Online quality monitoring and self-adaptive control method for coated paper production line

Abstract

The invention relates to an online quality monitoring and self-adaptive control method for a coated paper production line. The method comprises the following steps: acquiring a surface image of a coated paper coating, and carrying out microcosmic pinhole defect identification treatment on the surface image to obtain a defect feature list; the defect feature list comprises geometric feature parameters of each microscopic pinhole defect; inputting the defect feature list into a pre-trained hybrid prediction model to obtain a macroscopic moisture permeability prediction value; the hybrid prediction model comprises a physical sub-model based on a gas diffusion physical law and a data driving correction sub-model based on machine learning; and comparing the macroscopic moisture permeability performance predicted value with a preset target performance threshold value, and generating an upstream production process parameter adjustment amount according to a comparison result. By adopting the method, the qualification and stability of the barrier property of the product can be actively maintained in the continuous operation of the production line.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of laminated paper industry, and particularly relates to an online quality monitoring and self-adaptive control method for a laminated paper production line. BACKGROUND

[0002] With the development of high-performance laminated paper production and barrier performance guarantee technology, polyethylene (PE) or co-extrusion barrier material coating technology and machine vision online surface defect detection technology have emerged to provide key oxygen and moisture barrier performance and online quality monitoring means for laminated paper.

[0003] In the traditional method, a core barrier layer is constructed by PE or co-extrusion barrier material coating to resist oxygen and moisture penetration. Microscopic pinholes and through defects of micron level randomly generated in the production process due to impurities in raw materials, unstable melt rheological properties, uneven substrate surface or process parameter fluctuations are monitored online by using a machine vision system to identify basic information such as the number and area of defects, and quality control of the production process is performed in accordance with preset alarm rules to try to avoid the outflow of unqualified products.

[0004] However, the above method can only obtain surface data such as the number and area of defects by machine vision, and cannot quantitatively evaluate how much the specific defects with different sizes, shapes and distributions will cause the product water vapor transmission rate (WVTR) to rise, and lacks quantitative analysis capability of the correlation between defect severity and barrier performance. In addition, the alarm mechanism completely depends on historical experience threshold, has obvious hysteresis and conservatism, and often triggers an alarm after a large number of unqualified products have been produced, or causes excessive downtime adjustment due to too strict threshold setting, resulting in unnecessary quality loss and efficiency waste. Furthermore, the dynamic correlation between defect causes and final performance consequences cannot be established, and the process optimization direction cannot be deduced in reverse through defect data, making it difficult to achieve forward-looking and preventive production control. SUMMARY

[0005] Therefore, it is necessary to provide an online quality monitoring and self-adaptive control method for a laminated paper production line which can improve the consistency and qualified rate of high-value-added laminated paper products and reduce quality loss.

[0006] In a first aspect, the application provides an online quality monitoring and self-adaptive control method for a laminated paper production line, comprising:

[0007] acquiring a surface image of a laminated paper coating and performing micro pinhole defect identification processing on the surface image to obtain a defect feature list; the defect feature list contains geometric feature parameters of each micro pinhole defect;

[0008] The list of defect features is input into a pre-trained hybrid prediction model to obtain the macroscopic moisture permeability prediction value; the hybrid prediction model includes a physical sub-model based on the physical laws of gas diffusion and a data-driven correction sub-model based on machine learning.

[0009] The predicted macroscopic moisture permeability is compared with the preset target performance threshold, and the adjustment amount of upstream production process parameters is generated based on the comparison results.

[0010] In one embodiment, the surface image is processed to identify microscopic pinhole defects, resulting in a list of defect features, including:

[0011] The surface image is sequentially subjected to flat field correction and adaptive contrast enhancement processing to obtain a preprocessed enhanced image; the surface image is a high-resolution original grayscale image of the coating surface acquired under dark field illumination conditions.

[0012] An adaptive thresholding algorithm based on local gray-level statistics is applied to the enhanced image to obtain a binarized image;

[0013] Morphological closing operations are performed on the binarized image to connect adjacent pixels, and a connected component labeling algorithm is used to identify all independent connected regions to obtain candidate defects.

[0014] Candidate defects are filtered according to a preset physical size range to obtain a set of effective defect regions;

[0015] For each defect region in the defect region set, calculate the geometric features to obtain a list of defect features; the geometric features include equivalent diameter, perimeter and roundness.

[0016] In one embodiment, a list of defect features is input into a pre-trained hybrid prediction model to obtain macroscopic moisture permeability prediction values, including:

[0017] The list of defect features is input into the physical sub-model to obtain the individual permeability flux of each micro-pinhole defect. The individual permeability flux of all micro-pinhole defects is summed to obtain the preliminary macroscopic moisture permeability prediction value.

[0018] Statistical features are calculated based on the defect feature list. A comprehensive feature vector is obtained by combining the statistical features with the preliminary macroscopic moisture permeability prediction value and the real-time collected melt temperature and extrusion pressure process parameters. The statistical features include the total number of defects, the average size, and the size distribution variance.

[0019] The comprehensive feature vector is input into the data-driven correction sub-model to obtain the macroscopic moisture permeability prediction value; the data-driven correction sub-model corrects the preliminary macroscopic moisture permeability prediction value.

[0020] In one embodiment, a list of defect features is input into the physical sub-model to obtain the individual permeation flux of each microscopic pinhole defect, including:

[0021] The individual osmotic flux can be obtained using the following formula:

[0022]

[0023] in, The effective diffusion coefficient of water vapor inside the microscopic pinhole defect; This represents the cross-sectional area of ​​a microscopic pinhole defect. The difference in water vapor concentration across the coating is set according to standard test conditions. The depth of the microscopic pinhole defect; The equivalent diameter of the microscopic pinhole defect; This is a dimensionless constant characterizing the diffusion inlet effect.

[0024] In one embodiment, the data-driven correction sub-model is trained using the following method:

[0025] Historical production data is collected to form a training dataset; each training data in the training dataset includes a comprehensive feature vector generated from the surface image of the coated paper and process parameters, and the corresponding true moisture permeability value of the coated paper;

[0026] A data-driven correction sub-model is constructed based on a multi-layer feedforward neural network, and the multi-layer feedforward neural network is supervised and trained using a training dataset to obtain model parameters. The training corresponds to taking the comprehensive feature vector as input and the actual moisture permeability value as the target output, and minimizing the mean square error between the macroscopic moisture permeability performance prediction value and the actual moisture permeability value through the backpropagation algorithm until the model converges.

[0027] The data-driven correction sub-model is determined based on the model parameters.

[0028] In one embodiment, the predicted macroscopic moisture permeability is compared with a preset target performance threshold, and an adjustment amount for upstream production process parameters is generated based on the comparison result, including:

[0029] When the first control threshold is less than or equal to the predicted macroscopic moisture permeability value and less than the second warning threshold, the precise control mode is triggered. The precise control mode uses the deviation between the predicted macroscopic moisture permeability value and the first control threshold as input, and calculates the fine adjustment amount of the melt temperature setpoint through the proportional-integral-derivative control algorithm. The first control threshold is lower than the second warning threshold. The second warning threshold is lower than the product specification limit.

[0030] When the predicted value of macroscopic moisture permeability is greater than or equal to the second warning threshold, the enhanced control mode is triggered. The enhanced control mode corresponds to adding a compensation amount with a preset fixed amplitude on the basis of the fine adjustment amount, and at the same time generating an adjustment command for the matching relationship between the extrusion rate and the traction speed.

[0031] The adjustment command is converted into an industrial communication protocol message and sent to the temperature control module and transmission controller on the production line to drive the actuator to complete the adjustment of process parameters.

