A method for real-time monitoring of the operating status of corrugated cardboard printing equipment

By using a long short-term memory network to analyze the surface temperature and color density of the anilox roller in a corrugated cardboard printing equipment, the impact of anilox roller temperature changes on printing quality was resolved, enabling real-time monitoring and early warning of printing quality and improving the stability of printing hue.

CN123104227APending Publication Date: 2026-11-17上海鼎龙机械有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611359551.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-03
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies cannot accurately identify the cumulative effect of anilox roller surface temperature changes on ink transfer behavior during corrugated cardboard printing, leading to a deviation between printing quality judgment standards and actual ink transfer conditions, and affecting the color difference consistency between consecutive printing batches.

Method used

By acquiring the time-series records of the surface temperature of the anilox roller and the color density of the first corrugated cardboard print, the heat accumulation characteristics are extracted using a long short-term memory network, the cell wall film thickness is analyzed, and the ink transfer amount is predicted by combining the color density. This distinguishes between batches with obstructed and smooth ink transfer, and the color density benchmark is lowered for batches with obstructed ink transfer. Real-time comparison triggers a color drift warning.

Benefits of technology

It achieves closed-loop perception and adaptive monitoring of the ink transfer efficiency of the anilox roller, significantly improving the stability of printing hue and the accuracy of equipment operation status discrimination, and reducing the impact of hue drift.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104227A_ABST
    Figure CN123104227A_ABST
Patent Text Reader

Abstract

This application provides a method for real-time monitoring of the operating status of corrugated cardboard printing equipment, including: acquiring the time sequence record of the surface temperature of the anilox roller in a continuous corrugated cardboard printing batch and the color density of the first corrugated cardboard print in the same batch; analyzing the duration of the temperature rise in the heat accumulation characteristics to determine the cell wall film thickness; combining the cell wall film thickness and the color density of the first corrugated cardboard print in the same batch to comprehensively predict the ink transfer amount, so as to distinguish between printing batches with obstructed cell ink transfer and printing batches with smooth cell ink transfer; acquiring the color density reading of the current corrugated cardboard printing batch, comparing it with the adjusted color density reading batch by batch, screening out printing batches with color density readings lower than the adjusted color density readings, and outputting a hue drift warning signal for the batch.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for real-time monitoring of the operating status of corrugated cardboard printing equipment. Background Technology

[0002] Corrugated cardboard printing, as a core link in the packaging industry, directly impacts product appearance consistency and production efficiency through the stability of its printing quality. The anilox roller, a key component for ink transfer, experiences significant changes in its surface condition, particularly during prolonged continuous printing. Existing monitoring solutions primarily focus on measuring printing parameters at single moments, failing to adequately consider the cumulative effect of anilox roller surface temperature changes on ink transfer behavior. As printing batches progress, the anilox roller surface temperature continuously changes, but traditional methods still rely on fixed color density criteria to determine printing quality, leading to a deviation between the quality assessment standard and the actual ink transfer state. When the anilox roller surface temperature rises moderately, the ink viscosity within the cells decreases, and its fluidity improves, resulting in increased ink transfer per unit area and full ink coverage. However, with further temperature increases, accelerated solvent evaporation forms a dry film on the cell walls. This film gradually blocks the cell openings, reducing the actual amount of ink that can be transferred. This causes the same color density reading to represent drastically different cell ink transfer states in different temperature ranges. For example, in the first printing run, the anilox roller temperature was recorded at around 30 degrees Celsius, and the measured color density was 1.5, indicating smooth ink transfer through the cells. However, in the fifth batch, although the same color density value was measured, the anilox roller temperature had risen to 50 degrees Celsius, resulting in severe film formation on the cell walls and a significant decrease in the actual effective ink transfer, causing color shift in the printed product. Because there is a lack of quantitative assessment of how the degree of cell film formation changes with the anilox roller surface temperature, it is impossible to accurately identify batches with reduced ink transfer, preventing timely warnings of color shift and consequently affecting the color consistency between consecutive printing batches. Summary of the Invention

[0003] This invention provides a method for real-time monitoring of the operating status of corrugated cardboard printing equipment, mainly including: Obtain the time sequence record of the surface temperature of the anilox roller in a continuous corrugated board printing batch and the color density of the first corrugated board printed in the same batch; The time-series recording of the surface temperature of the anilox roller is input into a long short-term memory network. Combined with the initial ink transfer conditions characterized by the color density of the first corrugated cardboard print as a priori constraint, the heat accumulation characteristics that match the thermal response differences of the current batch of water-based inks on the porous printing surface of the corrugated cardboard are extracted. Analyze the duration of the temperature rise in the heat accumulation characteristics to determine the thickness of the conjunctival membrane in the cavity wall; By combining the thickness of the cell wall film with the color density of the first corrugated cardboard print in the same batch, the ink transfer amount can be predicted, and printing batches with obstructed cell ink transfer and printing batches with smooth cell ink transfer can be distinguished. For printing batches where ink transfer to cells is obstructed, the current cell opening size of the anilox roller is compared with the cell opening size specified at the factory. Based on the comparison results, the color density is reduced to obtain the adjusted color density reading. For printing batches where ink transfer to cells is smooth, the measured color density of the first sheet is assigned as the reference color density reading. Obtain the color density reading of the current printing batch and compare it with the adjusted color density reading batch by batch. Filter out printing batches with color density readings lower than the adjusted color density readings and output a hue drift warning signal.

[0004] Furthermore, obtaining the time sequence record of the surface temperature of the anilox roller in a continuous corrugated board printing batch and the color density of the first corrugated board printed in the same batch includes: The surface temperature of the anilox roller is collected, and the temperature values ​​at each sampling time are spliced ​​together in chronological order to form a temperature time sequence record and labeled with the printing batch number. A spectrophotometer is used to measure the color density of the first printed product of the batch, and the color density value is read as the color density of the first corrugated cardboard printed product, and stored in association with the temperature time sequence record according to the batch number.

[0005] Furthermore, the step of inputting the time-series recording of the surface temperature of the anilox roller into a long short-term memory network, combined with the initial ink transfer conditions characterized by the color density of the first corrugated cardboard print as a priori constraints, extracts the heat accumulation characteristics matching the thermal response differences of the current batch of water-based inks on the porous printing surface of the corrugated cardboard, including: The time-series record of the surface temperature of the anilox roller is divided according to a preset time step. The mean and slope of each time step are calculated to obtain the temperature state sequence. For the color density of the first corrugated cardboard print, the initial ink transfer condition vector is determined by combining the linear correspondence between the color block reflection density and the ink transfer amount. The temperature state sequence and the initial ink transfer condition vector are aligned with the printing batch number and sent to the memory unit frame by frame. The forget gate discards historical temperature components that are irrelevant to the current thermal state, and the input gate writes the roller thermal state components. The initial ink transfer condition vector is used to suppress candidate states that deviate from the initial transfer level, and the hidden state sequence is output. After the hidden state sequence is dimensionally reduced by a fully connected mapping layer, it is split into the ink-side heat-induced film formation tendency component and the printing surface-side heat-induced ink absorption fluctuation component according to the water-based ink solvent evaporation temperature threshold range. The heat accumulation feature is obtained by combining them.

[0006] Furthermore, the duration of the temperature rise in the heat accumulation characteristics is analyzed to determine the thickness of the cell wall conjunctiva, including: For the heat accumulation characteristics, a solvent evaporation trigger threshold is set along the time axis for the ink-side heat-induced film formation tendency component. Continuous time steps where the amplitude of the time axis component of the ink-side heat-induced film formation tendency component exceeds the solvent evaporation trigger threshold are merged into a heating duration period. The time step span and the average thermal intensity of each heating duration period are statistically analyzed to obtain the heating duration characteristics.

[0007] Furthermore, after statistically analyzing the time step span and the average thermal intensity of each heating period to obtain the characteristics of the heating period, the process includes: The time span in the heating period is divided into grades according to a preset segment interval and a corresponding reference conjunctival rate value is given. The reference conjunctival rate value is then corrected according to the average thermal intensity to obtain the graded conjunctival rate. The graded conjunctival rate is multiplied by the heating period feature segment by segment and then accumulated in chronological order. The membrane retreat amount calculated during the low temperature intermittent period is then deducted to obtain the cell wall conjunctival thickness.

