Gravure printing process parameter optimization method and system based on tension fluctuation analysis

Abstract

The invention provides an intaglio printing process parameter optimization method and system based on tension fluctuation analysis, and belongs to the technical field of printing process parameter optimizing.The method comprises the steps that tension fluctuation signals in the running process of an intaglio printing machine are collected, and corresponding printing process parameters are collected; preprocessing and feature extraction are carried out on the tension fluctuation signal to obtain a multi-dimensional feature vector, and the multi-dimensional feature vector comprises a time domain feature and a frequency domain feature; inputting the printing process parameters into a preset parameter-tension prediction model to obtain a predicted multi-dimensional feature vector; calculating a residual error between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and identifying an abnormal process parameter through residual error analysis; taking minimization of tension fluctuation and overprinting errors as optimization objectives, based on physical constraints of printing process parameters, utilizing an optimization algorithm to adjust and optimize abnormal process parameters, and generating an optimized process parameter set; and the optimized process parameter set is updated to the photogravure press to carry out printing process parameter optimization.

Description

Technical Field

[0001] This application relates to the field of printing process parameter optimization technology, and in particular to a method and system for optimizing gravure printing process parameters based on tension fluctuation analysis. Background Technology

[0002] In existing gravure printing process parameter adjustments, manual judgment and trial-and-error are employed. Operators rely on their experience to observe the appearance quality of the printed product, such as color deviation and registration accuracy, to determine whether the printing process parameters are suitable. If problems are found, parameters such as unwinding tension, rewinding tension, printing unit speed, inter-unit tension, and drying temperature are gradually adjusted through multiple trials to find a relatively suitable parameter combination. This method is inefficient and requires extensive testing and adjustments. Additionally, simple sensors are used to monitor some parameters, such as tension sensors to monitor tension magnitude. This monitoring method only obtains single parameter values ​​and lacks a comprehensive analysis of the interrelationships between parameters and tension fluctuations. It cannot accurately identify abnormal process parameters and cannot achieve effective optimization of printing process parameters.

[0003] Therefore, there is an urgent need for a method and system for optimizing gravure printing process parameters based on tension fluctuation analysis. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method and system for optimizing gravure printing process parameters based on tension fluctuation analysis.

[0005] A first aspect of this application provides a method for optimizing gravure printing process parameters based on tension fluctuation analysis, including: The tension fluctuation signal of the gravure printing machine during operation is collected, and the corresponding printing process parameters are collected simultaneously. The tension fluctuation signal is preprocessed and features are extracted to obtain a multi-dimensional feature vector, which includes time-domain features and frequency-domain features. The printing process parameters are input into a preset parameter-tension prediction model to obtain a predicted multi-dimensional feature vector; Calculate the residual between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and identify abnormal process parameters through residual analysis; With the goal of minimizing tension fluctuations and registration errors, and based on the physical constraints of the printing process parameters, an optimization algorithm is used to adjust and optimize the abnormal process parameters, generating an optimized set of process parameters. The optimized set of process parameters is updated to the gravure printing machine to optimize the printing process parameters.

[0006] A second aspect of this application provides a gravure printing process parameter optimization system based on tension fluctuation analysis, comprising: The data acquisition module is used to collect tension fluctuation signals during the operation of the gravure printing machine and simultaneously collect the corresponding printing process parameters. The feature extraction module is used to preprocess and extract features from the tension fluctuation signal to obtain a multi-dimensional feature vector, which includes time-domain features and frequency-domain features. The model prediction module is used to input the printing process parameters into a preset parameter-tension prediction model to obtain a predicted multi-dimensional feature vector. The abnormal parameter module is used to calculate the residual between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and to identify abnormal process parameters through residual analysis. The adjustment and optimization module is used to adjust and optimize the abnormal process parameters based on the physical constraints of the printing process parameters, with the optimization goal of minimizing tension fluctuations and registration errors, and to generate an optimized set of process parameters. The parameter update module is used to update the optimized process parameter set to the gravure printing machine to optimize the printing process parameters.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described gravure printing process parameter optimization method based on tension fluctuation analysis.

[0008] In a fourth aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for optimizing gravure printing process parameters based on tension fluctuation analysis.

[0009] The beneficial effects of the gravure printing process parameter optimization method and system based on tension fluctuation analysis provided in this application are as follows: This application analyzes and optimizes the tension fluctuation signal and printing process parameters during the operation of the gravure printing machine, avoiding the subjectivity of manual experience judgment and trial and error, thus ensuring the consistency and accuracy of the adjustment results; after preprocessing and feature extraction, a multi-dimensional feature vector is obtained, and a predicted multi-dimensional feature vector is obtained through a parameter-tension prediction model. The residual between the two is calculated to identify abnormal process parameters. Then, the abnormal process parameters are optimized and adjusted with the goal of minimizing tension fluctuation and registration error. This can effectively identify abnormal process parameters in the gravure printing process, generate an optimized process parameter set and update it to the printing machine, thereby reducing tension fluctuation and registration error and improving the quality and performance of gravure printing. Attached Figure Description

[0010] Figure 1 A flowchart illustrating a gravure printing process parameter optimization method based on tension fluctuation analysis provided in an embodiment of this application; Figure 2 A structural block diagram of a gravure printing process parameter optimization system based on tension fluctuation analysis provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0013] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for optimizing gravure printing process parameters based on tension fluctuation analysis, provided in one embodiment of this application. The method includes: S101: Collects tension fluctuation signals during the operation of the gravure printing machine and simultaneously collects the corresponding printing process parameters.

[0014] In this embodiment, the tension fluctuation signal is the data on the change in tension of the substrate material (e.g., plastic film, paper, aluminum foil) during the conveying, printing, and rewinding processes of the gravure printing press. It is a physical signal representing the stability of the printing process. Excessive tension will cause the substrate material to stretch and deform, while insufficient tension will result in wrinkles and misregistration. The amplitude and frequency of the signal fluctuation are directly related to the quality of the printed product. Printing process parameters are operational parameters that affect the gravure printing process and product quality. These parameters include unwinding tension, rewinding tension, speed of each printing unit, tension between printing units, and drying temperature. Synchronous acquisition involves acquiring the corresponding printing process parameters at the same time dimension as the tension fluctuation signal, ensuring that each set of tension data corresponds one-to-one with the process parameters at that moment. This provides a time-consistent dataset for establishing a parameter-tension model and identifying abnormal parameters.

[0015] S102: Preprocess and extract features from the tension fluctuation signal to obtain a multi-dimensional feature vector, which includes time-domain features and frequency-domain features.

[0016] In this embodiment, tension fluctuation signal preprocessing is a set of operations that perform noise reduction and outlier removal on the acquired raw tension fluctuation signal. The purpose is to eliminate interference factors such as sensor noise and instantaneous equipment vibration, obtaining a steady-state tension signal that accurately represents the tension changes during the printing process, thus providing a reliable data foundation for subsequent feature extraction. Feature extraction is the process of extracting indicators that represent the tension fluctuation pattern from the processed steady-state tension signal. It is divided into two categories: time-domain features and frequency-domain features. Through feature extraction, continuous signal data can be transformed into high-dimensional feature vectors, facilitating subsequent analysis and anomaly identification.

[0017] In this embodiment, the time-domain features are statistical features directly calculated from the time-dimensional data of the tension signal, representing the amplitude variation characteristics of tension on the time axis without the need for frequency domain conversion. These features include standard deviation, peak-to-peak value, root mean square value, and waveform factor. The frequency-domain features are features extracted from the frequency-dimensional data after performing a Fourier transform on the steady-state tension signal. They represent the frequency distribution pattern of tension fluctuations and can identify characteristic frequencies related to equipment failures (e.g., roller eccentricity) and process anomalies, including the energy proportion of characteristic frequency bands and significant spectral peak frequencies. The multi-dimensional feature vector is a vector formed by combining the extracted time-domain and frequency-domain features according to certain rules. It is a digital feature identifier of the tension fluctuation signal and can be directly input into a prediction model or used for residual analysis to achieve a quantitative description of the tension state.