[0032] In one embodiment, the method further includes:

[0033] Analyze the size distribution of micro-pinhole defects in the defect feature list. If the distribution of micro-pinhole defects within a preset size range exceeds a preset density threshold, generate a melt temperature feedforward compensation increment.

[0034] The melt temperature feedforward compensation increment and the fine-tuning amount are superimposed to obtain a composite adjustment command; the composite adjustment command is used to instruct the actuator to adjust the process parameters according to the superposition of the melt temperature feedforward compensation increment and the fine-tuning amount.

[0035] Secondly, this application also provides an online quality monitoring and adaptive control device for a coated paper production line, comprising:

[0036] The coated paper defect identification module is used to acquire surface images of the coated paper coating and to identify microscopic pinhole defects in the surface images to obtain a defect feature list; the defect feature list contains the geometric feature parameters of each microscopic pinhole defect;

[0037] The performance prediction module is used to input the list of defect features into the pre-trained hybrid prediction model to obtain the macroscopic moisture permeability prediction value; the hybrid prediction model includes a physical sub-model based on the physical laws of gas diffusion and a data-driven correction sub-model based on machine learning.

[0038] The control module is used to compare the predicted macroscopic moisture permeability with the preset target performance threshold, and generate the adjustment amount of upstream production process parameters based on the comparison result.

[0039] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the online quality monitoring and adaptive control method for any of the above-mentioned coated paper production lines.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the online quality monitoring and adaptive control method for any of the above-described coated paper production lines.

[0041] The online quality monitoring and adaptive control method for the above-mentioned coated paper production line acquires images of the coated paper surface online and identifies and extracts the geometric feature parameters of microscopic pinhole defects to construct a defect feature list. This list, reflecting the microscopic state of the defects, is then input in real time into a pre-trained hybrid prediction model that integrates gas diffusion physics mechanisms and data-driven correction. The model directly calculates the predicted value of the macroscopic moisture permeability of the corresponding coating. Finally, by comparing this predicted value with a preset target threshold in real time and dynamically generating adjustments to upstream process parameters, a coherent closed loop is achieved, from online perception of microscopic defects to real-time prediction of macroscopic performance and then to proactive adaptive control of the production process. This proactively maintains the qualified and stable barrier performance of the product during continuous operation of the production line, effectively overcoming the lag problem of traditional post-event detection and experience-based control. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart illustrating the online quality monitoring and adaptive control method for the coated paper production line of the present invention.

[0044] Figure 2 This is a step-by-step flowchart of step S101;

[0045] Figure 3 This is a flowchart illustrating the steps of step S102.

[0046] Figure 4 This is a structural diagram of the online quality monitoring and adaptive control device for the coated paper production line of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0048] In one embodiment, such as Figure 1 As shown, an online quality monitoring and adaptive control method for a coated paper production line is provided. This embodiment illustrates the method by applying it to a terminal. It is understood that this method can also be applied to a server, or to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0049] S101. Obtain a surface image of the coated paper and perform micro-pinhole defect identification processing on the surface image to obtain a defect feature list; the defect feature list contains the geometric feature parameters of each micro-pinhole defect.

[0050] The image acquisition chamber is schematically positioned at a critical point where the coating has partially cooled and solidified but has not yet been wound up. The chamber integrates a high-resolution linear CCD or CMOS camera. The linear camera is chosen based on the high-speed operation of the production line, and its scanning direction is perpendicular to the paper web's direction, enabling continuous panoramic scanning. Combined with a telecentric lens optical design, it eliminates perspective distortion inherent in traditional lenses, ensuring consistent dimensional accuracy for micron-level defects at different locations within the image. The illumination system employs a coaxial dark-field illumination architecture. Multiple high-brightness LED light sources are arranged in a ring around the lens at an incident angle of nearly 90 degrees, allowing light to be projected almost parallel onto the coating surface. For smooth, defect-free coating areas, the light undergoes specular reflection and cannot enter the vertically arranged lens, resulting in a completely black background. However, when microscopic pinholes, pits, or other defects exist on the surface, the rough sidewalls or bottom of the defects diffusely scatter the incident light. Some of this scattered light is captured by the lens, forming clear, bright spots on the dark background, achieving strong contrast separation between defects and the background.

[0051] After image acquisition, the analog signal is converted into a digital grayscale image sequence using a high-speed analog-to-digital converter. The original image is denoted as [image name missing]. ,in and These represent the horizontal and vertical pixel coordinates of the image, respectively. Preprocessing involves flat-field correction to eliminate camera dark current noise and illumination inhomogeneities from the lighting system. After flat-field correction, dynamic range optimization and filtering are performed to obtain the preprocessed image. Furthermore, although the preprocessed image has improved the contrast between the defect and the background, it still needs to be segmented using a binarization method to separate the defect region from the background. The binarized image may contain pixel breaks due to noise interference or incomplete defect edges. These breaks can segment a previously continuous defect region into multiple discontinuous parts, affecting the accuracy of subsequent defect identification. Optionally, morphological closing operations can be performed on the binarized image, and a connected component labeling algorithm can be used to analyze the image. Further, to eliminate false positives, upper and lower area thresholds can be set based on the physical characteristics of microscopic pinhole defects in the coated paper. Connected components with excessively large areas (potentially non-pinhole defects such as contamination or bubble clusters) or excessively small areas (potentially random noise) can be eliminated, retaining a set of candidate defect regions with equivalent diameters ranging from 1 micrometer to 500 micrometers. Once the effective set of defect regions is determined, the key geometric features of each defect region are extracted to obtain a list of defect features.

[0052] S102. Input the list of defect features into the pre-trained hybrid prediction model to obtain the macroscopic moisture permeability prediction value; the hybrid prediction model includes a physical sub-model based on the physical laws of gas diffusion and a data-driven correction sub-model based on machine learning.

[0053] In a schematic representation, the physical sub-model, based on gas diffusion theory, describes the influence mechanism of microscopic pinhole defects on moisture permeability. A data-driven correction sub-model compensates for deviations between the simplified assumptions of the physical model and actual operating conditions. The physical sub-model is constructed around the calculation of the permeation flux of a single defect. The transport of water vapor through defects in the coated paper is essentially a diffusion process, the mechanism of which is determined by the Knudsen number. Specifically, the Knudsen number is defined as the ratio of the mean free path of gas molecules to the characteristic size of the defect. When the defect pore size is much larger than the mean free path of molecules, Fick's diffusion law applies; when the pore size is close to or smaller than the mean free path of molecules, Knudsen's diffusion law applies. The typical range of coated paper thickness and pinhole size places the diffusion process in a transitional region. For example, a unified diffusion formula is used to describe a single defect. steady-state water vapor mass flow rate ,in, The effective diffusion coefficient of water vapor inside the defect, in units of This parameter comprehensively considers the effects of defect internal roughness, adsorption effect, and diffusion in the transition region, and can be initially estimated in conjunction with experimental data. The cross-sectional area of ​​the defect is expressed in units of 1000 ppm. For irregularly shaped defects, their actual area can be used as an approximation. As an alternative, the lateral dimension of pinhole defects is much smaller than the coating thickness, and the difference between the cross-sectional area and the actual area is negligible; The difference in water vapor concentration across the coating is expressed in units of... This value can be determined by the temperature and humidity conditions of the product's operating environment. When making online predictions, a constant value under industry standard test conditions is used to ensure the comparability of the prediction results. Defect depth, in units of That is, the apparent depth or penetration indicator factor; The diameter of the defect equivalent circle, in units of ; This is the inlet effect correction coefficient, used to characterize the additional diffusion resistance caused by streamline contraction at the defect inlet and outlet. Its value ranges from 0.5 to 2, and the specific value is determined experimentally based on the defect shape and flow state. The term represents the equivalent additional diffusion length, used to correct for the actual length of the diffusion path.