[0008] Furthermore, the method of combining the cell wall film thickness with the color density of the first corrugated cardboard print in the same batch to predict ink transfer volume, and distinguishing between printing batches with obstructed cell ink transfer and printing batches with smooth cell ink transfer, includes: The cell wall film thickness is normalized according to the original opening size calibrated by the anilox roller at the factory to obtain the cell opening blockage rate. Based on the color density of the first corrugated board print in the same batch, and combined with the linear correlation between color block reflection density and ink transfer amount, the baseline transfer amount per unit area is calculated. The baseline transfer amount per unit area is adjusted downwards using the blockage attenuation coefficient corresponding to the cell opening blockage rate to obtain the predicted ink transfer amount. The difference between the predicted ink transfer amount and the rated ink transfer amount calibrated by the anilox roller at the factory is used to determine the batch ink transfer amount deviation. Printing batches with a batch ink transfer amount deviation lower than the ink transfer judgment threshold are judged as printing batches with smooth cell ink transfer, while printing batches with a deviation not lower than the ink transfer judgment threshold are judged as printing batches with obstructed cell ink transfer.

[0009] Furthermore, for printing batches where ink transfer to the cells is obstructed, the current cell opening size of the anilox roller is compared with the factory-calibrated cell opening size. Based on the comparison result, the color density is reduced to obtain an adjusted color density reading, including: Non-contact measurement of multiple cell openings along the roller axis is performed, and the arithmetic mean of the multi-point measurement values ​​is taken as the current cell opening size of the anilox roller. The difference between the current cell opening size and the factory-calibrated cell opening size is calculated and divided by the factory-calibrated cell opening size to obtain the cell opening shrinkage rate. When the cell opening shrinkage rate exceeds the shrinkage response threshold, the color density is reduced by a preset color density reduction coefficient to obtain the adjusted color density reading.

[0010] Furthermore, the step of acquiring the color density reading of the current printing batch and comparing it batch by batch with the adjusted color density reading, filtering out printing batches with color density readings lower than the adjusted color density reading, and outputting a hue drift warning signal includes: Simultaneous measurement of solid color blocks of the current printing batch is performed to obtain color density readings; the color density readings are compared with the adjusted color density readings batch by batch number, and batches with color density readings lower than the adjusted color density readings are selected as hue drift candidate batches; for the hue drift candidate batches, a hue drift warning signal carrying the difference between the printing batch number and the color density is output.

[0011] Furthermore, after outputting a hue drift warning signal carrying the difference between the printing batch number and the color density for the hue drift candidate batch, the process includes: The hue drift warning signal is transmitted to the printing equipment control unit along the fieldbus. The printing equipment control unit reads the color density difference. When the color density difference exceeds the cleaning trigger threshold, the anilox roller cleaning program is activated. When the color density difference does not exceed the cleaning trigger threshold, the ink supply adaptive recalibration operation is activated.

[0012] Furthermore, the method also includes: Using the printing batch number as the primary key, the hue drift warning signal, cell wall film thickness, and heat accumulation characteristics are aligned, linked, and timestamps are added. These are then arranged in ascending order of timestamps to obtain time-series labels. The combined values ​​of cell wall film thickness and heat accumulation characteristics within the time-series labels are used as incremental samples to backfill the prior constraints. The long short-term memory network updates the prior constraints in incremental training mode. Based on the updated prior constraints, the heat accumulation characteristics of subsequent printing batches are superimposed with the hue drift warning batch distribution to obtain the operating status monitoring results, which include the current anilox roller ink transfer efficiency.

[0013] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a real-time monitoring method for the operating status of corrugated cardboard printing equipment. Addressing the problem of ink transfer obstruction and subsequent hue drift caused by solvent evaporation and film formation in the cells of the anilox roller on the porous printing surface of corrugated cardboard, which is difficult to detect in a timely manner, the method inputs the time-series recording of the anilox roller surface temperature into a long short-term memory network (LSTM). Using the initial ink transfer condition characterized by the color density of the first printed sheet of the same batch as a priori constraint, the method extracts heat accumulation features matching the differences in thermal response. Based on the duration of the temperature rise, the cell wall film thickness is inverted, and the ink transfer amount is comprehensively predicted by combining the first sheet's color density to distinguish between batches with obstructed and smooth ink transfer. For batches with obstructed ink transfer, the color density benchmark is lowered based on the cell opening size deviation. Then, the color density is compared batch by batch with the real-time color density to trigger a hue drift warning. The warning, film thickness, and heat accumulation features are encapsulated as time-series tags to feed back to the network and update the priori constraints, achieving closed-loop perception and adaptive monitoring of the anilox roller's ink transfer efficiency. This significantly improves the stability of printed hue and the accuracy of equipment operating status discrimination. Attached Figure Description

[0014] Figure 1 This is a flowchart of a method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to the present invention.

[0015] Figure 2 This is a schematic diagram of a method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to the present invention.

[0016] Figure 3 This is another schematic diagram of a method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to the present invention. Detailed Implementation

[0017] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0018] like Figures 1-3 This embodiment of a method for real-time monitoring of the operating status of corrugated cardboard printing equipment may specifically include: Step S101: Obtain the time sequence record of the surface temperature of the anilox roller in the continuous corrugated board printing batch and the color density of the first corrugated board printed in the same batch.

[0019] During the continuous operation of corrugated cardboard printing batches, the surface temperature of the anilox roller is collected by an infrared temperature sensor at a preset sampling period. The temperature values ​​at each sampling moment are sequentially spliced ​​to form a temperature time sequence record. Simultaneously, the corresponding printing batch number is labeled on the temperature time sequence record, resulting in roller surface temperature time sequence data aligned with the batch. For each printing batch associated with the roller surface temperature time sequence data, on the first corrugated cardboard printed in that batch, a spectrophotometer is used to measure the solid color patch on the surface of the printed product. The color density value of this solid color patch is read as the color density of the first printed product in that batch and is stored in association with the roller surface temperature time sequence data according to the batch number.

[0020] In one embodiment, for the ink transfer monitoring scenario of anilox roller on a continuous corrugated board printing production line, an infrared temperature sensor is installed above the side of the anilox roller, with the probe aligned with the roller surface and covering the actual contact area of ​​the printing plate along the axial direction, and the sampling period is set to read the roller surface temperature once every 2 to 5 seconds.

[0021] For example, a printing batch typically lasts 5 to 15 minutes, and all temperature sampling points collected within the batch are arranged in chronological order according to their timestamps, splicing them together to form a continuous temperature time-series record.

[0022] Understandably, since the transfer of water-based inks on the porous printing surface of corrugated cardboard is accompanied by solvent evaporation and heat generation from roller friction, this temperature time series record reflects the thermal changes of the surface of the current batch of anilox rollers during the ink transfer process.

[0023] Specifically, at the start of each printing batch, the printing equipment control unit synchronously assigns a unique printing batch number. This number is bound to the starting sampling time of the temperature timing record. After the batch ends, the temperature timing record, along with the number, is stored in the roller surface temperature timing database to obtain roller surface temperature timing data aligned with the batch. Preferably, the color density of the first printed product is collected at the first corrugated cardboard printed after the batch is started. The sampling point is selected from the solid color block on the surface of the printed product corresponding to the solid area of ​​the printing plate. This position has uniform ink coverage and is not affected by dot gain. The illumination source of the spectrophotometer illuminates the solid color block at a 45-degree angle. The reflected light is decomposed into components corresponding to the three primary color channels by the built-in filter channel, and then converted into an electrical signal by the photoelectric element and converted into a color density value according to the standard density formula. This value is read as the color density of the first printed product of the batch.

[0024] It should be noted that the first printed product is in the early stage of the batch, and the ink absorption state of the printing surface and the temperature of the roller have not yet been affected by long-term accumulation. This value can characterize the initial ink transfer condition of the batch and reflect the baseline level of the current anilox roller transferring ink to the porous printing surface of the corrugated cardboard at the beginning stage of the batch.

