models; The grayscale value is obtained by: obtaining the color data corresponding to each pixel in the printed product image according to the RGB color space corresponding to the printed product image; calculating the grayscale value corresponding to each pixel according to the color data; the printing area is obtained by: presetting a grayscale threshold, comparing the grayscale value of each pixel with the grayscale threshold, marking the pixel with a grayscale value less than the grayscale threshold as a printed point, and not marking the pixel with a grayscale value greater than or equal to the grayscale threshold; The number of printed dots is counted and marked as the printing quantity; the dot area is obtained, and the dot area is the area of ​​a pixel in the printed product image; the dot area is multiplied by the printing quantity to obtain the printing area.

3. The printing process optimization decision system based on big data according to claim 2 is characterized in that: The color value includes brightness, red-green axis and blue-yellow axis; The expression of the horizontal convolution kernel is: In the formula, G x is the horizontal convolution kernel; The expression of the vertical convolution kernel is: In the formula, G y is the vertical convolution kernel; The expression of the sharpness value is: Where G is the sharpness value, is the horizontal gradient value, is the vertical gradient value; The training process of the scratch detection model includes: Collect a number of printed product images in advance, where a is an integer greater than 1, and annotate each printed product image with the number of scratches; divide the annotated printed product images into a training set and a test set; use the training set to train the scratch detection model, and use the test set to test the scratch detection model; preset an error threshold, and when the mean of the prediction errors of all printed product images in the test set is less than the error threshold, output the scratch detection model; The color data includes a red channel value, a green channel value and a blue channel value; The expression of the gray value is: gray=0.2989×R+0.578×G+0.114×B; wherein gray is the gray value, R is the red channel value, G is the green channel value, and B is the blue channel value.

4. The printing process optimization decision system based on big data according to claim 3 is characterized in that: The method for identifying quality defect problems includes: Acquire a standard product image, and acquire a color value, a grayscale value, and a printing area corresponding to the standard product image; mark the color value corresponding to the standard product image as a first color value, and mark the color value corresponding to the printed product image as a second color value; subtract the first color value corresponding to each pixel from the second color value corresponding to the pixel, and acquire a color difference value of each pixel; mark the grayscale value corresponding to the standard product image as a first grayscale value, and mark the grayscale value corresponding to the printed product image as a second grayscale value; subtract the second grayscale value from the second grayscale value of each pixel, and acquire a grayscale difference value of each pixel; mark the printing area corresponding to the standard product image as a first area, and mark the printing area corresponding to the printed product image as a second area; subtract the first area from the second area, and acquire an area difference value; The sharpness value, the number of scratches, the number of stains, the color difference, the grayscale difference and the area difference are used as analysis data, and the analysis data are input into the trained quality inspection model to predict the corresponding inspection results; the inspection results are set labels, and the set labels are digital labels corresponding to the pre-constructed problem sets, and different problem sets have different corresponding digital labels; the method of pre-constructing the problem set is: obtain all quality defect problems; count the number of quality defect problems and mark them as the number of problems; combine all quality defect problems, and randomly select h quality defect problems in each combination process, h∈[0,H], H is the number of problems; take the quality defect problems combined each time as a set of problem sets, and obtain D sets of problem sets in total, D=2 H ; The D problem sets are all different, and the D problem sets cover all combinations of quality defect problems; according to the predicted set labels, the quality defect problems in the corresponding problem set are obtained.

5. The printing process optimization decision system based on big data according to claim 4 is characterized in that: The steps of tracing the causes of quality defects include: Step S101: convert all quality defect problems into vector form and mark them as problem vectors; Step S102: each question vector is used as a sample point, and the sample point corresponds to the question vector one by one; the number of categories g is preset, g sample points are randomly selected as center points, and a digital label is set for each center point in increasing order, and marked as ωk, k∈[1,g]; Step S103: Mark the sample points that are not used as the center points as classification points, and set a digital label for each center point in turn and mark it as ψc, c∈[1,Hg]; Step S104: Calculate the point distance from each classification point to each center point in sequence; Step S105: establishing corresponding g quality defect causes according to the g center points; Step S106: assigning the classification point ψc to the corresponding quality defect cause; Step S107: set c=c+1, and jump back to step S106; Step S108: looping steps S106 to S107 until c=Hg, then the loop ends and the process goes to step S109; Step S109: recalculate the new center point corresponding to each quality defect cause; Step S110: looping steps S104 to S109 until the new center point of each quality defect cause recalculated in step S109 is consistent with the new center point calculated in the previous loop, the loop ends, and the quality defect problems corresponding to the sample points corresponding to the k quality defect causes are obtained; Step S111: Mark the identified quality defect problem as a real-time problem, compare the real-time problem with the quality defect problem corresponding to each quality defect cause, obtain the quality defect cause corresponding to the quality defect problem that is the same as the real-time problem, and use it as the quality defect cause for traceability.