[0054] Furthermore, to establish the correlation between microscopic defects and macroscopic moisture permeability, the moisture permeability of an ideal defect-free coating was calculated. This value is determined by the permeability of the coating material itself. ,thickness and concentration difference Decision, that is ,in The water vapor permeability of the coating material is determined by the material's inherent physical properties and can be obtained through technical parameters provided by the material supplier or through laboratory testing. (Single defect) The contribution to macroscopic moisture permeability is manifested as additional permeable flux density. This refers to the increase in moisture permeability per unit area of ​​the sample caused by this defect, which is calculated as the water vapor mass flow rate of a single defect. Sample area in the current analysis region ratio Alternatively, defects can be viewed as additional parallel permeation channels, introducing permeability. Characterizing the penetration ability of a single defect, i.e. This parameter is independent of the concentration difference and is determined solely by the geometry of the defect and the properties of the diffusion medium. Furthermore, the permeability is converted into an equivalent additional permeable area. Even if the penetration contribution of this defect is equivalent to an area of The penetration effect of ideal defects in defect-free coatings, i.e. The total additional effective penetration area is obtained by summing the equivalent additional penetration areas of all defects. Based on the moisture permeability of an ideal, defect-free coating, the total moisture permeability predicted by the physical sub-model is [value missing]. This directly reflects the cumulative effect of defects on moisture permeability.

[0055] The role of the data-driven correction sub-model is to compensate for deviations between the simplifying assumptions of the physical sub-model and actual operating conditions. It directly modifies the predictions made by the physical model. As a foundation, defect statistical characteristics are used as input. These characteristics encompass macroscopic distribution information, including the total number of defects reflecting defect density. The average diameter, reflecting the overall size level of the defect; the standard deviation of the diameter, reflecting the degree of dispersion in defect size; and the percentage of the total area covered by the defect. And the defect spatial distribution uniformity index, which is calculated by dividing the sample area into several equal-area grids and calculating the coefficient of variation of the number of defects in each grid. The results show that a larger coefficient of variation indicates a more uneven defect distribution, and locally concentrated defects may lead to a nonlinear deterioration of moisture permeability. Optionally, process parameters, specifically melt temperature, can also be incorporated as contextual features into the input. It affects melt viscosity and surface tension, thus altering the probability of defect formation; extruder head pressure Affects the uniformity of coating formation; substrate tension Affects substrate smoothness; coating speed The input feature vector is affected by the coating's cooling and curing rate. .

[0056] Indicatively, the data-driven correction sub-model employs a 3-5 layer fully connected feedforward neural network (MLP). The number of neurons in the network input layer and the feature vector... The dimensions are consistent, including hidden neural layers. The ReLU activation function is used to avoid the gradient vanishing problem. The output layer is a single neuron, directly outputting the final macroscopic moisture permeability prediction value. To achieve mapping relationship Specifically, the list of defect features generated in real time. The physical sub-model is called for parallel computation, and the effective diffusion coefficient estimated in the initial step is substituted into it. and the correction coefficient for the inlet effect Calculate the value of each defect in turn. , , Summing yields And then calculate Simultaneously, the statistical characteristics of defects are calculated in real time. And read the current process parameters from the production line control system via the industrial bus. ,and They are assembled together into the input feature vector. ;Will The input is fed into a pre-trained correction neural network, which quickly completes inference through forward propagation and outputs the final macroscopic moisture permeability prediction value. .

[0057] S103. Compare the predicted macroscopic moisture permeability with the preset target performance threshold, and generate the adjustment amount of upstream production process parameters based on the comparison results.

[0058] Indicative, clearly defining product specification thresholds This refers to the upper limit of moisture permeability required by the market or customers, and the control target threshold is set based on this. ,Right now A safety margin of 10% to 30% can be reserved. This margin is used to offset the delayed effects of process adjustments, minor fluctuations in raw materials, and model prediction errors, preventing moisture permeability from exceeding the standard due to sudden disturbances. At the same time, a lower limit for the comfort zone should be set. This creates three control zones, for example, the comfort zone. Warning Zone and areas exceeding standards .

[0059] Optionally, the control strategy adopts a feedforward-feedback composite control mode. Feedforward compensation focuses on early intervention. Specifically, early regulation is achieved based on an empirical database of process parameters and defect trends. This database is constructed through historical production data mining and process experiments, collecting a large amount of historical production data, including combinations of process parameters, corresponding defect characteristic trends, and final moisture permeability. Association rule mining algorithms are used to analyze the correlation between process parameters and defect types. For example, it was found that when the melt temperature exceeds a certain threshold, the number of large-size pinhole defects increases significantly, while when the substrate tension fluctuation exceeds ±5%, the proportion of small-size dense defects increases. Simultaneously, orthogonal process experiments are designed to study the impact of changes in single process parameters on defect types, supplementing the database with correlation information for key parameters. The rules stored in the database are presented in the form of "if the defect trend is A, then adjust the process parameter B to range C". When a certain type of defect trend is detected online and meets the rule triggering conditions, even if the current moisture permeability prediction value is still in the comfort zone, the fine-tuning instruction is output according to the rule base. For example, when small-sized dense defects increase, the substrate tension is finely adjusted so that its real-time value moves closer to the historical optimal range. Early intervention is used to suppress the further development of defects and prevent the moisture permeability from entering the warning zone.

[0060] Furthermore, feedback control is the core control element, employing a multiple-input multiple-output (MIMO) fuzzy logic controller. This controller is suitable for nonlinear industrial processes with strong time delays, and the input variables include the current error. and error change rate ,in, For the current control cycle, For the previous control cycle, the error Reflects the degree of deviation between the current moisture permeability and the control target, the rate of error change. This reflects the deviation trend; positive indicates a decrease in deviation, and negative indicates an increase in deviation. Simultaneously, the current key process parameters ( The actual value of the input is used as an auxiliary input to correct the amplitude of the control output. For example, when the melt temperature is close to the upper limit, the temperature adjustment amount is appropriately reduced to avoid exceeding the safe operating range of the equipment. The input variables need to be fuzzified, converting precise numerical values ​​into fuzzy linguistic variables, each corresponding to a specific membership function.

[0061] The controller's output variables are the adjustments to various process parameters, including the melt temperature adjustment. Screw speed and traction speed matching adjustment amount Die head pressure adjustment amount Substrate tension adjustment amount The construction of the fuzzy rule base is based on process mechanisms and operational experience, such as "if..." For the upright and If the negative is small, then For negative small, "Negatively low" indicates that the current moisture permeability is far below the control target and the deviation is decreasing. A slight reduction in the melt temperature and the ratio of screw speed to traction speed is needed to avoid over-adjustment that could lead to a reverse increase in defects. "If..." For negative small and If it is negative, then For positive small, "Positive value: Small" indicates that the current moisture permeability is slightly higher than the control target and is still rising. A slight increase in melt temperature is needed to improve melt flowability, while a slight increase in substrate tension is needed to improve coating smoothness. The rule base must contain the output strategies corresponding to all combinations of input fuzzy linguistic variables to ensure coverage of various operating conditions. Through declarative processing, the fuzzy output quantities are converted into precise adjustment values.