[0025] In one possible implementation, the roller surface temperature time series data and the color density of the first printed product are established in a batch data table with the printing batch number as the primary key. Under the same number, a temperature time series and a color density value are simultaneously linked to form a searchable associated record by batch number, which can be called in the subsequent heat accumulation feature extraction stage.

[0026] Step S102: Input the time-series record of the surface temperature of the anilox roller into the long short-term memory network, and combine it with the initial ink transfer conditions of the corrugated board printing characterized by the color density of the first corrugated board print as a priori constraint to extract the heat accumulation characteristics that match the thermal response differences of the current batch of water-based inks on the porous printing surface of the corrugated board.

[0027] The time-series record of the surface temperature of the anilox roller is uniformly divided according to a preset time step. For the roller surface temperature sampling points within each time step, the mean and slope within that window are calculated to obtain a temperature state sequence reflecting the heating rate of the roller body. For the color density of the first printed product, based on the linear correspondence between the solid color block reflection density and the ink transfer amount, it is converted into a quantized initial ink transfer condition vector. The initial ink transfer condition vector records the unit area transfer amount and spreading consistency of water-based ink on the porous printing surface of corrugated cardboard at the beginning of the batch. The temperature state sequence and the initial ink transfer condition vector are aligned with the printing batch number as the link to obtain the input tensor of the long short-term memory network. The input tensor is fed frame by frame into the memory unit of the Long Short-Term Memory network in time step order. The forget gate reads the hidden state quantity of the previous moment and discards historical temperature components that are irrelevant to the current thermal state according to the continuity of the roller temperature rise trend. The input gate writes the new roller thermal state component according to the temperature mean and slope of the current time step. The initial ink transfer condition vector is connected to the candidate state calculation path of the input gate as an additional constraint term to suppress candidate states that deviate from the initial transfer level of the batch. After the memory unit is updated by gating, it outputs the hidden state quantity corresponding to the current time step. The hidden state quantities of consecutive time steps are concatenated along the time axis to obtain the hidden state sequence that carries the thermal evolution trajectory of the roller. For the hidden state sequence, the dimensionality of the weighted aggregation result of the end time step and the whole sequence is reduced by a fully connected mapping layer, and a multi-component vector is output that reflects the total amount of heat deposition, the steepness of temperature rise, and the proportion of high temperature maintenance period of the roller in the whole batch. According to the solvent evaporation temperature threshold range of water-based ink, the multi-component vector is split according to the thermal response channel to distinguish the ink side heat film formation tendency component and the printing surface side heat ink absorption fluctuation component. The two types of components are combined along the time axis according to the batch number to obtain the heat accumulation characteristics that match the thermal response difference of the current batch of water-based ink on the porous printing surface of corrugated cardboard.

[0028] In one embodiment, to address the ink transfer monitoring requirements of water-based inks on a continuous corrugated board printing production line, the time-series data of the roller surface temperature obtained in the preceding process and the color density of the first printed product are used as raw materials in the heat accumulation feature extraction stage. The core of this stage is to feed the time evolution process of the roller body temperature rise and the ink transfer conditions at the beginning of the batch into a long short-term memory network to extract a multi-component characterization that reflects the thermal response differences of the current batch.

[0029] Specifically, the temperature time series record is uniformly divided according to a preset time step, which is usually selected as a value between 10 seconds and 30 seconds.

[0030] For example, a time step of 20 seconds is used. Each time step contains 4 to 10 temperature sampling points. The arithmetic mean of the sampled values ​​within this window is calculated to reflect the average thermal state of the roller surface within the window. A linear fit is performed on the sampled values ​​within the same window over time, and the slope of the fitted line reflects the heating rate of the roller within that window. The mean and slope components are concatenated in order of time step to obtain a temperature state sequence reflecting the heating rate of the roller.

[0031] It should be noted that the mean component and the slope component are complementary in physical meaning; the former describes the absolute level of the thermal state, while the latter describes the rate of change of the thermal state.

[0032] For example, the average temperature of the rollers in the early stages of a batch may be 35 degrees Celsius with a steep slope; as the batch progresses, the average temperature rises to 45 degrees Celsius with a gentler slope. The temperature sequence, measured in time steps, represents the thermal trajectory of the rollers throughout the entire batch. Preferably, the conversion from the color density of the first printed sheet to the initial ink transfer vector is performed according to the linear correspondence between the solid color patch reflectance density and the amount of ink transferred.

[0033] Understandably, for a given model of water-based ink and the corrugated cardboard substrate, several sets of color density readings and measured ink transfer per unit area are collected offline using a standard step color chart. A linear reference curve is then fitted using the least squares method, and this curve is already fixed as a lookup table relationship in the offline stage.

[0034] For example, in step S101, spectrophotometer measurement points are arranged at equal intervals in a 3x3 grid within the solid color patch, and the arithmetic mean of the 9 readings is taken as the representative color density value of the sheet. The same set of readings is directly used to calculate the sample variance, avoiding measurement drift caused by additional point arrangement in subsequent steps. When the average color density reading is 1.5, the ink transfer amount per unit area is found to be approximately 1.45 grams per square meter according to the comparison relationship. The spreading consistency component is converted in a piecewise manner using the variance s of the 9 points as input: when s falls into the range of 0 to 0.005, it is taken as 1; when s falls into the range of 0.005 to 0.02, it is linearly and smoothly decreased from 1 to 0.5, that is, 1 is taken at the endpoint of 0.005 and 0.5 is taken at the endpoint of 0.02. For every 0.001 increase in variance within the range, the consistency decreases by approximately 0.0333; when s exceeds 0.02, it is taken as 0.3. The initial ink transfer condition vector consists of two components: the amount of transfer per unit area and the consistency of spreading. It records the initial ink transfer baseline of water-based ink on the porous printing surface of corrugated cardboard at the beginning of the batch.

[0035] Furthermore, the temperature state sequence and the initial ink transfer condition vector are aligned with the printing batch number as the link. The temperature state sequence serves as the main time axis, and the initial ink transfer condition vector is attached at the starting position of the time axis in the form of a constant value, thus obtaining the input tensor of the long short-term memory network.

[0036] In one possible implementation, the input tensor is fed frame by frame into the memory units of the Long Short-Term Memory (LSTM) network in time-step order. The LSM network adopts a two-layer stacked structure, with the hidden state dimension set between 32 and 64. The processing at each time step follows the classic gating mechanism of the LSM network: the forget gate reads the hidden state from the previous time step and the average temperature and slope of the current time step, outputting a forgetting coefficient between 0 and 1 to determine the retention ratio of historical thermal states in the current time step; the input gate generates candidate states based on the average temperature and slope of the current time step, writing new roller thermal components. Based on this, the initial ink transfer condition vector v0 is first projected onto a constraint embedding e of the same dimension as the candidate state c of the input gate through a linear mapping layer at each time step, and then concatenated with the candidate state c along the feature dimension before being fed into the writing path of the input gate to participate in the re-estimation of the candidate states. The splicing weight w is jointly determined by the baseline weight w0 and the deviation decay factor α. Initially, w0 is set to 0.8, and α is initially set to 0.2. At each time step, the candidate state c is linearly read out to obtain the implicit transfer level ht. The difference between ht and the batch initial transfer level h0 given by the initial ink transfer condition vector is used to obtain the deviation amplitude d = |ht - h0|. This deviation is then normalized according to the allowable fluctuation range of 0.05 for the batch initial transfer level, resulting in the relative deviation r = d / 0.05. When r is not greater than 1, the splicing weight remains w = w0, and the input gate is written normally; when r is greater than 1, the splicing weight is adjusted according to... The threshold is monotonically lowered, with a lower limit truncation of 0.2, thereby compressing the write amplitude of the input gate to candidate state c in the deviation direction and suppressing the accumulation of hot states in that direction. At the same time, the embedding e is constrained and superimposed back onto the write path with a fixed weight of 0.3, guiding the candidate state to regress towards the batch start transition level. If r is still greater than 1 for three consecutive time steps, α is increased to 0.4 to accelerate the suppression response, and then restored to 0.2 after r falls back to below 1.