6. The printing process optimization decision system based on big data according to claim 5 is characterized in that: In step S104, the expression of point distance is: Where D ck is the point distance from the classification point ψc to the center point ωk, ω k is the problem vector corresponding to the classification point ωk, ψ c is the problem vector corresponding to the classification point ψc; In step S106, the method of assigning the classification point ψc to the corresponding quality defect cause is as follows: presetting a distance threshold, comparing the point distance from the classification point ψc to each center point with the distance threshold respectively, marking the center point whose point distance is less than the distance threshold as the assigned point, and not marking the center point whose point distance is greater than or equal to the distance threshold; assigning the classification point ψc to the quality defect cause corresponding to the assigned point; In step S109, the method for calculating the new center point of each quality defect cause includes: In the formula, ω′ k is the new center point corresponding to the kth quality defect cause, ψ kr is the rth classification point in the kth quality defect cause, R k is the number of classification points in the kth quality defect cause, r∈[1,R k ].

7. The printing process optimization decision system based on big data according to claim 6 is characterized in that: The step of dynamically optimizing printing process parameters comprises: Step S201: determining printing process parameters based on the cause of the quality defect; Step S202: collecting printing process parameters in real time and marking them as real-time process parameters; Step S203: performing abnormality detection on the real-time process parameters to obtain abnormal process parameters; Step S204: obtaining a process parameter range corresponding to the abnormal process parameter; Step S205: constructing M groups of parameter sets based on the process parameter range, and setting different parameter labels for each group of parameter sets, the parameter labels are digital labels, and the range of the parameter labels is [1, M]; Step S206: randomly select a parameter label as a valid solution; Step S207: updating the valid solution and determining whether the valid solution applies the updated value; Step S208: adjusting the preset step size factor, and updating the valid solution according to the adjusted step size factor, and determining whether the valid solution applies the updated value; Step S209: determining the search phase and adjusting the preset disturbance factor; Step S210: updating the effective solution according to the adjusted disturbance factor, and determining whether the effective solution applies the updated value; Step S211: respectively calculating the defect coefficients of the effective solution in this iteration process and the effective solution in the previous iteration process, and calculating the defect difference; Step S212: looping steps S207 to S211 until the defect difference is less than a preset difference threshold, the loop ends, and the process proceeds to step S213; Step S213: Calculate the mean difference; Step S214: loop step S206 to step S213 until the mean difference is less than the preset difference threshold or the defect coefficient corresponding to the effective solution is 0, then the loop ends; obtain the parameter label corresponding to the effective solution and mark it as a valid label; dynamically optimize the printing process parameters according to the parameter set corresponding to the effective label.

8. The printing process optimization decision system based on big data according to claim 7 is characterized in that: In step S201, the method for determining printing process parameters includes: Obtain a historical adjustment set, which includes the quality defect causes identified at historical moments and the corresponding printing process parameters for dynamic optimization; use an association rule learning method to analyze the historical adjustment set to obtain a parameter correspondence set, which includes different quality defect causes and the printing process parameters corresponding to each quality defect cause; obtain the corresponding printing process parameters from the parameter correspondence set according to the traced quality defect causes; In step S203, the method for obtaining abnormal process parameters includes: Obtain historical process parameters, which are printing process parameters collected at historical moments; take each identical parameter in the historical process parameters as a calculation set, and the calculation set corresponds to the parameters in the historical process parameters one by one; calculate the mean and sample difference corresponding to each calculation set; subtract the mean corresponding to the corresponding calculation set from each parameter in the real-time process parameters, and then divide it by the standard deviation corresponding to the corresponding calculation set to obtain the abnormal coefficient corresponding to each parameter in the real-time process parameters; preset the coefficient range [-u,u]; if -u≤YC v ≤u, the corresponding parameter is not marked; if -u>YC v ∪YC v >u, the corresponding parameter is marked as an abnormal parameter; among them, YC v is the abnormal coefficient corresponding to the vth parameter in the real-time process parameters, ∪ represents or; all abnormal parameters are regarded as abnormal process parameters; In step S204, the method for obtaining the process parameter range corresponding to the abnormal process parameter includes: Add u times the corresponding standard deviation to the mean value corresponding to each abnormal parameter to obtain the maximum value of each abnormal parameter; subtract u times the corresponding standard deviation from the mean value corresponding to each abnormal parameter to obtain the minimum value of each abnormal parameter; construct the parameter range corresponding to each abnormal parameter based on the maximum and minimum values ​​of each abnormal parameter, and use all parameter ranges as process parameter ranges.