[0062] Optionally, the control decision-making and execution process needs to consider the dynamic characteristics and delay effects of the production line. The controller's control cycle is set to 1 second, synchronized with the image acquisition and moisture permeability prediction cycle, to ensure real-time decision-making. For example, a state judgment is first performed, i.e., when... Furthermore, when the process parameters are near the median of their historical optimal range, the current process parameters should be kept unchanged to avoid production fluctuations caused by frequent adjustments; when Upon entering the warning zone, the feedback controller is activated, according to... and Calculate small, incremental adjustments, keeping the adjustment range within 5% of the allowable fluctuation range of process parameters. For example, a single adjustment of melt temperature should not exceed ±3°C, and a single adjustment of substrate tension should not exceed ±5N. Ensure smooth adjustments to avoid drastic changes in coating quality. When this happens, the controller enters emergency intervention mode, increasing the adjustment range to 5%~10% of the allowable fluctuation range, simultaneously triggering an audible and visual alarm, and displaying a warning message on the human-machine interface, prompting the operator to pay attention to potential influencing factors such as raw material quality and equipment operating status; if the predicted value continues to exceed [a certain value] for 3 control cycles... It automatically generates product downgrade or isolation instructions, and marks the products in that section through the marking device on the production line to prevent unqualified products from flowing into the downstream process.

[0063] Commands are issued via an industrial Ethernet bus. Melt temperature adjustment commands are sent to the temperature control modules of each temperature zone of the extruder, achieving precise control of heating power through PID regulation. Screw speed and traction speed matching adjustment commands are sent to the frequency converter, adjusting the speed by changing the motor frequency. Die pressure adjustment commands are implemented by controlling the die gap adjustment mechanism or the melt pump speed. Substrate tension adjustment commands are sent to the tension controllers of the unwinding and rewinding drive motors, maintaining tension stability through closed-loop regulation. Because changes in coating quality after process parameter adjustments require feedback through coating formation, cooling and curing, image acquisition, and moisture permeability prediction, there is a process delay. The controller needs to integrate delay compensation logic and adopt the Smith predictor algorithm. By establishing a process delay model, it can predict the effect of adjustment commands in advance, correct the current control output, and avoid overshoot or oscillation caused by delay.

[0064] Optionally, for each shift, the online predicted moisture permeability data and the laboratory sampled measured moisture permeability data are collected, and the average prediction error is calculated. ,in This refers to the number of samples to be inspected. When the average prediction error exceeds the preset allowable range, the incremental learning of the data-driven correction sub-model is triggered. The model is fine-tuned using newly collected labeled data, the weights of the bottom layer network are frozen, and only the weights of the top layer are updated. This prevents the model from forgetting historical knowledge and allows the model to continuously adapt to changes in operating conditions such as raw material batches and equipment wear.

[0065] In the above-mentioned online quality monitoring and adaptive control method for coated paper production line, the surface image of the coated paper coating is first acquired, and microscopic pinhole defects are identified to obtain a defect feature list containing the geometric feature parameters of each defect. Then, the defect feature list is input into a pre-trained hybrid prediction model composed of a physical sub-model based on the physical laws of gas diffusion and a data-driven correction sub-model based on machine learning, thereby accurately obtaining the macroscopic moisture permeability prediction value. Finally, the prediction value is compared with the preset target performance threshold, and the upstream production process parameter adjustment amount is generated based on the comparison result. This realizes the quantitative correlation between microscopic defect features and macroscopic barrier performance, achieves forward-looking and preventive production control, effectively improves product quality stability, and reduces unnecessary quality losses and production efficiency waste.

[0066] In one embodiment, such as Figure 2 As shown, the surface image is processed to identify microscopic pinhole defects, resulting in a list of defect features, including:

[0067] S201. Perform flat field correction and adaptive contrast enhancement processing on the surface image in sequence to obtain the preprocessed enhanced image; the surface image is a high-resolution original grayscale image of the coating surface obtained under dark field illumination conditions.

[0068] To illustrate, the flattening correction process requires the use of pre-acquired standard uniform whiteboard images. and dark current image The correction formula is ,in is the gain constant, used to map the corrected image grayscale values ​​to the standard grayscale range (0-255). The standard whiteboard image is acquired under lighting conditions consistent with production conditions to ensure that it can truly reflect the image response under ideal uniform lighting, while the dark current image is acquired under conditions where the camera is blocked and there is no light, used to characterize the noise level of the camera sensor itself.

[0069] Furthermore, an adaptive histogram equalization (CLAHE) algorithm is employed to enhance local contrast. This involves dividing the image into multiple non-overlapping sub-regions and performing histogram equalization on each sub-region separately, avoiding local overexposure or underexposure issues that may occur with global histogram equalization. Subsequently, a 3x3 window median filter is used for noise smoothing. Median filtering, by replacing the center pixel value with the median grayscale value of pixels within the window, effectively suppresses random noise such as salt-and-pepper noise. Simultaneously, due to the small window size, the edge sharpness of pinhole defects is preserved to the maximum extent, avoiding distortion of defect morphology.

[0070] S202. Apply an adaptive threshold segmentation algorithm based on local gray-level statistics to the enhanced image to obtain a binarized image.

[0071] For example, an adaptive thresholding segmentation algorithm based on local gray-level mean and standard deviation is used. The entire image is traversed through a sliding window, and for each pixel within the window, its local gray-level mean is calculated. and standard deviation And dynamically set the threshold. ,in, To adjust the coefficients, based on the grayscale characteristics of the defects, pixels with grayscale values ​​below a threshold are identified as background, while pixels with grayscale values ​​above the threshold are identified as potential defect areas, thus generating a binary image. .

[0072] S203. Perform morphological closing operation on the binarized image to connect adjacent pixels, and use a connected component labeling algorithm to identify all independent connected regions to obtain candidate defects.

[0073] Furthermore, morphological closing operations are performed on the binary image, specifically dilation followed by erosion. Dilation fills in small holes caused by noise within the defect area, while erosion eliminates tiny protrusions at the defect edges, bridging minor breaks in the defect area and forming a complete and continuous region for each defect. Optionally, an eight-neighbor connected component labeling algorithm is used to analyze the binary image. This algorithm traverses all pixels in the image, marking interconnected white pixel regions as the same connected component, assigning a unique label to each component, and then calculating the basic attributes of each connected component, including area in pixels. Circumscribed rectangle parameters, centroid coordinates .

[0074] S204. Based on the preset physical size range, candidate defects are screened to obtain a set of effective defect areas.

[0075] Optionally, the candidate defects may include areas that do not conform to the physical characteristics of actual microscopic pinhole defects, such as tiny areas formed by noise accumulation or large areas caused by coating surface contamination. The screening is based on a preset physical size range, determined by the actual engineering characteristics of microscopic pinhole defects, covering a reasonable range of key physical parameters such as the equivalent diameter and area of ​​the defect. Specifically, the pixel size of the candidate defects is converted to actual physical size. For example, based on the camera resolution at the time of image acquisition, i.e., the actual physical size corresponding to a single pixel, the pixel area, pixel diameter, and other parameters of the candidate defects are converted into actual area and diameter. The converted actual physical parameters are compared with the preset range, and candidate defects exceeding the range are eliminated. For example, areas with an actual equivalent diameter less than the preset lower limit are determined to be false defects caused by noise interference and are eliminated; areas with an actual equivalent diameter greater than the preset upper limit are determined to be non-pinhole defects such as coating contamination and bubble clusters and are also eliminated. Through this physical size-based screening, only areas conforming to the physical characteristics of microscopic pinhole defects are retained, forming an effective set of defect areas.

[0076] S205. Calculate the geometric features for each defect region in the defect region set to obtain a list of defect features; the geometric features include equivalent diameter, perimeter and roundness.