[0018] S103: Input the printing process parameters into the preset parameter-tension prediction model to obtain the predicted multi-dimensional feature vector.

[0019] In this embodiment, the preset parameter-tension prediction model is a mathematical model that has been trained in advance based on a historical synchronous dataset. Its function is to establish a mapping relationship between printing process parameters and tension fluctuation feature vectors. By inputting any set of process parameters, the corresponding predicted value of the multi-dimensional feature vector of tension fluctuation can be output. The model type used in this embodiment is a BP neural network.

[0020] Specifically, the parameter-tension prediction model adopts a 3-layer BP neural network structure. The module composition, number of neurons, and connection relationships of each layer are as follows: Input layer: 6 neurons, corresponding to the parameter dimensions of the gravure printing process, specifically: 1st color group squeegee pressure (MPa), 1st color group printing speed (m / min), 1st color group winding tension setpoint (N), cumulative rotations of the printing roller (10,000 rpm), cumulative squeegee running time (h), and printing roller surface roughness (μm). The input layer has no activation function; it is only responsible for receiving the raw process parameter data and transmitting it to the hidden layer. Hidden layer: 12 neurons, using the ReLU activation function (to solve the gradient vanishing problem and adapt to the nonlinear characteristics of industrial data). The hidden layer is connected to the input layer through a fully connected method, with a weight matrix dimension of 6×12 and a bias vector dimension of 12×1, realizing the nonlinear feature mapping of the input process parameters. Output layer: Contains 4 neurons, corresponding to the multi-dimensional feature vector of the tension signal, specifically: normalized standard deviation, normalized peak-to-peak value, normalized energy percentage of the feature frequency band, and normalized significant spectral peak frequency. The output layer uses a linear activation function to ensure the continuity of the output value and adapt to the quantization requirements of the tension features. It is fully connected to the hidden layer, with a weight matrix dimension of 12×4 and a bias vector dimension of 4×1, ultimately outputting a predicted multi-dimensional feature vector. The connections between layers follow a unidirectional propagation rule from the input layer to the hidden layer and then to the output layer. During backpropagation, the error is passed layer by layer from the output layer to the input layer, updating the weights and biases of each layer.

[0021] In this embodiment, the predicted multi-dimensional feature vector is the output of the parameter-tension prediction model. Its dimensions and feature composition are completely consistent with the extracted multi-dimensional feature vector, including the same time-domain and frequency-domain features. It serves as the benchmark reference value for subsequent calculation of residuals and identification of abnormal process parameters.

[0022] S104: Calculate the residual between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and identify abnormal process parameters through residual analysis.

[0023] In this embodiment, the residual is the difference between the extracted multi-dimensional feature vector and the predicted multi-dimensional feature vector output by the parameter-tension prediction model. It is an indicator of prediction accuracy and process condition anomalies. Specifically, the difference between the two vectors is calculated element-wise based on their corresponding feature components (using absolute difference or squared difference). The resulting residual vector can intuitively represent the degree of prediction deviation for each tension feature.

[0024] In this embodiment, residual analysis is a series of analytical processes involving evaluation, threshold comparison, and correlation mapping of residual vectors. The purpose is to determine whether the prediction deviation is caused by random error or abnormal process parameters, and to pinpoint the specific abnormal process parameters. Abnormal process parameters are printing process parameters whose values ​​deviate from reasonable ranges, whose coupling relationships between parameters are unbalanced, or whose tension fluctuation characteristics are abnormal (residuals exceeding limits) due to equipment wear. Examples include abnormally high squeegee pressure or sudden changes in printing speed.

[0025] S105: With the goal of minimizing tension fluctuations and registration errors, based on the physical constraints of printing process parameters, an optimization algorithm is used to adjust and optimize abnormal process parameters, generating an optimized set of process parameters.

[0026] In this embodiment, the optimization objective (minimizing tension fluctuations and registration errors) is the direction for adjusting printing process parameters. This requires simultaneously reducing two key indicators: tension fluctuations and registration errors are coupled. For example, excessive tension fluctuations can exacerbate registration errors, necessitating multi-objective optimization. The physical constraints of printing process parameters serve as the boundary conditions for parameter adjustment. These constraints stem from equipment performance, material properties, and printing mechanisms. They include the value range of individual parameters (e.g., the upper limit of printing speed), the coupling relationships between parameters (e.g., the matching limit between doctor blade pressure and printing roller speed), and constraints ensuring stable system operation. For example, tension fluctuations must not exceed the material's tolerance limits to prevent equipment failure or product scrapping caused by parameter adjustments.

[0027] In this embodiment, the optimization algorithm is a calculation method used to find the optimal combination of printing process parameters within the physical constraints. This embodiment uses the particle swarm optimization algorithm, which efficiently searches for the optimal solution and avoids blind trial and error.

[0028] Specifically, the Particle Swarm Optimization (PSO) algorithm adopts a three-layer architecture of population-particle-fitness evaluation. The functions and scenario adaptability of each layer are as follows: Population layer: As the global search carrier of the algorithm, it includes several sets of candidate process parameter combinations, i.e., particles. The population size is set according to the dimension of the optimization parameters. In this scenario, the optimization parameters are 4-color group squeegee pressure, printing speed, and tension setting values, totaling 12 dimensions. The population size is set to 100 to ensure the coverage of the search space; Particle layer: Each particle corresponds to a complete set of process parameter adjustment schemes. The particle's position vector is directly mapped to the 12-dimensional process parameter value. The velocity vector controls the direction and magnitude of parameter adjustment. Particles need to track their own historical best position (pbest) and the global best position of the population (gbest); Fitness evaluation layer: As the connecting module between the algorithm and the printing scenario, it inputs the process parameters corresponding to the particles, calls the parameter-tension prediction model to simulate global tension fluctuations and registration errors, calculates the comprehensive objective function value and converts it into fitness, and updates the particle position. Each level achieves global optimization through a loop logic of position update-fitness calculation-optimal tracking, and data transfer between levels follows a closed-loop process of species-particle-fitness evaluation-population update.

[0029] In this embodiment, abnormal process parameters are those identified as causing abnormal tension fluctuations, such as excessively high squeegee pressure or sudden changes in printing speed, and are the targets for adjustment and optimization. The optimized set of process parameters is a combination of process parameters that meets physical constraints and minimizes tension fluctuations and registration errors, obtained after solving the optimization algorithm. This set of process parameters can be directly updated and executed on the printing press.

[0030] S106: Update the optimized process parameter set to the gravure printing machine to optimize the printing process parameters.

[0031] In this embodiment, the optimized process parameter set is a complete combination of process parameters that satisfies physical constraints, minimizes tension fluctuations and registration errors, obtained by solving the optimization algorithm. This combination of process parameters includes adjusted abnormal parameters and related coupling parameters, with annotations such as parameter names, units, and value ranges. It is a standardized parameter data set that can be directly executed.

[0032] Updating a gravure printing press involves transmitting the optimized set of process parameters to the press's control system, such as a PLC system, via a communication link. This replaces the original parameter operation process, ensuring precise parameter delivery and effectiveness. Printing process parameter optimization refers to the printing press operating according to the updated optimized parameters, adjusting operating states such as squeegee pressure and printing speed to ultimately suppress tension fluctuations, reduce registration errors, and improve print quality and production stability.