[0037] Specifically, the additional constraint term participates in the backpropagation weight update along with the temperature input component during the network training phase. This allows the memory unit to both gradually accumulate the continuous heating effect of the roller along the time axis and constrain thermal fluctuations that deviate from the initial ink transfer baseline of the batch, preventing the network from misjudging short-term noise-type temperature fluctuations as heat accumulation. After gating and updating, the memory unit outputs the hidden state quantity corresponding to the current time step. The hidden state quantities of consecutive time steps are concatenated along the time axis to obtain a hidden state sequence that carries the thermal evolution trajectory of the roller. For the hidden state sequence, a fully connected mapping layer is used to concatenate and reduce the dimensionality of the hidden state quantity at the end time step and the result of the entire sequence after being weighted and aggregated according to attention weights. This outputs a multi-component vector reflecting the total heat deposition, heating steepness, and proportion of high-temperature maintenance period of the roller throughout the entire batch. The training samples of the network are derived from the roller temperature measurement curves and process logs of the same batch retained from historical printing batches. The sampling step size is 2 seconds, and a single sample covers approximately 800 to 1200 time steps of a complete batch. The training set and validation set are divided in an 8:2 ratio. Each sample's supervisory label consists of three components: y1, the total heat deposition calibrated by the time integral of the roller temperature relative to the initial ink transfer baseline within the batch; y2, the temperature rise steepness calibrated by the maximum slope of the temperature rise segment; and y3, the proportion of the high-temperature maintenance period calibrated by the percentage of the total batch duration where the roller temperature exceeds the batch baseline threshold by 5 degrees Celsius. All three are obtained offline from historical batch measured curves and normalized to zero mean and unit variance. During training, the weighted mean square error between the network's output multi-component vector and the label vector is used as the main loss Lmain=w1*(y1). hat -y1) 2 +w2*(y2 hat -y2) 2 +w3*(y3 hat -y3) 2 , where y1 hat y2 hat y3 hatFor the corresponding output components of the network, the weight coefficients w1, w2, and w3 are preset to 0.4, 0.3, and 0.3 respectively, based on the numerical magnitudes of the three components. Simultaneously, the additional constraint term penalizing hot-state fluctuations deviating from the baseline is incorporated into the total loss L = Lmain + λ * Lreg in the form of a regularization term Lreg. The regularization coefficient λ is set to 0.05 to 0.1 to simultaneously converge the forget gate and input gate parameters during gradient backpropagation. The optimizer uses Adam, with an initial learning rate of 1e-3, a batch size of 32, 100 training epochs, and early termination when the validation set loss does not decrease for 5 consecutive epochs, thus obtaining the weight parameters used for deployment.

[0038] For example, the total heat deposition component corresponds to the energy amplitude of the entire hidden state sequence, the temperature rise steepness component corresponds to the peak statistics of the slope component, and the high temperature maintenance period proportion component corresponds to the time step proportion of the roller surface temperature exceeding 45 degrees Celsius. Further, the multi-component vector is split according to the thermal response channel based on the water-based ink solvent evaporation temperature threshold range.

[0039] For example, the lower limit of the solvent evaporation temperature threshold range is set at 40 degrees Celsius, and the upper limit is set at 55 degrees Celsius. Below the lower limit, solvent evaporation is insufficient, and the cell walls are not easy to form a film. Above the upper limit, solvent evaporation is intense, and the cell walls are easy to form a film quickly. The steepness of the temperature rise is characterized by the ratio of the temperature difference between adjacent sampling points to the sampling interval, in degrees Celsius per second. The preset amplitude is 1.2 degrees Celsius per second, reflecting the degree of temperature jump in a short period of time. The component classification is determined step-by-step in the following order to ensure mutual exclusion and completeness: First, consider the temperature drop. Components whose temperature does not fall within the 40-55 degree Celsius range are classified as heat-induced ink absorption fluctuation components on the printing surface side. This is because such temperature range cannot enter the stable film-forming window and can only disturb the ink absorption of the pores. Next, consider the steepness. Components that fall within the range but have a temperature rise steepness exceeding 1.2 degrees Celsius per second are also classified as heat-induced ink absorption fluctuation components on the printing surface side. This is because short-term impacts are insufficient to support continuous film formation and instead manifest as ink absorption disturbances. Finally, consider the proportion. Components that fall within the range and have a temperature rise steepness not exceeding 1.2 degrees Celsius per second, if the proportion of the high-temperature maintenance period is not less than 0.4, are classified as heat-induced film-forming tendency components on the ink side, reflecting the accumulation tendency of cell wall film formation. If the proportion of the high-temperature maintenance period is less than 0.4, they are classified as heat-induced ink absorption fluctuation components on the printing surface side, reflecting the fluctuation of the pore ink absorption state of the porous printing surface of corrugated cardboard. For example, if a component has a temperature of 48 degrees Celsius, a steepness of 0.6 degrees Celsius per second, and a high-temperature percentage of 0.55, it is included in the ink side; if it is also 48 degrees Celsius but has a steepness of 1.5 degrees Celsius per second, it is included in the printing surface side regardless of its percentage. The two types of components are combined along the time axis according to the batch number to obtain the heat accumulation characteristics that match the thermal response differences of the current batch of water-based ink on the porous printing surface of corrugated cardboard. This can characterize the differentiated influence of the roller temperature rise in the current batch on the two paths of ink film formation on the ink side and ink absorption on the printing surface side. In another embodiment of determining the multi-component vector, taking a printing process of 60 seconds for a batch as an example, the temperature sensor collects the roller surface temperature every 1 second, forming a temperature state sequence of 60 time steps. The initial ink transfer condition vector corresponding to this batch is a three-dimensional vector, representing the initial ink layer thickness, initial roller surface wettability, and reference ink transfer pressure, respectively. The measured values ​​can be set to [0.75, 0.60, 0.80], and each component has been normalized to the range of 0 to 1. This vector is bound to the batch number and is attached as a constant value to the input of each time step at the starting position of the time axis. It is concatenated with the average temperature and temperature slope of the current step to form a five-dimensional input tensor for each step. The network adopts a two-layer long short-term memory structure with a hidden state dimension of 48. The inputs of the 60 time steps of the entire batch are sent to the memory unit frame by frame in sequence. In actual operation, it is assumed that the average roller surface temperature of the 10th time step is 32.5 degrees Celsius, and the temperature slope compared to the previous moment is a temperature increase of 0.8 degrees Celsius per minute.The forget gate reads the hidden state from the previous moment and the average temperature and slope of the current step size, outputting a forgetting coefficient of 0.92. This means that 92% of the thermal information accumulated in the previous moment is retained, and only 8% is forgotten. This aligns with the thermophysical characteristics of the roller body, which has high thermal inertia and does not experience sudden temperature changes. The input gate generates candidate states based on the same input, representing the heat accumulation components to be added at the current moment. During this process, the initial ink transfer condition vector participates in the calculation as an additional constraint. For example, based on the initial transfer level of this batch (which can be determined by the initial ink layer thickness of 0.75 and the reference transfer rate of 0.70), the transfer level implied by the current candidate state, after network mapping, is 0.73, with a deviation of only 0.03, within the preset allowable deviation range of ±0.10. Therefore, the write amplitude of the input gate remains normal, allowing the heating effect to be written into the memory cell. Conversely, if a short-lived temperature spike occurs at time step 25, with the average temperature suddenly rising to 38.0 degrees Celsius and the slope increasing by 3.5 degrees Celsius per minute, the implicit transfer level corresponding to the candidate state triggered by this abnormal data is mapped to 0.88, deviating from the batch initial transfer level of 0.70 by 0.18, exceeding the preset range of ±0.10. At this point, the component related to the transfer level in the initial ink transfer condition vector will have a suppressive effect on the splicing weights, causing the activation value of the input gate to decrease from the normal 0.75 to 0.30. The actual writing amplitude of the candidate state is thus significantly reduced, with only a very small amount of abnormal thermal components being recorded in the cell state. This mechanism ensures that brief noise spikes are not misjudged by the network as substantial accumulation of heat in the roller, while genuine continuous heating, such as a stable temperature and slope of 0.3 degrees Celsius per minute from time steps 30 to 50, accumulates normally because the deviation remains within 0.05. During network training, the initial ink transfer condition vector and the temperature input component participate in backpropagation. When the actual heat deposition on the roller in a training sample is high, the gradient update strengthens the weight of the corresponding component in the ink transfer condition vector, making the constraint boundary more adaptive; conversely, if the sample label shows low heat accumulation, the corresponding weight is weakened. After multiple iterations, the network learns how to balance preserving the long-term heating trend and suppressing instantaneous disturbances. After all 60 time steps have been processed, the hidden state variables output by each step are concatenated in chronological order to form a complete hidden state sequence. In the output stage, on the one hand, the terminal hidden state of the 60th step is extracted, which reflects the thermal information of the roller at the end of the batch; on the other hand, attention weights are calculated for each of the 60 hidden states in the entire sequence.