[0077] Furthermore, for each candidate defect region Calculate the equivalent circle diameter ,in The actual area of ​​the defect is calculated by multiplying the pixel area of ​​the connected region by the actual physical area of ​​a single pixel. A boundary tracking algorithm is used to traverse the edge pixels of the defect region, calculating the pixel perimeter of the defect contour, and then converting it to the physical perimeter based on the actual pixel size. Circularity Used to characterize the degree to which the shape of a defect closely resembles a circle, i.e. The value ranges from 0 to 1, with values ​​closer to 1 indicating a defect shape closer to a circle. Elongation Defined as the ratio of the major axis to the minor axis of the minimum bounding ellipse of the defect region, it is calculated after obtaining the lengths of the major and minor axes through a minimum bounding ellipse fitting algorithm, and is used to distinguish between circular and elongated defects. Furthermore, the effective penetration width is... As a morphological characteristic quantity, when the roundness of the defect When the value exceeds a set threshold, it is determined to be an approximately circular defect. When roundness If the value is less than or equal to a set threshold, it is judged as an irregular or elongated defect, and the perimeter is used directly. As the effective penetration width. The apparent depth or penetration indicator of the defect. The method combines offline calibration with online hypothesis testing. Artificial defect samples of known depth are prepared, and their dark-field images are acquired under identical detection conditions. A mapping relationship between the average grayscale value of the defect and its depth is established. During online detection, this mapping relationship is used to estimate the depth. If calibration data is lacking, it can be preliminarily assumed that all detected pinholes are penetrating defects. equal to coating thickness This assumption can be corrected in subsequent hybrid models through a data-driven component. Finally, a defect feature list is formed by integrating the feature parameters of all defects for the currently acquired image frames. ,in .

[0078] In one embodiment, such as Figure 3 As shown, the list of defect features is input into a pre-trained hybrid prediction model to obtain macroscopic moisture permeability prediction values, including:

[0079] S301. Input the list of defect features into the physical sub-model to obtain the individual permeability flux of each microscopic pinhole defect, and sum the individual permeability flux of all microscopic pinhole defects to obtain the preliminary macroscopic moisture permeability prediction value.

[0080] In a schematic physical sub-model, each pinhole defect is treated as an independent parallel permeation channel. Water vapor transport through the defects follows the diffusion law of the transition zone. Taking into account the combined effects of Fick diffusion and Knudsen diffusion, the individual permeation flux is obtained. The calculation of the individual permeation flux is based on the geometric characteristics of the defect. The effective space of the permeation channel is reflected by the equivalent cross-sectional area of ​​the defect. The diffusion path length is characterized by the defect depth and the inlet effect correction term. Combining the effective diffusion coefficient of water vapor inside the defect and the concentration difference on both sides of the coating, the steady-state water vapor mass flow rate of a single defect, i.e., the individual permeation flux, is obtained.

[0081] To obtain the individual permeation flux of each microscopic pinhole defect. Then, the total permeation flux of all defects in the current analysis area is obtained by summation. ,in This represents the total number of defects in the defect characteristic list. Converting the total permeability flux to permeability per unit area yields the preliminary macroscopic permeability performance prediction. .

[0082] S302. Calculate statistical features based on the defect feature list, and obtain a comprehensive feature vector by combining the statistical features with the preliminary macroscopic moisture permeability prediction value and the real-time collected melt temperature and extrusion pressure process parameters. The statistical features include the total number of defects, average size, and size distribution variance.

[0083] The calculation of statistical characteristics aims to extract the distribution characteristics of defect groups from a macroscopic perspective, compensating for the group effects that cannot be reflected by the geometric parameters of individual defects. Total number of defects. The defect density is obtained by counting directly from the defect feature list. Generally, the higher the defect density, the more significant the negative impact on moisture permeability. The average size is calculated using the arithmetic mean of the equivalent diameters of all defects. Size distribution variance is used to describe the dispersion of the equivalent diameter of a defect. The larger the variance, the more obvious the difference in defect size. There may be a situation where a small number of large-sized defects dominate the overall moisture permeability. This nonlinear effect needs to be included in the model through statistical characteristics.

[0084] The preliminary macroscopic moisture permeability prediction, as the output of the physical model, carries core information based on defect geometry and diffusion theory, while the introduction of real-time process parameters provides the model with operating context. Melt temperature and extrusion pressure are key process parameters affecting coating quality: melt temperature directly affects melt viscosity and surface tension; abnormal temperatures may lead to poor melt flow or degradation, thus forming specific types of pinhole defects; the stability of extrusion pressure affects the uniformity of coating thickness; pressure fluctuations can easily cause localized weak areas in the coating, indirectly affecting the penetration characteristics of defects. Both process parameters can be acquired in real time by sensors on the production line, with the sampling frequency synchronized with the image acquisition cycle to ensure the temporal correlation between parameters and defect characteristics.

[0085] The construction of the comprehensive feature vector organically integrates defect group statistical information, preliminary prediction results from the physical model, and real-time operating parameters to form input data with complete dimensions and complementary information, specifically consisting of... ,in For real-time melt temperature, This refers to the real-time extrusion pressure.

[0086] S303. Input the comprehensive feature vector into the data-driven correction sub-model to obtain the macroscopic moisture permeability prediction value; the data-driven correction sub-model corrects the preliminary macroscopic moisture permeability prediction value.

[0087] The physical sub-model idealizes defect morphology and internal states during its derivation, potentially leading to discrepancies between initial predictions and actual moisture permeability. The data-driven correction sub-model, by learning from extensive measured data, captures the nonlinear effects of these complex factors. (Illustratively, the constructed comprehensive feature vector is shown.) The input is fed into a pre-trained modified sub-model, which then outputs the final macroscopic moisture permeability prediction value through forward propagation. The essence of the correction process is that the model dynamically calibrates the initial predictions of the physical sub-model based on multi-dimensional information in the comprehensive feature vector. Specifically, when statistical features show that the defect size distribution has large dispersion or the process parameters deviate from the optimal range, the model will appropriately adjust the initial predictions according to the rules learned during training to match the influence of complex factors in the actual scenario. The final output macroscopic moisture permeability prediction value retains the theoretical interpretability of the physical model and integrates the generalization ability of the data-driven model, achieving high-precision real-time prediction of the macroscopic moisture permeability of coated paper.

[0088] In one embodiment, a list of defect features is input into the physical sub-model to obtain the individual permeation flux of each microscopic pinhole defect, including:

[0089] The individual osmotic flux can be obtained using the following formula:

[0090]

[0091] in, The effective diffusion coefficient of water vapor inside the microscopic pinhole defect; This represents the cross-sectional area of ​​a microscopic pinhole defect. The difference in water vapor concentration across the coating is set according to standard test conditions. The depth of the microscopic pinhole defect; The equivalent diameter of the microscopic pinhole defect; This is a dimensionless constant characterizing the diffusion inlet effect.

[0092] In one embodiment, the data-driven correction sub-model is trained using the following method:

[0093] S11. Collect historical production data to form a training dataset; each training data in the training dataset includes a comprehensive feature vector generated from the surface image of the coated paper and process parameters, and the corresponding true moisture permeability value of the coated paper.

[0094] Indicatively, the data collection process needs to cover typical operating conditions of the production line, including different raw material batches, melt temperature gradients, extrusion pressure ranges, coating speed levels, etc., to capture the correlation between defect characteristics, process parameters, and moisture permeability under diverse operating conditions. Each training dataset must satisfy a one-to-one correspondence between input and output. That is, the generation of the comprehensive feature vector must be synchronously linked to the surface image analysis results of the coated paper and real-time process parameters. The surface image is processed to extract defect statistical features, combined with the preliminary macroscopic moisture permeability prediction value output by the physical sub-model, and then integrated with the key process parameters collected in real-time during production to jointly constitute a comprehensive feature vector with complete dimensions. This ensures that the vector can fully carry the characteristics of the defect group, the preliminary judgment of the physical model, and the background information of the operating conditions. The corresponding actual moisture permeability value needs to be obtained through standard laboratory testing methods, using industry-recognized cup method or infrared sensor method to measure the moisture permeability of the collected coated paper samples. For example, the cup method calculates the moisture permeability by monitoring the desiccant weight gain rate under constant temperature and humidity conditions, while the infrared sensor method utilizes the absorption characteristics of water vapor to specific wavelengths of infrared light to achieve real-time detection. The measured actual moisture permeability value serves as the target label for model training. Optionally, the collected historical production data needs to be preprocessed to remove outliers and ensure the reliability of the training dataset.