[0033] As can be seen from the above, this application analyzes and optimizes the tension fluctuation signals and printing process parameters during the operation of the gravure printing machine, avoiding the subjectivity of manual experience judgment and trial and error, thus ensuring the consistency and accuracy of the adjustment results. After preprocessing and feature extraction, multi-dimensional feature vectors are obtained. The predicted multi-dimensional feature vectors are obtained through the parameter-tension prediction model. The residual between the two is calculated to identify abnormal process parameters. Then, the abnormal process parameters are optimized and adjusted with the goal of minimizing tension fluctuations and registration errors. This can effectively identify abnormal process parameters in the gravure printing process, generate an optimized set of process parameters and update it to the printing machine, thereby reducing tension fluctuations and registration errors and improving the quality and performance of gravure printing.

[0034] In one embodiment of this application, the tension fluctuation signal is preprocessed and feature extracted to obtain a multi-dimensional feature vector, including: The tension fluctuation signal is processed by outlier removal and moving average filtering to obtain the steady-state tension signal; Based on the steady-state tension signal, the statistical characteristics of the steady-state tension signal are calculated as time-domain features. The statistical characteristics include the standard deviation, peak-to-peak value, root mean square value, and waveform factor of the signal. The spectrum is obtained by performing a fast Fourier transform on the steady-state tension signal; Extract the energy proportion of preset characteristic frequency bands and the frequencies corresponding to spectral peaks with amplitudes greater than a first frequency threshold from the spectrum, and use them as frequency domain features; Time-domain features and frequency-domain features are combined in a predetermined order to form a multi-dimensional feature vector.

[0035] In this embodiment, the outlier removal method uses the 3σ criterion, which removes extreme data points from the original tension fluctuation signal that do not conform to physical laws due to factors such as sensor failure, instantaneous equipment vibration, and external interference. The purpose is to eliminate noise interference to subsequent signal analysis. Moving average filtering involves selecting a fixed-length sliding window, calculating the arithmetic mean of the tension signal data within the window, and replacing the signal value at the center of the window with the average value. By sliding window by window, the entire signal is smoothed, effectively filtering out high-frequency noise and obtaining a stable signal curve. The steady-state tension signal, after outlier removal and moving average filtering, eliminates random interference and high-frequency noise, accurately representing the tension variation of the substrate material during gravure printing and serving as the foundational data for feature extraction.

[0036] In this embodiment, the time-domain features are statistical indicators directly calculated from the time-dimensional data of the steady-state tension signal, without the need for frequency domain conversion. They can intuitively represent the amplitude variation characteristics of tension on the time axis, including four categories: standard deviation, peak-to-peak value, root mean square value, and waveform factor. The Fast Fourier Transform (FFT) is a mathematical algorithm that converts a time-domain signal into a frequency-domain signal. It can decompose a continuous tension fluctuation signal into sinusoidal components of different frequencies, obtaining the frequency-amplitude distribution law of the signal, i.e., the spectrum, which is a means of extracting frequency-domain features.

[0037] In this embodiment, the spectrum is a graph showing the relationship between frequency and amplitude obtained after the steady-state tension signal undergoes a Fast Fourier Transform (FFT). The horizontal axis represents frequency, and the vertical axis represents the signal amplitude at the corresponding frequency, clearly demonstrating the frequency composition of tension fluctuations. The preset characteristic frequency band is a frequency range pre-defined based on the operating characteristics of the gravure printing press (e.g., the rotation frequency of the printing roller and the vibration frequency range of the transmission system). The signal components within this band are directly related to printing process abnormalities and equipment malfunctions. The first frequency threshold is a pre-set amplitude judgment standard used to filter meaningful spectral peaks in the spectrum. Only spectral peaks with amplitudes greater than the first frequency threshold will have their corresponding frequencies extracted as frequency domain features, avoiding interference from invalid low-frequency noise. Frequency domain features are indicators extracted from the spectrum that represent the frequency distribution pattern of tension fluctuations. In this embodiment, they include two categories: the energy proportion of the preset characteristic frequency band and the frequencies corresponding to significant spectral peaks, which can identify characteristic frequencies related to equipment malfunctions.

[0038] As can be seen from the above, this embodiment obtains a steady-state tension signal by performing outlier removal and moving average filtering on the tension fluctuation signal, which can eliminate abnormal interference in the signal and provide stable data for subsequent analysis; calculating the statistical characteristics of the steady-state tension signal as time-domain features can represent the tension fluctuation characteristics from the time dimension; performing a fast Fourier transform on the steady-state tension signal to obtain the spectrum and extracting frequency-domain features can analyze the tension fluctuation characteristics from the frequency dimension; combining the time-domain features and frequency-domain features in a predetermined order to form a multi-dimensional feature vector can comprehensively and accurately describe the tension fluctuation situation, providing a foundation for subsequent analysis.

[0039] In one embodiment of this application, the extracted time-domain features and frequency-domain features are combined according to a predetermined order, including: The time-domain features and frequency-domain features are normalized separately. The normalized time-domain features are arranged in a fixed order of standard deviation, peak-to-peak value, root mean square value, and waveform factor. The normalized frequency-domain features are arranged in ascending order of energy proportion of characteristic frequency bands and significant spectral peak frequency. The two are then combined to form the basic feature vector. Calculate the correlation coefficient between the variance of each component in the basic eigenvector and the tension fluctuation; Redundant feature components with variances less than a preset first threshold and correlation coefficients less than a preset second threshold are removed to obtain a multi-dimensional feature vector.

[0040] In this embodiment, normalization is a linear transformation operation that maps the original values ​​of time-domain and frequency-domain features to the interval [0,1] or [-1,1]. The purpose is to eliminate dimensional differences between different features; for example, the unit of tension standard deviation is N, and the unit of energy percentage is %. This prevents features with large numerical ranges from dominating subsequent analysis. The basic feature vector is an initial feature vector formed by directly concatenating the normalized time-domain features and frequency-domain features according to a preset order. It includes all extracted feature components without redundancy filtering and forms the basis for subsequent feature dimensionality reduction. Variance is the degree of numerical dispersion of a component in the basic feature vector across multiple samples. The smaller the variance, the smaller the variation of the feature component under different operating conditions, and the weaker its ability to distinguish tension fluctuation states. The correlation coefficient is the degree of linear correlation between a component in the basic feature vector and the tension fluctuation index (e.g., tension standard deviation). Its value ranges from [-1,1]. The closer the absolute value is to 1, the stronger the correlation; the closer it is to 0, the weaker the correlation.

[0041] In this embodiment, a preset first threshold is a critical value of variance, used to determine whether a feature component has no discriminative power. A preset second threshold is a critical value of the absolute value of the correlation coefficient, used to determine whether a feature component is unrelated to tension fluctuations. Redundant feature components are those that simultaneously satisfy both a variance less than the first threshold and an absolute value of the correlation coefficient less than the second threshold. These features do not contribute to the representation of tension fluctuation states, and removing them can reduce vector dimensions and improve computational efficiency.

[0042] As can be seen from the above, this embodiment eliminates the influence of different feature dimensions by normalizing the time-domain and frequency-domain features respectively, making the features comparable; the normalized time-domain and frequency-domain features are arranged in a fixed order and concatenated into a basic feature vector, which facilitates processing and analysis; the correlation coefficient between the variance of each component of the basic feature vector and the tension fluctuation is calculated and redundant feature components are removed, which can reduce the data dimension and reduce the computational complexity, while retaining features that are strongly correlated with the tension fluctuation, thereby improving the accuracy and efficiency of analysis and optimization.