[0040] For example, the sum of attention weights for the 30th to 50th step interval, where the heating rate is relatively fast and stable, reaches 0.65, while the sum of weights for the edge steps of the starting and ending segments is only 0.35. After weighted aggregation, a global thermal evolution summary vector is obtained. The terminal hidden state is concatenated with this summary vector and then fed into a fully connected mapping layer for dimensionality reduction, outputting a three-dimensional component vector. The first dimension of this vector has a value of 0.82, representing a normalized estimate of the total heat deposition of the rollers in the entire batch; the second dimension has a value of 2.4, calculated by dividing the maximum temperature rise rate by the average temperature rise rate, characterizing the steepness of the heating process, with a larger value indicating a more concentrated and intense heating; the third dimension has a value of 0.58, indicating that the roller surface temperature exceeds the threshold of 35 degrees Celsius for 58% of the time. Combining these three components can comprehensively quantify the overall thermal evolution of the rollers in this batch.

[0041] Step S103: Analyze the duration of the temperature rise in the heat accumulation characteristics to determine the thickness of the cell wall conjunctiva.

[0042] The time axis component reflecting the tendency of ink to form a film under heat in the heat accumulation characteristics is extracted and expanded according to the time step. A solvent evaporation trigger threshold is set for this component along the time axis, which is a reference to the lower limit of the amplitude of the component of the solvent evaporation point of water-based ink. Continuous time steps with component amplitude exceeding the trigger threshold are merged to obtain one or more heating duration periods within the batch. For each heating duration period, the time step span between its start time step and end time step is calculated, and the mean value of the ink-side tendency to form a film under heat within this span is read simultaneously to obtain the heating duration characteristics that carry the dual-dimensional information of the duration span and the thermal intensity. Based on the film-forming mechanism of water-based ink forming a dry film on the cell wall upon heating, the time span of the heating period is divided into three levels: short, medium, and long, according to a preset segmentation interval. A baseline film-forming rate value is given for each level, with a higher baseline film-forming rate value corresponding to a higher time span level. For the heating period characteristics within the same level, the baseline film-forming rate value is further multiplied and corrected based on its average thermal intensity. The larger the average thermal intensity, the larger the multiplication factor, resulting in a segmented film-forming rate that reflects the rate of solvent evaporation and deposition on the cell wall. The segmented film-forming rate is bound to the heating period characteristics segment by segment along the time axis to obtain a sequence of cell wall film growth rates. For the cell wall film growth rate sequence, multiply it segment by segment along the printing batch time axis with the time span of the corresponding heating duration to obtain the film thickness increment within the heating duration. Then, accumulate the film thickness increments of multiple heating durations within the same batch in chronological order. Based on the accumulated result, calculate the film retreat amount according to the amplitude of the heat-induced film formation tendency component of the ink side during the low-temperature intermittent period. Subtract the film retreat amount from the accumulated result to obtain the thickness of the dry film layer stably deposited on the cell wall at the end of the current printing batch, which is taken as the cell wall film thickness.

[0043] In one embodiment, to meet the quantitative requirements of the cell wall film thickness in a continuous corrugated cardboard printing production line, based on the physical mechanism of solvent evaporation and gradual deposition of water-based ink on the cell wall to form a dry film layer under heating conditions, the heat accumulation characteristics obtained in the preceding steps are used as raw materials to estimate the thickness of the dry film layer stably deposited on the cell wall at the end of the current printing batch.

[0044] It is understandable that when the surface temperature of the roller rises above the evaporation point of the water-based ink solvent, the water and alcohol solvent molecules in the ink inside the cells escape more quickly and form a thin film along the cell wall. The longer the temperature is maintained, the thicker the film stacks.

[0045] Specifically, the component of ink-side heat-induced film formation tendency in the heat accumulation characteristics is expanded along the time axis step by step. This component corresponds to an amplitude at each time step on the time axis, and the magnitude of the amplitude reflects the driving intensity of the roller surface temperature on the solvent evaporation film formation process within that time step. Furthermore, a solvent evaporation trigger threshold is set for this time axis component, and this threshold is selected with reference to the lower limit of the component amplitude of the solvent evaporation initiation point of water-based inks.

[0046] For example, taking the roller surface temperature of approximately 40 degrees Celsius, corresponding to the solvent evaporation point of the water-based ink, as a reference, the value is determined by the statistical value of the component amplitude mapped to this temperature from the preceding steps, with an exemplary value of 0.3. Time steps where the component amplitude is less than this threshold are considered periods of insufficient solvent evaporation drive and are not included in the calculation of the film formation process.

[0047] It should be noted that for time steps where the component amplitude exceeds the solvent evaporation trigger threshold, adjacent positions are merged along the time axis. A slight fluctuation is allowed between two adjacent time steps exceeding the threshold; otherwise, it is considered the termination of a heating duration. The merging operation is repeated along the entire batch time axis to obtain one or more heating durations within the batch.

[0048] For example, a printing batch may have 2 to 4 heating duration periods, corresponding to the initial startup phase, steady-state operation phase, and subsequent heat accumulation phase, respectively. For each heating duration period, the time step span between its start and end time steps is calculated, and the arithmetic mean of the ink-side heat-induced film formation tendency component within this span is simultaneously read as the average thermal intensity of that segment. The combination of the time span and the average thermal intensity constitutes the heating duration characteristic of that segment.

[0049] In one possible implementation, the time span of the heating period is divided into three categories according to a preset segmentation interval: time spans of less than 30 seconds are classified as short time periods, time spans between 30 and 90 seconds are classified as medium time periods, and time spans exceeding 90 seconds are classified as long time periods. Each category is given a baseline conjunctival rate value.

[0050] For example, the baseline conjunctival rate value for the short-term range is 0.02 micrometers per second, the baseline conjunctival rate value for the medium-term range is 0.05 micrometers per second, and the baseline conjunctival rate value for the long-term range is 0.09 micrometers per second, showing a tiered pattern where the higher the tier, the higher the baseline value. This reflects the solvent evaporation and film formation pattern where a thicker dry film layer is deposited on the cell wall per unit time as the heating duration continues longer. Furthermore, for the heating duration characteristics within the same tier, the baseline conjunctival rate value for that tier is multiplied and corrected based on its average thermal intensity. The multiplier correction follows the relationship r=r0*(1+a*(h-h0)), where r0 is the baseline conjunctival rate value for that tier, h is the average thermal intensity of the current segment, h0 is the intra-tier reference value of the average thermal intensity of that tier, which is obtained by averaging the average thermal intensity of multiple batches of heating duration characteristics accumulated during the offline calibration stage and then solidifying it, and a is the sensitivity coefficient, with a value ranging from 0.5 to 1.5. When the average thermal intensity is higher than the reference value within the range, the multiplication factor is greater than 1; otherwise, it is less than 1. r represents the corrected film formation rate for this segment, resulting in a segmented film formation rate reflecting the rate of solvent evaporation and deposition on the cell walls. This segmented film formation rate is then bound segment by segment along the time axis to the characteristics of the heating period, yielding a sequence of cell wall film growth rates. Preferably, for the cell wall film growth rate sequence, it is multiplied segment by segment along the printing batch time axis by the span of the corresponding heating duration to obtain the film thickness increment within that heating duration.