[0095] S12. Construct a data-driven correction sub-model based on a multi-layer feedforward neural network, and use the training dataset to supervise the training of the multi-layer feedforward neural network to obtain the model parameters; the training corresponds to taking the comprehensive feature vector as input and the actual moisture permeability value as the target output, and minimizing the mean square error between the macroscopic moisture permeability prediction value and the actual moisture permeability value through the backpropagation algorithm until the model converges.

[0096] The model training employs a supervised learning model. The training process uses a comprehensive feature vector as input data and the corresponding true moisture permeability as the target output. The network parameters are iteratively optimized using a backpropagation algorithm. The core objective of the training is to minimize the mean squared error between the model's predicted values ​​and the true moisture permeability values. The mean squared error loss function... ,in, The number of samples in the training dataset, For the model to the first The predicted permeability value for each sample. For the first The true moisture permeability value of each sample. The backpropagation algorithm calculates the gradient of the loss function with respect to the parameters of each network layer, updates the parameters along the gradient descent direction, and gradually reduces the prediction error. During training, a convergence criterion is set: when the loss function value of several consecutive iterations is lower than a preset threshold, or when the number of iterations reaches a preset upper limit, the model training is considered to have converged, and parameter updates are stopped. The network parameters obtained at this time are the parameters of the trained model.

[0097] S13. Determine the data-driven correction sub-model based on the model parameters.

[0098] After the model training converges, the final network parameters, including the weight matrices between the input and hidden layers, between each hidden layer, and between the hidden and output layers, as well as the bias vectors of each neuron, are permanently stored. Further, once the model is finalized, offline validation is required. A reserved validation dataset is used to test the model's prediction accuracy, ensuring that the model maintains stable correction effects even under unseen operating conditions. After successful validation, the model can be deployed to the real-time computing unit on the production line to achieve dynamic and accurate correction of the initial predictions from the physical sub-model.

[0099] In one embodiment, the predicted macroscopic moisture permeability is compared with a preset target performance threshold, and an adjustment amount for upstream production process parameters is generated based on the comparison result, including:

[0100] S21. When the first control threshold ≤ the macroscopic moisture permeability prediction value < the second warning threshold, the precise control mode is triggered; the precise control mode uses the deviation between the macroscopic moisture permeability prediction value and the first control threshold as input, and calculates the fine adjustment amount of the melt temperature setpoint through the proportional-integral-derivative control algorithm; the first control threshold is lower than the second warning threshold; the second warning threshold is lower than the product specification limit.

[0101] As an illustration, the first control threshold, the second warning threshold, and the product specification limit should follow a hierarchical relationship: first control threshold < second warning threshold < product specification limit. The first control threshold, serving as the initiation point for preventative control, is set at 70%–80% of the product specification limit, reserving sufficient buffer space for adjustment. The second warning threshold, serving as the trigger boundary for enhanced intervention, is set at 85%–90% of the product specification limit to prevent further deterioration of moisture permeability to the point of exceeding the limit. When the predicted macroscopic moisture permeability falls between the first control threshold and the second warning threshold, it indicates that although the current product moisture permeability is not exceeding the limit, it has deviated from the ideal range and requires slight correction through precise control mode to avoid production condition fluctuations due to excessive adjustments.

[0102] Specifically, the core of the precise control mode is the use of a proportional-integral-derivative (PID) control algorithm. The input to the algorithm is the deviation between the predicted macroscopic moisture permeability and the first control threshold. ,in This is a real-time predicted macroscopic moisture permeability value. The first control threshold is set at 100°C. The PID algorithm uses a proportional (P) component to output an adjustment amount in real time based on the magnitude of the deviation, quickly responding to deviations in moisture permeability; an integral (I) component accumulates historical deviations, gradually eliminating static errors and ensuring that moisture permeability stably returns to the target range; and a derivative (D) component predicts trends based on the rate of change of deviation, applying reverse adjustments in advance to suppress overshoot. Through the synergistic effect of these three components, the algorithm calculates the fine-tuning amount of the melt temperature setpoint. The magnitude of the fine-tuning is strictly controlled within ±2% to ±3% of the allowable fluctuation range of the melt temperature. For example, when the melt temperature setpoint is 200°C, the single fine-tuning amount does not exceed ±4°C to ±6°C, ensuring a smooth and gradual adjustment and avoiding drastic changes in melt viscosity due to sudden temperature changes, which could lead to new coating defects.

[0103] S22. When the predicted value of macroscopic moisture permeability is greater than or equal to the second warning threshold, the enhanced control mode is triggered. The enhanced control mode corresponds to adding a compensation amount with a preset fixed amplitude on the basis of the fine adjustment amount, and at the same time generating an adjustment command for the matching relationship between the extrusion rate and the traction speed.

[0104] When the predicted macroscopic moisture permeability reaches or exceeds the second warning threshold, it indicates that the current moisture permeability is approaching the product specification limit. Simple small temperature adjustments are unlikely to quickly reverse the trend, and a more targeted intervention using an enhanced control mode is necessary. The core logic of the enhanced control mode is a combination of basic fine-tuning, compensation enhancement, and coordinated adjustment. First, the melt temperature fine-tuning amount calculated using the PID algorithm is used. On this basis, a preset fixed-amplitude compensation amount is superimposed. The compensation amount needs to be set in conjunction with process experimental data, and is usually 50% to 100% of the fine-tuning amount. The total adjustment range does not exceed ±5% to ±8% of the allowable fluctuation range of the melt temperature, ensuring both the intensity of intervention and avoiding exceeding the safe operating range of the equipment. At the same time, the enhanced control mode needs to generate adjustment instructions for the matching relationship between the extrusion rate and the traction speed to directly affect the forming quality of the coating. The ratio of the extrusion rate to the traction speed determines the theoretical thickness of the coating and the shear stress on the melt. If the ratio is too large, it can easily lead to an excessively thick coating and internal stress concentration, resulting in defects. If the ratio is too small, it may cause an excessively thin coating and localized missed coatings. The adjustment commands are implemented by fine-tuning the screw speed to change the extrusion rate or the traction motor frequency to change the traction speed. For example, when the moisture permeability is too high, the extrusion rate is appropriately reduced or the traction speed is increased to bring the ratio of extrusion rate to traction speed closer to the historical optimal range, optimizing the uniformity and density of the coating and reducing the formation of pinhole defects at the source. The coordinated implementation of temperature adjustment and speed matching adjustment forms a dual intervention mechanism to quickly suppress the upward trend of moisture permeability and prevent the product from being downgraded or isolated due to excessive moisture permeability.

[0105] S23. The adjustment command is converted into an industrial communication protocol message and sent to the temperature control module and transmission controller of the production line to drive the actuator to complete the process parameter adjustment.