[0043] In one embodiment of this application, the residual between a multi-dimensional feature vector and a predicted multi-dimensional feature vector is calculated, and abnormal process parameters are identified through residual analysis, including: Arrange the multi-dimensional feature vector and the predicted multi-dimensional feature vector in the same order, subtract them element by element and take the absolute value to obtain the initial residual vector; Each residual value in the initial residual vector is compared with the preset anomaly judgment threshold of the corresponding feature component. The feature component whose residual value is greater than the corresponding preset anomaly judgment threshold is marked as an abnormal feature. Based on the preset abnormal feature-process parameter lookup table, one or more process parameters directly associated with the abnormal feature are output as abnormal process parameters.

[0044] In this embodiment, the initial residual vector is a set of vectors formed by calculating the absolute difference between the actual multi-dimensional feature vector and the predicted multi-dimensional feature vector element by element according to their corresponding components. Its dimensions are consistent with the feature vectors, and the residual value of each component directly represents the degree of prediction deviation for the corresponding feature, serving as data for anomaly feature identification. The preset anomaly judgment threshold is a pre-set residual critical value for each feature component in the multi-dimensional feature vector. The preset anomaly judgment threshold needs to be determined based on the residual statistical patterns of historical samples (e.g., the 75th percentile) and engineering experience. The thresholds for different feature components vary due to differences in physical meaning and fluctuation characteristics.

[0045] In this embodiment, abnormal features are the feature components in the initial residual vector whose residual values ​​are greater than the corresponding preset abnormality judgment threshold. The actual values ​​of these features deviate from the predicted values ​​by more than a reasonable range, caused by abnormal process parameters or equipment malfunctions. The abnormal feature-process parameter lookup table is a mapping table established based on gravure printing mechanism analysis and historical fault data, clarifying the potential abnormal process parameters corresponding to different abnormal features. It serves as the basis for locating parameter abnormalities from feature anomalies. Abnormal process parameters are the printing process parameters that cause tension fluctuation anomalies, determined through the abnormal feature reverse matching lookup table, and are the objects of subsequent parameter optimization.

[0046] As can be seen from the above, this embodiment identifies abnormal process parameters through residual analysis, which can accurately identify the process parameters that cause abnormal tension fluctuations, providing a basis for subsequent optimization and adjustment. This helps to improve the stability and printing quality of gravure printing process and reduce registration errors. At the same time, by subtracting the predicted multi-dimensional feature vector from the multi-dimensional feature vector arranged in the same order and taking the absolute value, the initial residual vector is obtained. Then, it is compared with the preset abnormal judgment threshold to mark abnormal features and output abnormal process parameters, making the identification process of abnormal process parameters more scientific and accurate.

[0047] In one embodiment of this application, identifying abnormal process parameters through residual analysis further includes: Collect images of printed materials output from each printing unit, detect periodically occurring defects using machine vision algorithms, and generate visual detection results for registration errors. Cepstral analysis was performed on the steady-state tension signal to extract the quister frequency components and their amplitudes corresponding to the mechanical rotation cycle of each printing roller, which were used as special features of the printing roller status. When the visual inspection result of overprinting error shows periodic defects in a specific color group, and the absolute value of the residual between the actual observed value of the corresponding frequency component in the dedicated feature vector and the corresponding component in the predicted multi-dimensional feature vector is greater than the preset plate roller health threshold, a mechanical health abnormality diagnosis conclusion for the plate roller of that color group is generated. Mechanical health anomaly diagnosis conclusions, as a special type of equipment status anomaly parameter, are incorporated into the abnormal process parameter set and trigger maintenance warning instructions that include specific color group locations and suspected fault types.

[0048] In this embodiment, printed image acquisition involves using an industrial camera to capture images of the output printed materials from each color group (e.g., color groups 1-4) of the gravure printing press. The acquisition location is after the color group drying unit and before the rewinding unit. The image resolution and frame rate must be adapted to the printing speed, for example, 2 megapixels and 50 frames per second, for subsequent defect detection. The machine vision algorithm is an image processing algorithm used for printed defect detection. In this embodiment, it involves template matching and frequency domain analysis. By comparing the acquired image with a standard defect-free template, periodically recurring defects such as stripes and misregistration are identified, and the location, period, and type of defects are output. The visual detection result of the registration error is quantitative data of the multi-color group registration accuracy output by the machine vision algorithm, including lateral registration error, longitudinal registration error, and the periodic characteristics of the defects, such as the defect period being consistent with the rotation period of a certain color group's printing roller. This serves as the visual basis for determining color group anomalies.

[0049] Specifically, the image processing algorithm adopts a four-layer serial architecture of image preprocessing, feature extraction, defect detection, and result output. The functions of each layer are deeply adapted to the printing scenario: Image preprocessing layer: The module includes grayscale conversion, Gaussian filtering, and image enhancement. Its main function is to eliminate the interference of surface reflection and noise on the detection of printed materials, and to unify the brightness and contrast of the image; Feature extraction layer: It includes two sub-modules: edge detection and template matching. Edge detection is used to extract the contour features of each color group pattern on the printed material, and template matching is used to locate the corresponding position of the standard pattern and the pattern to be detected; Defect detection layer: It is a periodic defect identification module. It extracts the frequency features of defects through frequency domain analysis and compares them with the rotation frequency of each printing roller to determine whether it is a periodic overprinting defect; Result output layer: It quantifies the detected overprinting deviation and outputs key information such as overprinting error value (horizontal / vertical), defect location (color group number), and defect period, forming the visual detection result of overprinting error. Data is transmitted unidirectionally at each level in the order of preprocessing - feature extraction - defect detection - result output. The output of the previous level serves as the input of the next level, ensuring the continuity and accuracy of the detection process.

[0050] In this embodiment, cepstrum analysis involves performing a Fourier transform on the spectrum of the steady-state tension signal. Its purpose is to extract periodic components from the spectrum, effectively identifying frequency components related to the mechanical rotation of the printing roller. This avoids noise interference in direct spectrum analysis and is a means of diagnosing mechanical faults in the printing roller. The quiescent frequency component is a characteristic frequency corresponding to the mechanical rotation period of the printing roller. Its value is equal to the reciprocal of the printing roller's rotational speed, F=n / 60, where n is the printing roller's rotational speed in r / min. The amplitude change of this frequency component directly indicates the mechanical state of the printing roller, such as eccentricity or wear. The printing roller condition-specific feature is a combination of the quiescent frequency component extracted from the cepstrum and its corresponding amplitude, specifically used to diagnose the mechanical health of the printing roller.

[0051] In this embodiment, the preset printing roller health threshold is a critical value set for the residual characteristics of the printing roller's condition. It is determined by statistical data of the residual characteristics during normal operation of the printing roller. If the residual is greater than the preset printing roller health threshold, it is determined that the printing roller has a mechanical abnormality. The mechanical health abnormality diagnosis conclusion is a specific diagnostic result output based on the visual detection results of the overprinting error and the analysis of the residual characteristics of the printing roller. It includes the location of the abnormal color group and the suspected fault type, such as printing roller eccentricity or bearing wear. Equipment status-type abnormal parameters are abnormalities caused by mechanical failures of the equipment. They are different from process parameter setting abnormalities. These abnormalities cannot be resolved by adjusting process parameters and need to be eliminated through equipment maintenance. Therefore, they are included in the abnormal process parameter set and trigger maintenance warnings.

[0052] As can be seen from the above, this embodiment can generate visual inspection results of registration error by collecting images of printed materials to detect periodic defects. It can extract special features of the printing roller status by performing cepstral analysis on the steady-state tension signal. It can diagnose mechanical health abnormalities of the printing roller by using the visual inspection results of registration error and the residual of the special feature vector. It can incorporate the diagnostic conclusions into the abnormal process parameter set and trigger maintenance warning commands, which helps to discover equipment problems in a timely manner and carry out maintenance. It can also identify abnormal process parameters more comprehensively and accurately, providing a basis for process parameter optimization.