[0051] For example, if a heating period of 60 seconds has a baseline conjunctival velocity of 0.05 micrometers per second and a multiplier-corrected conjunctival velocity of 0.06 micrometers per second, then the conjunctival thickness increment for that period is 3.6 micrometers. The conjunctival thickness increments of multiple heating periods within the same batch are accumulated sequentially according to time to obtain the cumulative conjunctival thickness within the batch.

[0052] It should be noted that during the low-temperature intermittent period, the solvent film layer on the cell wall will undergo partial recovery and redissolution due to the drop in roller surface temperature, resulting in slight degradation of the film layer. The calculation of the film layer regression amount is based on the product of the span of the low-temperature intermittent period and the average difference between the amplitude of the component of the ink side's tendency to form a film under heat and the threshold value of solvent evaporation during that period, and multiplied by a conversion factor. The conversion factor is exemplarily taken as 0.02 micrometers per second per unit difference, where a unit difference refers to the case where the component amplitude is lower than the threshold by 1 amplitude unit. This formula calculates the amount of film retraction. T represents the amount of film retraction. low V represents the span of the low-temperature intermittent period, n represents the number of sampling points within that period, and V th H represents the solvent evaporation trigger threshold. ink , i represents the amplitude of the ink-side heat-induced film formation tendency component at the i-th sampling point, K conv This represents the conversion factor. The main body of the formula is the product of the time span and the average difference, multiplied by the conversion factor. Subtracting the film retreat amount from the cumulative film thickness yields the thickness of the stable dried film deposited on the cell wall at the end of the current printing batch, which is taken as the cell wall film thickness. The cell wall film thickness, in numerical form, characterizes the cumulative degree of blockage of the cell openings by the deposited film under the current batch's roller temperature rise conditions, reflecting the decrease in the actual amount of ink that can be transferred through the cells.

[0053] Step S104: Combine the cell wall film thickness with the color density of the first corrugated cardboard print in the same batch to predict the ink transfer amount, so as to distinguish between printing batches with blocked cell ink transfer and printing batches with smooth cell ink transfer.

[0054] Based on the relationship between film thickness and opening size in the cell geometry, the film thickness of the cell wall is normalized according to the original opening size calibrated at the factory of the anilox roller to obtain the cell opening clogging rate. For the color density of the first corrugated cardboard print in the same batch, it is converted into a baseline transfer amount per unit area according to a pre-defined linear correlation between solid color block reflection density and ink transfer amount. The cell opening clogging rate and the baseline transfer amount per unit area are aligned according to the printing batch number to obtain batch ink transfer status data carrying both film clogging and color density dimensions. Based on the inhibitory relationship between the cell opening clogging rate and the baseline transfer amount per unit area in the batch ink transfer status data, the baseline transfer amount per unit area is adjusted downwards using the clogging attenuation coefficient corresponding to the cell opening clogging rate as the downward adjustment ratio. The clogging attenuation coefficient increases with the increase of the cell opening clogging rate, resulting in the predicted ink transfer amount for the current printing batch. The predicted ink transfer amount is then linked to the batch ink transfer status data along the time axis on a batch-by-batch basis. The batch ink transfer deviation is obtained by comparing the predicted ink transfer amount with the rated ink transfer amount specified by the factory of the anilox roller. If the batch ink transfer deviation is lower than the preset ink transfer judgment threshold, the printing batch is determined to be a printing batch with smooth cell ink transfer. If the batch ink transfer deviation is not lower than the preset ink transfer judgment threshold, the printing batch is determined to be a printing batch with obstructed cell ink transfer, thus obtaining the ink transfer status classification result covering the current continuous printing batches.

[0055] In one embodiment, in order to classify the ink transfer status of batches on a continuous corrugated board printing production line, the thickness of the cell wall film obtained in the previous step and the color density of the first corrugated board print of the same batch are used as two raw materials to calculate the predicted ink transfer amount of the current printing batch, and thereby distinguish between printing batches with blocked cell ink transfer and printing batches with smooth cell ink transfer.

[0056] Specifically, based on the rule that the film thickness occupies the opening size in the cell geometry, the film thickness of the cell wall is normalized according to the original opening size specified by the factory of the anilox roller.

[0057] For example, if the factory-calibrated initial opening size of a certain type of anilox roller is 25 micrometers, and the calculated cell wall film thickness at the end of the batch is 5 micrometers, then the normalized cell opening blockage rate is 0.2, reflecting the proportion of the cell opening cross-section occupied by the deposited film layer at the end of the current batch. Furthermore, for the color density of the first corrugated cardboard print in the same batch, the baseline transfer amount per unit area is converted according to the linear correlation between the solid color patch reflection density and the ink transfer amount obtained in advance offline using a standard step color chart.

[0058] For example, when the color density reading of the first sheet is 1.5, the baseline transfer amount per unit area is approximately 1.45 grams per square meter, obtained by looking up the table according to the linear correlation. This reflects the baseline level of ink transfer from the cells to the corrugated board printing surface at the beginning of the batch. The cell opening clogging rate and the baseline transfer amount per unit area are linked as the same record in the batch data table with the printing batch number as the primary key, resulting in batch ink transfer status data that carries both film clogging and color density dimensions.

[0059] In one possible implementation, based on the inhibition relationship between the cell opening clogging rate and the baseline transfer amount per unit area in the batch ink transfer status data, the clogging attenuation coefficient corresponding to the cell opening clogging rate is used as the adjustment ratio to perform a downward correction on the baseline transfer amount per unit area. The clogging attenuation coefficient is obtained by looking up a table using a monotonic mapping pre-fitted offline for the cell opening clogging rate, following the relationship t=t0*(1-c*b), where t0 is the baseline transfer amount per unit area, b is the cell opening clogging rate, c is the clogging attenuation coefficient, and t is the predicted ink transfer amount after downward correction. The clogging attenuation coefficient is looked up in a table segmented by cell opening clogging rate. When the clogging rate is less than 0.1, c is 0.8; when the clogging rate falls between 0.1 and 0.3, c is 1.0; and when the clogging rate exceeds 0.3, c is 1.2. The values ​​for each segment are obtained by fitting multiple batches of measured data from the anilox roller model and the water-based ink model during the offline calibration phase.

[0060] For example, when the cell opening clogging rate is 0.2 and the clogging attenuation coefficient is 1.0, the predicted ink transfer amount is 0.8 times the baseline transfer amount per unit area. The predicted ink transfer amount is linked to the batch ink transfer status data along the time axis on a batch-by-batch basis. Preferably, the batch ink transfer amount deviation is obtained by subtracting the predicted ink transfer amount from the rated ink transfer amount specified by the anilox roller at the factory and dividing the absolute value of the difference by the rated ink transfer amount.

[0061] For example, if the rated ink transfer amount of the anilox roller is 1.5 grams per square meter at the factory, and the predicted ink transfer amount for the current batch is 1.16 grams per square meter, the batch ink transfer amount deviation is approximately 0.227. Further, a preset ink transfer judgment threshold is set for the batch ink transfer amount deviation.

[0062] For example, 0.15 is used. If the deviation of the batch ink transfer amount is lower than the ink transfer determination threshold, the printing batch is determined to be a printing batch with smooth cell ink transfer; if the deviation of the batch ink transfer amount is not lower than the ink transfer determination threshold, the printing batch is determined to be a printing batch with obstructed cell ink transfer.

[0063] It should be noted that the above classification process is performed batch by batch for multiple printing batches on a continuous printing production line to obtain the ink transfer status classification results covering the current continuous printing batch. The ink transfer status category of each batch is recorded with the printing batch number as the link, reflecting the fluctuation of ink transfer status between batches caused by roller temperature rise and film accumulation during continuous printing.

[0064] Step S105: For printing batches where ink transfer to the cells is obstructed, compare the current cell opening size of the anilox roller with the cell opening size specified at the factory. Based on the comparison result, reduce the color density of the first corrugated cardboard print in the same batch to obtain the adjusted color density reading for the batch. For printing batches where ink transfer to the cells is smooth, assign the measured color density of the first print in the batch as the reference color density reading for the batch.