9. The printing process optimization decision system based on big data according to claim 8, characterized in that: In step S205, the method for constructing M sets of parameter sets includes: A value is randomly selected from each parameter range within the process parameter range, and a set of parameter sets is constructed, and a total of M sets of parameter sets are constructed, and the M sets of parameter sets are all different; The method of step S207: updating the valid solution comprises: In the formula, is the updated valid solution, C1∈[0,1], N(0,1) is a random number in the standard normal distribution, and x is a valid solution; Methods for determining whether a valid solution applies the updated value include: The updated effective solution is marked as the updated solution, and the defect coefficients corresponding to the updated solution and the effective solution are calculated respectively. The defect coefficient corresponding to the updated solution is marked as the first coefficient, and the defect coefficient corresponding to the effective solution is marked as the second coefficient; the first coefficient is compared with the second coefficient; if the first coefficient is greater than the second coefficient, the effective solution maintains the value before the update; if the first coefficient is less than or equal to the second coefficient, the effective solution applies the updated value; the defect coefficient is calculated by: real-time collection of all parameters in the printing process and marking them as process comprehensive parameters; obtaining the parameter label corresponding to the effective solution, replacing the abnormal process parameters in the process comprehensive parameters with the parameter set corresponding to the parameter label, using the replaced process comprehensive parameters as test data, inputting the test data into the trained defect analysis model, and predicting the corresponding defect coefficient; the training process of the defect analysis model is consistent with the training process of the quality inspection model, and both are deep neural network models; In step S208, the method for adjusting the preset step size factor includes: If the updated value is applied in step S207, then In the formula, is the adjusted step size factor, p is the preset step size factor, γ is the adjustment factor, sig(f) is the improvement factor, f old is the second coefficient, f new is the first coefficient; If the updated value is not applied to the valid solution in step S207, then Where pen(f) is the penalty factor, Methods for updating the valid solution based on the adjusted step size factor include: In the formula, is the valid solution after updating according to the adjusted step size factor, x′ is the valid solution after executing step S207, and x best is the effective solution with the minimum defect coefficient in the iterative process, C2∈[0,1]; The method of determining whether the valid solution applies the updated value is the same as the method in step S207.

10. The printing process optimization decision system based on big data according to claim 9, characterized in that: In step S209, the method for determining the search phase includes: The search phase includes the early and late stages; The number of cycles of step S207 to step S211 is counted and marked as the number of iterations; a number threshold is preset, and the number of iterations is compared with the number threshold; if the number of iterations is greater than the number threshold, the exploration stage is judged to be the early stage; if the number of iterations is less than the number threshold, the exploration stage is judged to be the late stage; Methods for adjusting the preset disturbance factor include: If the exploration phase is the early phase, then In the formula, is the adjusted disturbance factor, q is the preset disturbance factor, e is the natural constant, and t is the number of iterations; If the exploration phase is in the late stage, then In the formula, T is the number threshold; In step S210, the method for updating the effective solution according to the adjusted disturbance factor includes: In the formula, is the valid solution updated according to the adjusted disturbance factor, x″ is the valid solution after executing step S208, and f best is the defect coefficient corresponding to the effective solution with the minimum defect coefficient in the iterative process, f x″ is the defect coefficient corresponding to the valid solution after executing step S208; The method for determining whether the updated value is applied to the valid solution is consistent with the method in step S207; In step S211, the defect difference is calculated by subtracting the defect coefficient corresponding to the effective solution in the previous iteration from the defect coefficient corresponding to the effective solution in the current iteration, and taking the absolute value to obtain the defect difference; In step S213, the method for calculating the mean difference is: adding the defect differences calculated in this iteration process in sequence, and then dividing by the number of iterations to obtain the defect mean; calculating the defect mean in the previous iteration process, subtracting the defect mean corresponding to the previous iteration process from the defect mean corresponding to this iteration process, and taking the absolute value to obtain the mean difference.

Images