[0053] In one embodiment of this application, before comparing each residual value in the initial residual vector with a preset anomaly judgment threshold for the corresponding feature component, the method further includes: Based on the cumulative running time and wear status of preset components, the preset abnormality judgment thresholds of each feature component are adjusted; the preset components include the printing roller and the scraper. The cumulative number of rotations of the printing roller and the historical data of the pressure and angle of the scraper are obtained, and a comprehensive health index is calculated through a preset wear model. If the overall health index is less than the preset health threshold, the preset abnormality judgment threshold of the corresponding feature component will be increased based on the first step.

[0054] In this embodiment, the preset components are core components directly related to tension fluctuations in gravure printing and prone to wear due to long-term operation. Specifically, this refers to the printing roller and doctor blade. Wear on these two types of components directly alters ink transfer efficiency and the contact state of the printing material, thus causing a shift in tension fluctuation characteristics. Using a fixed threshold would lead to misjudgments. The cumulative running time is a statistical value of the running duration of the preset components. The cumulative revolutions are the core quantitative indicator of wear on the printing roller, representing the total number of rotations from when it was put into use to the current moment, and are a key parameter for assessing the degree of wear on the roller surface. The wear model is a mathematical model constructed based on historical operating data and wear detection data of the preset components. It represents the correlation between the degree of component wear and tension characteristic deviation, outputting an index that comprehensively represents the health status of the components.

[0055] In this embodiment, the comprehensive health index is a normalized value representing the overall health status of the printing roller and doctor blade, calculated using a wear model, and its value ranges from [0,1]. The closer the value is to 1, the better the component's health status; the closer it is to 0, the more severe the wear. The preset health threshold is a critical value used to determine whether a preset component is in a wear-prone state, determined by statistical analysis of historical fault data. When the comprehensive health index is less than the preset health threshold, it is determined that component wear has affected tension fluctuation characteristics, and the anomaly judgment threshold needs to be adjusted. The first step length is the adjustment range of the anomaly judgment threshold, used to control the degree to which the anomaly judgment threshold is increased, avoiding over-adjustment that leads to missed anomaly detection.

[0056] Specifically, the first step length is calculated through a linear correlation between the comprehensive health index and the step length. The more severe the component wear, the larger the first step length for adjusting the anomaly judgment threshold. The calculation formula is: First step length = Base step length × (Preset health threshold - Current comprehensive health index / Preset health threshold). The base step length is determined by historical data and is set to 10%; the first step length ranges from [5%, 20%].

[0057] For example, with a base step size of 10% and a preset health threshold of 0.3, the current comprehensive health index is calculated to be 0.274 using the wear model. Therefore, the first step size = 10% × 0.3 - 0.274 / 0.3 = 10% × 0.087 = 0.87%. Since the calculated step size is less than the minimum effective step size of 5%, the final first step size is set to 5%.

[0058] As can be seen from the above, this embodiment adjusts the preset abnormal judgment threshold according to the cumulative running time and wear status of the printing roller and the scraper, which makes the identification of abnormal process parameters more consistent with the actual operating conditions of the equipment; calculating the comprehensive health index and raising the threshold when it is less than the preset health threshold can avoid misjudgment caused by normal wear of the equipment and improve the accuracy of abnormal process parameter identification.

[0059] In one embodiment of this application, with the optimization objective of minimizing tension fluctuations and registration errors, and based on the physical constraints of printing process parameters, an optimization algorithm is used to adjust and optimize abnormal process parameters, generating an optimized set of process parameters, including: A comprehensive objective function is constructed, with the objectives of minimizing tension fluctuations and minimizing overprinting errors. Determine the adjustment range for each abnormal process parameter to obtain physical constraints; the adjustment range includes the upper and lower limits of the abnormal process parameter itself and the linear or nonlinear coupling relationship between the parameters. Within the search space defined by physical constraints, a genetic algorithm is used to iteratively search for abnormal process parameters with the goal of minimizing the comprehensive objective function. When the genetic algorithm reaches the convergence condition, it outputs the optimal combination of parameters found so far, which serves as the optimized set of process parameters.

[0060] In this embodiment, the comprehensive objective function is a function that integrates the two optimization objectives of minimizing tension fluctuation and minimizing registration error into a single mathematical expression. The priority of the two objectives is balanced through weight allocation; for example, in high-precision printing, increasing the weight of registration error is the optimization target of the algorithm. Physical constraints are the boundary conditions that need to be satisfied when adjusting abnormal process parameters, including two types: first, single-parameter constraints, such as the upper and lower limits of printing speed, which originate from equipment performance and material characteristics; second, parameter coupling constraints, such as the linear relationship between the printing roller speed and the printing speed, which originates from the printing mechanism and ensures stable equipment operation after parameter adjustment.

[0061] In this embodiment, the search space is a set of parameter value ranges defined by the physical constraints of all abnormal process parameters. The optimization algorithm searches for the optimal solution only within this space, avoiding invalid or dangerous parameter combinations. Genetic algorithms are heuristic optimization algorithms that simulate biological evolution. They find the optimal parameter combination within the search space through selection, crossover, and mutation iterations. They are suitable for multi-objective, nonlinearly constrained process parameter optimization scenarios, exhibiting strong robustness and global search capabilities.

[0062] Specifically, the genetic algorithm adopts a four-layer architecture: population-individual-genetic operation-fitness evaluation. The functions of each layer are deeply adapted to the gravure printing process parameter optimization scenario: Population layer: Serving as the carrier of the global search, it consists of several sets of candidate process parameter combinations (individuals). The population size is set according to the dimensions of the optimized parameters. In this scenario, the optimization of 4-color group doctor blade pressure, printing speed, and tension settings totals 12 dimensions, and the population size is set to 100 to fully cover the search space; Individual layer: Each individual corresponds to a complete set of abnormal process parameter adjustment schemes. The individual's gene sequence is directly mapped to a 12-dimensional process parameter. The process parameters are encoded using real numbers, eliminating the need for number system conversion and directly adapting to the continuous value characteristics of the parameters. The genetic operation layer includes three modules: selection, crossover, and mutation. The selection module uses a roulette wheel algorithm to select high-quality individuals, the crossover module uses single-point crossover to achieve high-quality gene recombination, and the mutation module introduces new genes through random perturbation. The fitness evaluation layer, serving as the core link between the algorithm and the printing scenario, takes the process parameters corresponding to each individual as input, calls the parameter-tension prediction model to simulate global tension fluctuations and registration errors, calculates the comprehensive objective function value, and converts it into fitness, providing a decision-making basis for genetic operations. Each layer operates according to a closed-loop logic of population initialization, fitness evaluation, genetic operations, and population update, allowing the algorithm to gradually approach the global optimum.

[0063] In this embodiment, the convergence condition is the critical condition for determining when the genetic algorithm stops iterating. This includes: the number of iterations reaching a preset value, the objective function value not significantly decreasing over multiple generations, and the optimal solution remaining stable, allowing the algorithm to terminate promptly after finding the optimal solution. The optimal parameter combination is the set of process parameters that minimizes the comprehensive objective function value within the physical constraints after the genetic algorithm converges; it is the final output optimization result.

[0064] As can be seen from the above, this embodiment can clarify the optimization direction by constructing a comprehensive objective function aimed at minimizing tension fluctuations and registration errors; determine the adjustment range of abnormal process parameters as a physical constraint to ensure that the parameter adjustments are within a reasonable range; use a genetic algorithm to iteratively search for abnormal process parameters in the constraint space to efficiently find the optimal solution; and output the optimal parameter combination as the optimized process parameter set when the algorithm converges, which can realize the optimization of gravure printing process parameters and reduce tension fluctuations and registration errors.