[0065] For printing batches determined to have obstructed ink transfer in the cells, a white light confocal displacement sensor installed on the side of the anilox roller performs non-contact measurements of multiple cell openings along the roller's axial direction. The arithmetic mean of the measurements taken during the batch's operation is used as the current cell opening size of the anilox roller. The factory-calibrated cell opening size of the anilox roller is read from the anilox roller's factory calibration file and aligned with the current cell opening size using the same measurement aperture. The difference between the current cell opening size and the factory-calibrated cell opening size is calculated, and the difference is divided by the factory-calibrated cell opening size to obtain the cell opening shrinkage rate. If the cell opening shrinkage rate exceeds a preset shrinkage response threshold, the color density reduction coefficient corresponding to the cell opening shrinkage rate is used as the reduction ratio to adjust the color density of the first corrugated cardboard print in the same batch. The color density reduction coefficient increases as the cell opening shrinkage rate increases, resulting in the adjusted color density reading for the batch, which serves as the baseline color density reading for that batch. For printing batches determined to have smooth ink transfer in the cells, there is no need to compare the cell opening size. The measured color density of the first sheet of the batch is used directly without any reduction processing to obtain the reference color density reading for the batch. The reference color density readings of the hindered batches and the reference color density readings of the smooth batches are merged into the same reference color density sequence with the printing batch number as the link, to obtain a set of batch reference color density readings covering the current consecutive printing batches.

[0066] In one embodiment, for printing batches on a continuous corrugated board printing production line that are classified into two categories by previous stages: blocked ink transfer or smooth ink transfer, a differentiated color density processing path is executed for each batch. A set of batch reference color density readings is obtained by using the printing batch number as a link, which is used as a reference baseline for subsequent hue drift warning.

[0067] Specifically, for printing batches determined to have obstructed ink transfer in the cells, a white light confocal displacement sensor is used to perform non-contact measurement of the cell opening size.

[0068] Understandably, the white light confocal displacement sensor utilizes the principle that different wavelengths of broadband white light are focused at different depths after passing through a dispersive objective lens. Based on the one-to-one correspondence between the wavelength of the recovered light and the depth of the focal point, the distance difference between the probe and the top surface of the wall of the measured mesh cell and the bottom surface of the mesh cell is obtained, reflecting the cross-sectional size of the opening position of the mesh cell.

[0069] For example, the probe is arranged along the axial direction of the anilox roller on the side of the roller body. During the batch operation phase, it scans 30 to 50 evenly distributed anilox cavity opening positions on the roller surface, records the cross-sectional dimension reading of each anilox cavity opening position, and then takes the arithmetic mean of the multiple readings as the current anilox roller cavity opening size. Further, the anilox roller cavity opening size calibrated at the factory is obtained by looking up the anilox roller factory calibration file according to the roller body model code.

[0070] For example, the factory-calibrated cell opening size of a 150 lines per inch anilox roller is 25 micrometers, and this value is archived along with the inspection report upon roller delivery. The current cell opening size of the anilox roller and the factory-calibrated cell opening size of the anilox roller are aligned with the same measuring diameter and the same depth reference to form a pair of directly subtractable dimensional readings, which are combined to form the cell opening size comparison data for this batch.

[0071] In one possible implementation, a difference quotient operation is performed on the comparison data of the anilox roller opening size. The opening size difference is obtained by subtracting the current opening size of the anilox roller from the factory-calibrated opening size of the anilox roller. The difference is then divided by the factory-calibrated opening size of the anilox roller to obtain the opening shrinkage rate of the anilox roller opening.

[0072] For example, the factory-calibrated cell opening size of the anilox roller is 25 micrometers. When the current cell opening size of the anilox roller is 22 micrometers, the cell opening shrinkage rate is 0.12. Preferably, a preset shrinkage response threshold is set for the cell opening shrinkage rate.

[0073] For example, a value of 0.05 is used. If the cell opening shrinkage rate is less than or equal to the shrinkage response threshold of 0.05, it is considered that the change in the cell opening size of the roller is within an acceptable range, and no reduction processing is triggered; if the cell opening shrinkage rate is strictly greater than 0.05, the color density reduction coefficient corresponding to the cell opening shrinkage rate is used as the reduction ratio to apply to the color density of the first corrugated cardboard printed in the same batch for reduction processing. The color density reduction coefficient is obtained by looking up a table in segments according to the cell opening shrinkage rate. To avoid boundary overlap, the segments are all determined by the left-open and right-closed method. That is, when the shrinkage rate is in the range of greater than 0.05 and less than or equal to 0.10, the color density reduction coefficient is 0.05; when the shrinkage rate is in the range of greater than 0.10 and less than or equal to 0.20, the color density reduction coefficient is 0.10; when the shrinkage rate is greater than 0.20, the color density reduction coefficient is 0.18. Taking a shrinkage rate of exactly 0.10 as an example, it is classified into the previous category, with a value of 0.05; when the shrinkage rate is exactly 0.20, it is classified into the middle category, with a value of 0.10. Overall, it shows a segmented pattern where the larger the shrinkage rate, the greater the reduction.

[0074] For example, when the first color density reading is 1.5 and the cell opening shrinkage rate is 0.12, the color density reduction factor is 0.10. After the reduction process, the adjusted color density reading of this batch is 1.35, which is used as the reference color density reading of this batch.

[0075] It should be noted that for printing batches determined to have smooth ink transfer, the cell opening shrinkage rate of such batches does not exceed the shrinkage response threshold, the change in the roller cell opening size is within an acceptable range, and the correspondence between the ink transfer amount and the color density of the first printed sheet remains within the factory calibration range. Therefore, there is no need to call the white light confocal displacement sensor to perform cell opening size comparison; the measured color density of the first sheet of the batch is directly assigned without any reduction processing to obtain the reference color density reading for that batch. Furthermore, the reference color density readings of the hindered batches and the reference color density readings of the smooth batches are merged into the same reference color density sequence using the printing batch number as the link, resulting in a set of batch reference color density readings covering the current continuous printing batches. The set of batch reference color density readings records the color density reference baseline of each batch after differential processing in numerical form at the batch level, reflecting the equivalent compensation result of the change in the roller cell opening size on the ink transfer capability during continuous printing.

[0076] Step S106: Obtain the color density reading of the current corrugated cardboard printing batch, compare it with the adjusted color density reading batch by batch, filter out the printing batches with color density readings lower than the adjusted color density readings, and output a hue drift warning signal for the batch.

[0077] During the operation of the current corrugated cardboard printing batch, a spectrophotometer is used to synchronously measure the solid color blocks of the printed products after the batch is turned on, obtaining the real-time color density reading of the current printing batch. The real-time color density reading is then compared batch by batch with the adjusted color density reading of the batch, according to the printing batch number. The adjusted color density reading is taken from the corresponding batch reading in the previous set of reference color density readings. If the real-time color density reading is lower than the adjusted color density reading, the batch is selected as a candidate batch for hue drift. For the candidate batches for hue drift, a hue drift warning signal carrying the printing batch number and the color density difference is output. The hue drift warning signal is transmitted along the fieldbus to the printing equipment control unit, which reads the color density difference value in the signal. If the color density difference value exceeds a preset cleaning trigger threshold, the anilox roller cleaning program is initiated; if the color density difference value does not exceed the preset cleaning trigger threshold, the ink supply adaptive recalibration operation is initiated.

[0078] In one embodiment, to address the online maintenance requirement for batch-to-batch color consistency on a continuous corrugated board printing production line, the batch reference color density reading set obtained from the preceding steps is used as a reference baseline. Real-time color density acquisition and batch-by-batch comparison are performed on the current printing batch to identify candidate batches for color drift and output a color drift warning signal required for closed-loop adjustment.

[0079] Specifically, during the operation of the current corrugated cardboard printing batch, the solid color blocks of the printed products after the batch is turned on are synchronously measured by an online spectrophotometer arranged in the receiving section to obtain the real-time color density reading of the current printing batch; the real-time color density reading is compared with the adjusted color density reading of the batch according to the printing batch number, and the adjusted color density reading is taken from the corresponding batch reading in the previous reference color density reading set.