[0106] This example illustrates how melt temperature adjustment commands are matched with extrusion rate-traction speed adjustment commands. These commands are converted according to industrial communication protocols such as PROFINET and EtherCAT. The chosen protocol must meet the real-time requirements of the production line, ensuring that command transmission delays are controlled within milliseconds to prevent control failures due to transmission lag. The converted message contains key information such as target parameter settings, adjustment rates, and execution priorities. The adjustment rate controls the smoothness of the actuator's movement; for example, a melt temperature adjustment rate of 1°C / s prevents sudden temperature rises and falls, while a traction speed adjustment rate of 0.5 m / min² ensures smooth paper web operation. The converted protocol message is sent to the production line's temperature control module and drive controller via an industrial Ethernet bus. Upon receiving the melt temperature adjustment command, the temperature control module uses its internal PID control logic to control the power output of the heating unit until the target setting is reached. The drive controller, upon receiving the extrusion rate and traction speed adjustment commands, adjusts the corresponding motor inverter frequency, changing the screw speed or traction motor speed. Real-time speed feedback signals are collected via encoders to form a closed-loop control, ensuring precise speed parameter matching.

[0107] In one embodiment, the method further includes:

[0108] S31. Analyze the size distribution of micro-pinhole defects in the defect feature list. If the distribution of micro-pinhole defects within the preset size range exceeds the preset density threshold, generate a melt temperature feedforward compensation increment.

[0109] Defect size distribution analysis can identify potential risks that may degrade moisture permeability by mining the size clustering characteristics of defect groups, thus compensating for the lag in feedback control. Specifically, based on the characteristics of the coating material and production process experience, a preset size range is defined. This range typically focuses on the defect size interval that is most sensitive to the impact on moisture permeability, such as medium-sized pinholes of 10 to 50 micrometers. These defects have a certain permeation channel area and are easily generated in batches due to fluctuations in process parameters. Based on the equivalent diameter parameter in the defect feature list, all valid defects are categorized and statistically analyzed by size. The proportion of defects within the preset size range to the total number of defects is calculated, or the unit area distribution density of defects within this range is determined.

[0110] The preset density threshold needs to be set based on historical production data and process experiment results. It is determined by analyzing the change pattern of moisture permeability when defects are dense in this size range. For example, when the proportion of defects in the preset size range exceeds 30%, or the distribution density per unit area exceeds 50 defects / square meter, it is judged to meet the density condition. When the detection result exceeds this density threshold, it indicates that a trend of batch generation of defects of a specific size has appeared in the current production process. If only subsequent feedback adjustments are relied upon, defects may continue to accumulate due to process delays. At this time, it is necessary to generate a melt temperature feedforward compensation increment. The value of the compensation increment is dynamically adjusted according to the degree of defect density. That is, the higher the density, the larger the compensation increment, but it must be strictly controlled within ±1% to ±2% of the safe fluctuation range of melt temperature to avoid new process fluctuations caused by excessive feedforward adjustment.

[0111] S32. The melt temperature feedforward compensation increment and the fine-tuning amount are superimposed to obtain a composite adjustment command; the composite adjustment command is used to instruct the actuator to adjust the process parameters according to the superposition of the melt temperature feedforward compensation increment and the fine-tuning amount.

[0112] Indicatively, when feedforward compensation is triggered, the generated melt temperature feedforward compensation increment is algebraically superimposed with the fine-tuning amount calculated by the PID algorithm. If the two adjustments are in the same direction (e.g., both increasing or both decreasing temperature), the total adjustment after superposition is the sum of their absolute values. If the adjustments are in opposite directions, the feedforward compensation increment is used to correct the magnitude of the fine-tuning amount, ensuring that the total adjustment conforms to the process optimization direction. An upper limit for the total adjustment must be set during the superposition process to avoid exceeding the allowable fluctuation range of the melt temperature and to ensure the stability of the production process.

[0113] After the composite adjustment command is generated, its execution priority needs to be clarified. That is, the superposition result of the feedforward compensation increment and the fine adjustment amount is used as the final melt temperature adjustment target. The command must include key information such as the superimposed target set value and adjustment rate. After the command is sent to the temperature control module of the production line, the actuator will adjust the temperature according to the superimposed adjustment amount. This not only suppresses the batch generation of sensitive size defects in advance through feedforward compensation, but also accurately corrects the deviation of the current moisture permeability through feedback fine adjustment, thereby achieving synergistic optimization of process parameters.

[0114] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0115] Based on the same inventive concept, this application also provides an online quality monitoring and adaptive control device for a coated paper production line, used to implement the online quality monitoring and adaptive control method for the coated paper production line described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the online quality monitoring and adaptive control device for coated paper production lines provided below can be found in the limitations of the online quality monitoring and adaptive control method for coated paper production lines described above, and will not be repeated here.

[0116] In one exemplary embodiment, such as Figure 4 As shown, an online quality monitoring and adaptive control device for a coated paper production line is provided, comprising:

[0117] The coated paper defect identification module 401 is used to acquire a surface image of the coated paper coating and to identify micro-pinhole defects in the surface image to obtain a defect feature list; the defect feature list contains the geometric feature parameters of each micro-pinhole defect.

[0118] The performance prediction module 402 is used to input the list of defect features into the pre-trained hybrid prediction model to obtain the macroscopic moisture permeability prediction value; the hybrid prediction model includes a physical sub-model based on the physical laws of gas diffusion and a data-driven correction sub-model based on machine learning.

[0119] The control module 403 is used to compare the predicted value of macroscopic moisture permeability with the preset target performance threshold, and generate the adjustment amount of upstream production process parameters based on the comparison result.

[0120] In one embodiment, the coated paper defect identification module 401 is further configured to:

[0121] The surface image is sequentially subjected to flat field correction and adaptive contrast enhancement processing to obtain a preprocessed enhanced image; the surface image is a high-resolution original grayscale image of the coating surface acquired under dark field illumination conditions.

[0122] An adaptive thresholding algorithm based on local gray-level statistics is applied to the enhanced image to obtain a binarized image;

[0123] Morphological closing operations are performed on the binarized image to connect adjacent pixels, and a connected component labeling algorithm is used to identify all independent connected regions to obtain candidate defects.

[0124] Candidate defects are filtered according to a preset physical size range to obtain a set of effective defect regions;

[0125] For each defect region in the defect region set, calculate the geometric features to obtain a list of defect features; the geometric features include equivalent diameter, perimeter and roundness.

[0126] In one embodiment, the performance prediction module 402 is further configured to:

[0127] The list of defect features is input into the physical sub-model to obtain the individual permeability flux of each micro-pinhole defect. The individual permeability flux of all micro-pinhole defects is summed to obtain the preliminary macroscopic moisture permeability prediction value.

[0128] Statistical features are calculated based on the defect feature list. A comprehensive feature vector is obtained by combining the statistical features with the preliminary macroscopic moisture permeability prediction value and the real-time collected melt temperature and extrusion pressure process parameters. The statistical features include the total number of defects, the average size, and the size distribution variance.

[0129] The comprehensive feature vector is input into the data-driven correction sub-model to obtain the macroscopic moisture permeability prediction value; the data-driven correction sub-model corrects the preliminary macroscopic moisture permeability prediction value.

[0130] In one embodiment, a model training module is also included, for:

[0131] Historical production data is collected to form a training dataset; each training data in the training dataset includes a comprehensive feature vector generated from the surface image of the coated paper and process parameters, and the corresponding true moisture permeability value of the coated paper;

[0132] A data-driven correction sub-model is constructed based on a multi-layer feedforward neural network, and the multi-layer feedforward neural network is supervised and trained using a training dataset to obtain model parameters. The training corresponds to taking the comprehensive feature vector as input and the actual moisture permeability value as the target output, and minimizing the mean square error between the macroscopic moisture permeability performance prediction value and the actual moisture permeability value through the backpropagation algorithm until the model converges.

[0133] The data-driven correction sub-model is determined based on the model parameters.