[0065] In one embodiment of this application, an optimization algorithm is used to adjust and optimize abnormal process parameters, employing a decoupled global optimization strategy based on transfer function identification, including: When constructing the parameter-tension prediction model, system identification technology is used to establish a multi-input multi-output dynamic coupling model based on historical operating data to characterize the impact of changes in process parameters of each printing unit on the global tension system. This model can predict the transmission and coupling effect of process parameter adjustments of any unit on its own and the tension characteristics of all downstream units. The comprehensive objective function is defined as the weighted sum of the global cost of tension fluctuation and the global cost of overprinting error. The global cost of tension fluctuation is calculated by the sum of the weighted norms of the tension feature vectors of each unit, and the global cost of overprinting error is comprehensively evaluated by the virtual registration deviation calculated by the parameter-tension prediction model based on the simulated cross-unit tension coupling results, or by fusing with the overprinting error data fed back by real-time visual inspection. The physical constraints are extended to include: the independent feasible domains of each process parameter, the coupling inequality constraints between key parameters based on the principle of equipment dynamics, and the system stable operation boundary derived from the stability criterion of the dynamic coupling model; During the iterative search process of the genetic algorithm, for each set of candidate abnormal process parameter adjustment values, a multi-input multi-output dynamic coupling model is invoked to simulate and calculate the changes in all tension characteristics from the adjustment point to the final unit, and to evaluate its impact on the comprehensive objective function. This ensures that the final optimized process parameter set is the solution that enables the entire gravure printing machine to achieve the global optimal performance under all hard constraints.

[0066] In this embodiment, the decoupling global optimization strategy based on transfer function identification is an optimization method for multi-unit coupled systems. It establishes a dynamic coupling model between process parameters and global tension and registration error through system identification, quantifying the transmission effect of parameter adjustments between units. This breaks the coupling correlation between units, achieving global optimization rather than local optimization. System identification technology uses historical operating data (input: process parameters; output: tension characteristics, registration error) and mathematical modeling methods (such as Kalman filtering) to derive the system's input-output relationship. It is a means of constructing a multi-input multi-output dynamic coupling model. The multi-input multi-output dynamic coupling model (MIMO) takes the process parameters of each printing unit as input, such as squeegee pressure, printing speed, and tension setpoint, and outputs a dynamic model of the tension characteristics and registration error of each unit. It can accurately predict the chain reaction of parameter adjustments in a certain unit on its own and all downstream units' tension and registration accuracy, characterizing the coupling effect between units.

[0067] In this embodiment, the global cost of tension fluctuation is a comprehensive index representing the degree of tension fluctuation in each printing unit of the entire machine. It is calculated by the sum of the weighted norms of the tension feature vectors of each unit. The weights are positively correlated with the degree of influence of the unit on the final printed quality; for example, the weight of the subsequent color group is higher than that of the preceding color group. The global cost of registration error is a comprehensive index representing the registration accuracy of the entire machine. There are two calculation methods: one is a virtual registration deviation assessment based on the simulation of cross-unit tension coupling results using a MIMO dynamic coupling model; the other is a calculation that integrates with real-time registration error data detected by machine vision to improve the accuracy of the assessment.

[0068] In this embodiment, the physical constraint extension adds two types of constraints to the original single-parameter feasible region: one is the coupling inequality constraint based on key parameters of equipment dynamics, such as the matching relationship between the printing roller speed and the printing speed; the other is the system stable operation boundary based on the stability criterion of the dynamic coupling model, such as ensuring that the parameter adjustment range ensures that the model pole is located in the left half of the complex plane to avoid system oscillation. Candidate abnormal process parameter adjustment values ​​are each set of parameters to be evaluated generated during the genetic algorithm iteration process. Their impact on global tension and registration error needs to be simulated using the MIMO dynamic coupling model before calculating the comprehensive objective function value. The globally optimal performance is the optimized set of process parameters that, while satisfying all physical constraints, simultaneously optimizes the tension fluctuations and registration errors of all units in the entire machine, rather than optimizing the performance of a single unit.

[0069] As can be seen from the above, this embodiment, by constructing a multi-input multi-output dynamic coupling model using system identification technology, can predict the transmission and coupling effects of any unit's process parameter adjustment on its own and downstream unit's tension characteristics; the comprehensive objective function is defined as the weighted sum of the global cost of tension fluctuation and the global cost of registration error, which can comprehensively consider tension fluctuation and registration error; physical constraints are extended to ensure that the process parameter adjustment is within a reasonable range; the dynamic coupling model is called in the iterative search of the genetic algorithm to simulate and calculate the changes in tension characteristics and evaluate the impact on the comprehensive objective function, which can generate an optimized set of process parameters that satisfies all hard constraints and makes the entire gravure printing machine achieve the global optimal performance.

[0070] In one embodiment of this application, after calculating the residual between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and identifying abnormal process parameters through residual analysis, the method further includes: Calculate the overall residual norm of the multi-dimensional feature vector and the predicted multi-dimensional feature vector; If the overall residual norm is greater than the preset model update threshold, the output will specify the model parameter optimization requirements for the parameter-tension prediction model. In response to the need for model parameter optimization, the particle swarm optimization algorithm is used to optimize the parameters of the parameter-tension prediction model, which is a neural network model.

[0071] In this embodiment, the overall residual norm is a scalar value obtained by calculating the norm of the residual vector between the actual multi-dimensional feature vector and the predicted multi-dimensional feature vector. It is a comprehensive index representing the overall deviation of the two sets of vectors, rather than the deviation of a single component. The larger the calculated result, the more significant the overall prediction deviation. The preset model update threshold is a critical value for determining whether the parameter-tension prediction model needs to be updated, and it is determined by statistical analysis of historical model prediction accuracy data. When the overall residual norm is greater than the preset model update threshold, it indicates that the parameter-tension prediction model can no longer accurately fit the current process-tension correlation, and the model parameters need to be optimized.

[0072] In this embodiment, the requirement for model parameter optimization is a command signal triggered by excessive prediction deviation in the parameter-tension prediction model. It adjusts the internal parameters of the parameter-tension prediction model and serves as the trigger condition for initiating parameter optimization. Particle swarm optimization (PSO) is a swarm intelligence optimization algorithm that simulates the foraging behavior of bird flocks. It finds the optimal solution by updating the positions of particles in the solution space (tracking individual optima and global optima). It is suitable for parameter optimization in neural network models and features fast convergence and simple implementation. The neural network model parameters are the training parameters of the parameter-tension prediction model, including the weights and biases between neurons in each layer. These are the optimization objects of the PSO algorithm; adjusting the training parameters can improve the prediction accuracy of the parameter-tension prediction model.

[0073] Specifically, the Particle Swarm Optimization (PSO) algorithm for neural network parameter optimization adopts a four-layer architecture: particle layer, optimal tracking layer, fitness evaluation layer, and parameter update layer. This architecture optimizes the weights and biases of the parameter-tension prediction neural network. Each layer's function is deeply integrated with the neural network parameter optimization scenario: Particle layer: The algorithm's search unit. Each particle corresponds to a complete set of neural network parameters to be optimized (input-hidden layer weights, hidden layer biases, hidden-output layer weights, and output layer biases). The particle's position vector dimension is consistent with the total number of neural network parameters to be optimized (in this embodiment, a 3-layer BP neural network has 148 parameters to be optimized, and the particle position vector is 148-dimensional). Optimal tracking layer: Includes two sub-modules: individual optimal (pbest) and global optimal (gbest). Individual optimal records the historical search results of each particle. The optimal position (corresponding to the optimal combination of neural network parameters) is recorded globally, providing directional guidance for the search of all particles. The fitness evaluation layer is the core link between the algorithm and the neural network. It inputs the neural network parameters corresponding to the particles, loads them into the parameter-tension prediction model, calculates the mean square error between the model's predicted and actual values ​​using test set data, and converts the error value into a fitness value using the fitness function (fitness function = 1 / 1 + MSE). A higher fitness value indicates better corresponding neural network parameters. The parameter update layer adjusts the particle's search direction and step size using the velocity-position update formula based on the particle's current position, velocity, and individual and global optimal positions, causing the particles to approach the optimal parameter combination. Simultaneously, parameter constraint verification is embedded to ensure that the updated parameters are within a reasonable range.