[0134] In one embodiment, the control module 403 is further configured to:

[0135] When the first control threshold is less than or equal to the predicted macroscopic moisture permeability value and less than the second warning threshold, the precise control mode is triggered. The precise control mode uses the deviation between the predicted macroscopic moisture permeability value and the first control threshold as input, and calculates the fine adjustment amount of the melt temperature setpoint through the proportional-integral-derivative control algorithm. The first control threshold is lower than the second warning threshold. The second warning threshold is lower than the product specification limit.

[0136] When the predicted value of macroscopic moisture permeability is greater than or equal to the second warning threshold, the enhanced control mode is triggered. The enhanced control mode corresponds to adding a compensation amount with a preset fixed amplitude on the basis of the fine adjustment amount, and at the same time generating an adjustment command for the matching relationship between the extrusion rate and the traction speed.

[0137] The adjustment command is converted into an industrial communication protocol message and sent to the temperature control module and transmission controller on the production line to drive the actuator to complete the adjustment of process parameters.

[0138] In one embodiment, a feedforward compensation control module is also included, for:

[0139] Analyze the size distribution of micro-pinhole defects in the defect feature list. If the distribution of micro-pinhole defects within a preset size range exceeds a preset density threshold, generate a melt temperature feedforward compensation increment.

[0140] The melt temperature feedforward compensation increment and the fine-tuning amount are superimposed to obtain a composite adjustment command; the composite adjustment command is used to instruct the actuator to adjust the process parameters according to the superposition of the melt temperature feedforward compensation increment and the fine-tuning amount.

[0141] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps in the above method embodiments.

[0142] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0143] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0144] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. An online quality monitoring and adaptive control method for a tissue paper production line, characterized in that, The method comprises: acquiring a surface image of a coated kraft paper and performing micro-pinhole defect recognition processing on the surface image to obtain a defect feature list; the defect feature list contains geometric feature parameters of each micro-pinhole defect; inputting the defect feature list into a pre-trained hybrid prediction model to obtain a macroscopic moisture permeability prediction value; the hybrid prediction model comprises a physical sub-model based on the physical law of gas diffusion and a data-driven correction sub-model based on machine learning; comparing the macroscopic moisture permeability prediction value with a preset target performance threshold value, and generating an upstream production process parameter adjustment amount according to the comparison result.

2. The method of claim 1, wherein, The micro-pinhole defect recognition processing on the surface image to obtain a defect feature list comprises: performing flat field correction and adaptive contrast enhancement processing on the surface image in sequence to obtain a pre-processed enhanced image; the surface image is a high-resolution original gray image of the coating surface acquired under dark field illumination conditions; applying an adaptive threshold segmentation algorithm based on local gray scale statistics to the enhanced image to obtain a binary image; performing morphological closing operation to connect adjacent pixel points on the binary image, and using a connected domain labeling algorithm to identify all independent connected regions to obtain candidate defects; screening the candidate defects according to a preset physical size range to obtain an effective defect region set; calculating geometric features for each defect region in the defect region set to obtain the defect feature list; the geometric features include equivalent diameter, perimeter and circularity.

3. The method of claim 2, wherein, The inputting of the defect feature list into the pre-trained hybrid prediction model to obtain a macroscopic moisture permeability prediction value comprises: inputting the defect feature list into the physical sub-model to obtain individual permeation fluxes of each micro-pinhole defect, and summing up the individual permeation fluxes of all micro-pinhole defects to obtain a preliminary macroscopic moisture permeability prediction value; calculating statistical features based on the defect feature list, and obtaining a comprehensive feature vector according to the statistical features and the preliminary macroscopic moisture permeability prediction value combined with real-time collected melt temperature and extrusion pressure process parameter combinations; the statistical features include total number of defects, average size and size distribution variance; inputting the comprehensive feature vector into the data-driven correction sub-model to obtain the macroscopic moisture permeability prediction value; the data-driven correction sub-model corrects the preliminary macroscopic moisture permeability prediction value.

4. The method of claim 3, wherein, The inputting of the defect feature list into the physical sub-model to obtain individual permeation fluxes of each micro-pinhole defect comprises: the individual permeation fluxes are obtained by the following formula: wherein, D is the effective diffusion coefficient of water vapor inside the microscopic pinhole defect; A is the cross-sectional area of the microscopic pinhole defect; is the water vapor concentration difference across the coating set according to standard test conditions; is the depth of the microscopic pinhole defect; is the equivalent diameter of the microscopic pinhole defect; is a dimensionless constant characterizing the entrance effect of diffusion.

5. The method of claim 3, wherein, The data-driven correction sub-model is trained by the following method: collecting historical production data to form a training data set; each training data of the training data set includes a comprehensive feature vector generated by a surface image of a coated kraft paper and process parameters, and a real moisture permeability value of the coated kraft paper corresponding to the comprehensive feature vector; constructing the data-driven correction sub-model based on a multi-layer feedforward neural network, and performing supervised training on the multi-layer feedforward neural network using the training data set to obtain model parameters; the training corresponds to taking the comprehensive feature vector as input and the real moisture permeability value as target output, and minimizing the mean square error between the macroscopic moisture permeability performance prediction value and the real moisture permeability value through a back propagation algorithm until the model converges; determining the data-driven correction sub-model according to the model parameters.

6. The method of claim 1, wherein, comparing the macroscopic moisture permeability performance prediction value with a preset target performance threshold value, and generating an upstream production process parameter adjustment amount according to a comparison result, including: when a first control threshold value ≤ the macroscopic moisture permeability performance prediction value < a second early warning threshold value, triggering an accurate control mode; the accurate control mode corresponds to taking a deviation of the macroscopic moisture permeability performance prediction value from the first control threshold value as input, and calculating a fine adjustment amount of a melt temperature set value through a proportional-integral-derivative control algorithm; the first control threshold value is lower than the second early warning threshold value; the second early warning threshold value is lower than a product specification limit value; when the macroscopic moisture permeability performance prediction value ≥ the second early warning threshold value, triggering an intensive control mode; the intensive control mode corresponds to increasing a preset fixed amplitude compensation amount on the basis of the fine adjustment amount, and simultaneously generating an adjustment instruction for a matching relationship between an extrusion rate and a traction speed; converting the adjustment instruction into an industrial communication protocol message and sending it to a temperature control module and a transmission controller driving an execution mechanism of a production line to complete process parameter adjustment.

7. The method of claim 6, wherein, The method further includes: analyzing a size distribution of the micro pinhole defects in the defect feature list, and if the micro pinhole defect distribution in a preset size range exceeds a preset number density threshold value, generating a melt temperature feedforward compensation increment; superimposing the melt temperature feedforward compensation increment and the fine adjustment amount to obtain a composite adjustment instruction; the composite adjustment instruction is used to instruct the execution mechanism to adjust the process parameters by the superimposed amount of the melt temperature feedforward compensation increment and the fine adjustment amount.

8. An on-line quality monitoring and adaptive control device for a coated paper production line, characterized in that The device includes: a coated paper defect identification module configured to acquire a surface image of a coated paper coating layer, and identify micro pinhole defects in the surface image to obtain a defect feature list; the defect feature list includes geometric feature parameters of each micro pinhole defect; a performance prediction module configured to input the defect feature list into a pre-trained hybrid prediction model to obtain a macroscopic moisture permeability performance prediction value; the hybrid prediction model includes a physical sub-model based on a gas diffusion physical law and a data-driven correction sub-model based on machine learning; a control module configured to compare the macroscopic moisture permeability performance prediction value with a preset target performance threshold value, and generate an upstream production process parameter adjustment amount according to a comparison result. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1 to 7.

Images