[0074] As can be seen from the above, this embodiment calculates the overall residual norm after identifying abnormal process parameters. When the overall residual norm is greater than the preset model update threshold, it outputs the model parameter optimization requirements. The particle swarm optimization algorithm is used to optimize the parameters of the neural network model. This can promptly detect the deviation of the parameter-tension prediction model and optimize its parameters, thereby improving the model prediction accuracy and enhancing the effect of gravure printing process parameter optimization.

[0075] In one embodiment of this application, before optimizing the parameters of the parameter-tension prediction model using the particle swarm optimization algorithm, the method further includes: The search space of the particle swarm optimization algorithm is determined based on the number and range of parameters to be optimized in the neural network model. The combination of hyperparameters for the particle swarm optimization algorithm is determined based on the magnitude of the overall residual norm.

[0076] In this embodiment, the parameters to be optimized are the trainable parameters in the parameter-tension prediction neural network model, mainly including the weight matrix and bias vector between neurons in each layer. These parameters directly determine the prediction accuracy of the model and are the optimization objects of the particle swarm optimization algorithm. The search space is the set of value ranges of particles (corresponding to a set of model parameters) in the particle swarm optimization algorithm. It is defined by the number of parameters to be optimized and the reasonable value range of a single parameter. It is the optimization boundary of the algorithm, making the parameter optimization process smaller than the physical meaning and engineering reasonable range of the model.

[0077] In this embodiment, the hyperparameter combination is the set of control parameters of the particle swarm optimization algorithm itself, such as the number of particles, the number of iterations, the learning factor, and the inertia weight. These parameters do not participate in model prediction but directly affect the search efficiency and optimization effect of the algorithm. The hyperparameter combination needs to be adjusted according to the overall residual norm to adapt to different levels of model bias. The magnitude of the overall residual norm represents the overall prediction bias of the parameter-tension prediction model: the larger the norm, the more significant the model bias, and hyperparameters that are more biased towards global search need to be configured; the smaller the norm, the slighter the model bias, and hyperparameters that are more biased towards local refinement can be configured.

[0078] As can be seen from the above, this embodiment determines the search space of the particle swarm optimization algorithm based on the number and value range of the parameters to be optimized in the neural network model, so that the algorithm can perform parameter optimization within a suitable range and improve the optimization efficiency. The hyperparameter combination of the particle swarm optimization algorithm is determined based on the magnitude of the overall residual norm, which allows the algorithm to adjust its own parameters according to the actual situation, so as to more accurately optimize the parameters of the parameter-tension prediction model, improve the accuracy of model prediction, and thus improve the effect of gravure printing process parameter optimization.

[0079] In one embodiment of this application, the hyperparameters include inertia weights and learning factors; determining the hyperparameters of the particle swarm optimization algorithm based on the magnitude of the overall residual norm includes: If the overall residual norm is greater than or equal to the preset first residual threshold, the reference value of the inertia weight is reduced based on the first adjustment step size, and the reference value of the learning factor is increased based on the second adjustment step size. If the overall residual norm is less than the preset second residual threshold, the reference value of the inertia weight is increased based on the third adjustment step size, and the reference value of the learning factor is decreased based on the fourth adjustment step size.

[0080] In this embodiment, the inertia weight is one of the hyperparameters of the particle swarm optimization algorithm, representing the degree to which a particle retains the velocity of the previous generation: the larger the inertia weight, the stronger the particle's global search ability, and the better it escapes local optima; the smaller the inertia weight, the stronger the local search ability, and the better it optimizes. The learning factor, also known as the acceleration constant, includes the individual learning factor and the global learning factor, which control the degree to which a particle approaches its own historical best position and the global best position of the population, respectively: the larger the learning factor, the faster the particle converges, but too large a factor can lead to getting trapped in local optima.

[0081] In this embodiment, the first residual threshold and the second residual threshold are critical values ​​for dividing the overall residual norm interval: the first residual threshold corresponds to severe model deviation, and the second residual threshold corresponds to slight model deviation. The first residual threshold being greater than the second residual threshold is the criterion for triggering hyperparameter adjustment. The reference value is the initial baseline value of the inertia weight and learning factor, determined by engineering experience. Hyperparameter adjustments are all based on this value, and it forms the basis for the adjustments. The first adjustment step size, the second adjustment step size, the third adjustment step size, and the fourth adjustment step size are the adjustment ranges of the hyperparameters. The first and third adjustment step sizes correspond to the adjustment range of the inertia weight; the second and fourth adjustment step sizes correspond to the adjustment range of the learning factor. The step size should balance the adjustment effect with the algorithm stability, avoiding excessively large ranges that could cause the algorithm to diverge.

[0082] Specifically, the step size for adjusting the inertia weight (first and third adjustment steps) should be controlled between 5% and 10%. The inertia weight is sensitive to the stability of the algorithm; an excessively large step size can cause the search to go out of control. The step size for adjusting the learning factor (second and fourth adjustment steps) should be controlled between 10% and 20%. The learning factor has a more significant impact on the convergence speed, so the step size can be appropriately increased.

[0083] As can be seen from the above, this embodiment adjusts the inertia weight and learning factor of the particle swarm optimization algorithm according to the magnitude of the overall residual norm. When the overall residual norm is greater than or equal to the preset first residual threshold, the inertia weight reference value is reduced and the learning factor reference value is increased. When the overall residual norm is less than the preset second residual threshold, the inertia weight reference value is increased and the learning factor reference value is reduced. This allows the particle swarm optimization algorithm to better adapt to different residual conditions, more accurately optimize the parameters of the parameter-tension prediction model, and thus improve the accuracy and efficiency of gravure printing process parameter optimization.

[0084] Corresponding to the gravure printing process parameter optimization method based on tension fluctuation analysis in the above embodiment, Figure 2 This is a structural block diagram of a gravure printing process parameter optimization system based on tension fluctuation analysis, provided as an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2The gravure printing process parameter optimization system 20 based on tension fluctuation analysis includes: a data acquisition module 21, a feature extraction module 22, a model prediction module 23, an abnormal parameter module 24, an adjustment and optimization module 25, and a parameter update module 26.

[0085] Among them, the data acquisition module 21 is used to acquire the tension fluctuation signal of the gravure printing machine during operation and simultaneously acquire the corresponding printing process parameters. Feature extraction module 22 is used to preprocess and extract features from tension fluctuation signals to obtain multi-dimensional feature vectors, which include time-domain features and frequency-domain features. The model prediction module 23 is used to input printing process parameters into a preset parameter-tension prediction model to obtain a predicted multi-dimensional feature vector. The abnormal parameter module 24 is used to calculate the residual between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and to identify abnormal process parameters through residual analysis. The adjustment and optimization module 25 is used to adjust and optimize abnormal process parameters based on the physical constraints of printing process parameters, with the optimization goal of minimizing tension fluctuations and registration errors, and to generate an optimized set of process parameters. The parameter update module 26 is used to update the optimized process parameter set to the gravure printing machine for printing process parameter optimization.

[0086] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the data acquisition module 21, feature extraction module 22, model prediction module 23, abnormal parameter module 24, adjustment and optimization module 25, and parameter update module 26 are shown.

[0087] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0088] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0089] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0090] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the gravure printing process parameter optimization method based on tension fluctuation analysis provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0091] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0092] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0093] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0094] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0095] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0096] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0097] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0098] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing gravure printing process parameters based on tension fluctuation analysis, characterized in that, include: The tension fluctuation signal of the gravure printing machine during operation is collected, and the corresponding printing process parameters are collected simultaneously. The tension fluctuation signal is preprocessed and features are extracted to obtain a multi-dimensional feature vector, which includes time-domain features and frequency-domain features. The printing process parameters are input into a preset parameter-tension prediction model to obtain a predicted multi-dimensional feature vector; Calculate the residual between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and identify abnormal process parameters through residual analysis; With the goal of minimizing tension fluctuations and registration errors, and based on the physical constraints of the printing process parameters, an optimization algorithm is used to adjust and optimize the abnormal process parameters, generating an optimized set of process parameters. The optimized set of process parameters is updated to the gravure printing machine to optimize the printing process parameters.

2. The method for optimizing gravure printing process parameters based on tension fluctuation analysis according to claim 1, characterized in that, The preprocessing and feature extraction of the tension fluctuation signal yields a multi-dimensional feature vector, including: The tension fluctuation signal is subjected to outlier removal and moving average filtering to obtain a steady-state tension signal; Based on the steady-state tension signal, the statistical characteristics of the steady-state tension signal are calculated as time-domain characteristics, including the standard deviation, peak-to-peak value, root mean square value, and waveform factor of the signal. The spectrum is obtained by performing a fast Fourier transform on the steady-state tension signal; Extract the energy proportion of the preset characteristic frequency band and the frequency corresponding to the spectral peak with an amplitude greater than the first frequency threshold from the spectrum, and use them as frequency domain features; The time-domain features and the frequency-domain features are combined in a predetermined order to form the multi-dimensional feature vector.

3. The method for optimizing gravure printing process parameters based on tension fluctuation analysis according to claim 2, characterized in that, The step of combining the extracted time-domain features and frequency-domain features in a predetermined order includes: The time-domain features and frequency-domain features are normalized respectively; The normalized time-domain features are arranged in a fixed order of standard deviation, peak-to-peak value, root mean square value, and waveform factor. The normalized frequency-domain features are arranged in ascending order of energy proportion of the characteristic frequency band and significant spectral peak frequency. The two are then combined to form a basic feature vector. Calculate the correlation coefficient between the variance of each component in the basic feature vector and the tension fluctuation; The redundant feature components with variances less than a preset first threshold and correlation coefficients less than a preset second threshold are removed to obtain the multi-dimensional feature vector.

4. The method for optimizing gravure printing process parameters based on tension fluctuation analysis according to claim 1, characterized in that, The calculation of the residual between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and the identification of abnormal process parameters through residual analysis, includes: Arrange the multi-dimensional feature vector and the predicted multi-dimensional feature vector in the same order, subtract them element by element and take the absolute value to obtain the initial residual vector; Each residual value in the initial residual vector is compared with a preset anomaly judgment threshold for the corresponding feature component. Feature components with residual values ​​greater than the corresponding preset anomaly judgment threshold are marked as anomalous features. According to the preset abnormal feature-process parameter lookup table, one or more process parameters directly associated with the abnormal feature are output as the abnormal process parameters.

5. The method for optimizing gravure printing process parameters based on tension fluctuation analysis according to claim 4, characterized in that, Before comparing each residual value in the initial residual vector with a preset anomaly detection threshold for the corresponding feature component, the method further includes: Based on the cumulative operating time and wear status of preset components, the preset abnormality judgment thresholds for each feature component are adjusted; the preset components include printing rollers and scrapers; The cumulative number of revolutions of the printing roller and the historical data of the pressure and angle of the scraper are obtained, and a comprehensive health index is calculated through a preset wear model. If the comprehensive health index is less than the preset health threshold, then the preset abnormality judgment threshold of the corresponding feature component is increased based on the first step.

6. The method for optimizing gravure printing process parameters based on tension fluctuation analysis according to claim 1, characterized in that, The optimization objective is to minimize tension fluctuations and registration errors. Based on the physical constraints of the printing process parameters, an optimization algorithm is used to adjust and optimize the abnormal process parameters, generating an optimized set of process parameters, including: A comprehensive objective function is constructed, which aims to minimize tension fluctuation and minimize registration error. The adjustment range of each abnormal process parameter is determined to obtain the physical constraint; the adjustment range includes the upper and lower limits of the abnormal process parameter itself and the linear or nonlinear coupling relationship between the parameters; Within the search space defined by the physical constraints, a genetic algorithm is used to iteratively search for the abnormal process parameters with the goal of minimizing the comprehensive objective function. When the genetic algorithm reaches the convergence condition, it outputs the currently searched optimal parameter combination as the optimized process parameter set.

7. The method for optimizing gravure printing process parameters based on tension fluctuation analysis according to claim 1, characterized in that, After calculating the residual between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and identifying abnormal process parameters through residual analysis, the method further includes: Calculate the overall residual norm of the multidimensional feature vector and the predicted multidimensional feature vector; If the overall residual norm is greater than a preset model update threshold, then the model parameter optimization requirements for the parameter-tension prediction model are output. In response to the need for model parameter optimization, the parameter-tension prediction model is optimized using a particle swarm optimization algorithm. The parameter-tension prediction model is a neural network model.

8. The method for optimizing gravure printing process parameters based on tension fluctuation analysis according to claim 7, characterized in that, Before optimizing the parameters of the parameter-tension prediction model using the particle swarm optimization algorithm, the method further includes: The search space of the particle swarm optimization algorithm is determined based on the number and value range of the parameters to be optimized in the neural network model. The hyperparameter combination of the particle swarm optimization algorithm is determined based on the magnitude of the overall residual norm.

9. The method for optimizing gravure printing process parameters based on tension fluctuation analysis according to claim 8, characterized in that, The hyperparameters include inertia weights and learning factors; determining the hyperparameters of the particle swarm optimization algorithm based on the magnitude of the overall residual norm includes: If the overall residual norm is greater than or equal to the preset first residual threshold, the reference value of the inertia weight is reduced based on the first adjustment step size, and the reference value of the learning factor is increased based on the second adjustment step size. If the overall residual norm is less than the preset second residual threshold, the reference value of the inertia weight is increased based on the third adjustment step size, and the reference value of the learning factor is decreased based on the fourth adjustment step size.

10. A gravure printing process parameter optimization system based on tension fluctuation analysis, characterized in that, include: The data acquisition module is used to collect tension fluctuation signals during the operation of the gravure printing machine and simultaneously collect the corresponding printing process parameters. The feature extraction module is used to preprocess and extract features from the tension fluctuation signal to obtain a multi-dimensional feature vector, which includes time-domain features and frequency-domain features. The model prediction module is used to input the printing process parameters into a preset parameter-tension prediction model to obtain a predicted multi-dimensional feature vector. The abnormal parameter module is used to calculate the residual between the multi-dimensional feature vector and the predicted multi-dimensional feature vector, and to identify abnormal process parameters through residual analysis. The adjustment and optimization module is used to adjust and optimize the abnormal process parameters based on the physical constraints of the printing process parameters, with the optimization goal of minimizing tension fluctuations and registration errors, and to generate an optimized set of process parameters. The parameter update module is used to update the optimized process parameter set to the gravure printing machine to optimize the printing process parameters.

Images