[0080] For example, the current batch number is batch 5, the adjusted color density reading is 1.35, and the real-time color density reading is 1.28. Since the real-time color density reading is lower than the adjusted color density reading, batch 5 is determined to be a candidate batch for hue drift. Further, for the candidate batch for hue drift, a hue drift warning signal is encapsulated using two fields: the printing batch number and the color density difference. The color density difference is the difference between the adjusted color density reading and the real-time color density reading.

[0081] For example, the color density difference of the fifth batch mentioned above is 0.07. The hue drift warning signal is transmitted to the printing equipment control unit along the fieldbus, and the printing equipment control unit reads the color density difference value in the signal to determine the bifurcation.

[0082] For example, a preset cleaning trigger threshold of 0.05 is set for the color density difference value. If the color density difference value exceeds the cleaning trigger threshold, the anilox roller cleaning program is activated. Water-based cleaning liquid is sprayed onto the cell walls of the roller surface by a spray pipe arranged on the side of the anilox roller. The doctor blade scrapes off the dried film layer deposited on the cell walls and resets the ink path to the initial working pressure. If the color density difference value does not exceed the cleaning trigger threshold, the ink supply adaptive recalibration operation is activated. An alternating progressive strategy of first increasing the speed and then increasing the pressure is adopted. That is, the speed of the ink fountain roller is first increased by one level in increments of 5 revolutions per minute from the current level. After 20 seconds of stable operation, the real-time color density is collected. If the deviation between the reading and the target reference color density is still greater than the convergence tolerance of 0.02, the doctor blade pressure is increased by one level in increments of 0.05 MPa, and this process is repeated. An upper limit protection value is set for the above alternation process. The ink fountain roller speed shall not exceed 1.3 times the initial setting value, and the doctor blade pressure shall not exceed 0.5 MPa. Once either parameter reaches the upper limit, the current recalibration is stopped and the process is switched to the anilox roller cleaning program. When the deviation between the real-time color density reading and the target reference color density converges to within 0.02 for the first time, the current speed and pressure combination is locked and the recalibration is exited to compensate for the short-term ink transfer attenuation caused by the roller body temperature rise and maintain the color consistency between batches during continuous printing.

[0083] Step S107: The hue drift warning signal, the corresponding cell wall film thickness and heat accumulation characteristics are encapsulated as a time-series tag. At the same time, the long short-term memory network is triggered to update the prior constraints using the time-series tag, and output the corrugated board printing equipment operation status monitoring result containing the current anilox roller ink transfer efficiency.

[0084] After acquiring the hue drift warning signal, the corresponding cell wall film thickness, and the heat accumulation characteristics, the three are aligned and linked using the printing batch number as the primary key. For the same printing batch number, a timestamp is added to the hue drift warning signal, cell wall film thickness, and heat accumulation characteristics according to the warning occurrence time, and they are arranged in ascending order along the printing batch time axis to obtain a time-series label reflecting the dual-dimensional information of the batch's operating status and the hue drift event. Based on the time-series label, the combined values ​​of the cell wall film thickness and heat accumulation characteristics within the time-series label are used as incremental samples to fill the prior path corresponding to the initial ink transfer condition vector in the Long Short-Term Memory (LSTM) network. The LSTM network updates the prior constraint weights on the input gate candidate state calculation path in incremental training mode to obtain the prior constraint conditions reflecting the latest roller ink transfer attenuation law. Based on the updated prior constraints, the batch distribution of the heat accumulation features of the long short-term memory network in subsequent printing batches is extracted and superimposed with the hue drift warning signal. This yields a characterization of the current anilox roller ink transfer efficiency, which carries the triple indicators of cell wall film thickness, heat accumulation features, and hue drift warning batch. This characterization is then incorporated into the printing equipment operation status monitoring data table along the time axis to obtain the corrugated board printing equipment operation status monitoring results that include the current anilox roller ink transfer efficiency.

[0085] In one embodiment, to meet the closed-loop monitoring requirements of the anilox roller ink transfer efficiency in a continuous corrugated board printing production line, the hue drift warning signal obtained in the preceding step, the corresponding cell wall film thickness, and the heat accumulation characteristics are used as three raw materials and fed into the long short-term memory network output by the preceding step. The prior constraints are incrementally updated, and the current anilox roller ink transfer efficiency is calculated and fed into the printing equipment operation status monitoring data table to obtain the corrugated board printing equipment operation status monitoring results.

[0086] Specifically, three-source alignment and connection are performed for the hue drift warning signal, cell wall film thickness and heat accumulation characteristics under the same printing batch number.

[0087] For example, taking the 5th printing batch as an example, the hue drift warning signal of this batch carries a color density difference of 0.07. The cell wall film thickness of this batch is recorded as the dry film thickness obtained by subtracting the film layer retreat amount after accumulating the entire batch, with an exemplary value of 8.5 micrometers. The heat accumulation characteristics of this batch include three components: total heat deposition, temperature rise steepness, and the proportion of high temperature maintenance period. These three sources are linked as the same record in the batch data table with the 5th printing batch as the primary key. Furthermore, a timestamp is added to each linked record according to the time of the warning occurrence, with the timestamp accuracy down to the second, and they are arranged in ascending order along the printing batch time axis to form a time sequence label reflecting the dual-dimensional information of batch operation status and hue drift event. The time sequence label carries the hue drift occurrence trajectory in the continuous printing process in the form of batch-by-batch entries.

[0088] In one possible implementation, incremental sample backfilling is performed on the cell wall conjunctival thickness and heat accumulation characteristics in the time-series labels.

[0089] It is understandable that the initial ink transfer condition vector mentioned in the preceding steps plays the role of a priori constraint in the long short-term memory network, and is connected to the input gate candidate state calculation path in the form of an additional constraint term. The incremental sample backfilling method is to combine the three components of the cell wall membrane thickness and heat accumulation characteristics of each batch record in the time series label into a priori correction sample, and use the hue drift warning signal value of that batch as the deviation label of that sample.

[0090] For example, if the color density difference of the batch falls between 0.03 and 0.05 and is accompanied by a hue drift warning, it is determined to be a slight deviation from the operating condition, and the deviation label is set to 1; if the difference is less than 0.03 and there is no hue drift warning, it is determined to be a stable operating condition, and the deviation label is set to 0; if the difference reaches or exceeds 0.05, the batch of samples is determined to have crossed the cleaning trigger line and will no longer enter the LSTM training set, but will be handled by the anilox roller cleaning procedure described in S106. After cleaning, the re-collected batch is entered into the database with a new printing batch number, and the label is re-determined according to the above two ranges, thereby ensuring that the cleaning event and the label generation do not overlap. The prior correction sample, together with the deviation label, is written into the prior path corresponding to the initial ink transfer operating condition vector in the long short-term memory network with the printing batch number as the primary key.

[0091] Preferably, in the incremental training mode, the long short-term memory network performs parameter updates on the prior constraint weights on the input gate candidate state calculation path, with an update step size of 0.001 to 0.005, to avoid historical operating condition knowledge being quickly covered by new samples.

[0092] For example, in incremental training mode, a backpropagation process using small batches of samples is employed. For the currently accumulated prior correction samples, the scalar output of the fully connected layer at the end of the Long Short-Term Memory network after Sigmoid activation is used as the hue shift tendency output, with a value range of 0 to 1. This value is on the same scale as the 0 or 1 value of the bias label, and the mean squared error is used as the loss function between the two. Based on the backpropagation gradient, the prior constraint weights are iteratively adjusted with a finite step size. The learning rate is 0.001, and the number of iterations is between 20 and 50. After the update, the prior constraint conditions reflecting the latest ink transfer attenuation law of the roller are obtained. Furthermore, for subsequent printing batch operation stages, the Long Short-Term Memory network re-executes heat accumulation feature extraction based on the updated prior constraints. The extracted heat accumulation features are superimposed and aligned with the batch distribution of the hue drift warning signal in the batch dimension. Specifically, an alignment matrix M is constructed with the batch number as the row index. The first few columns of the matrix are filled with the values ​​of each component of the heat accumulation feature. For example, if an 8-dimensional feature is taken, the first 8 columns are filled in sequentially. The last column is filled with the intensity value of the hue drift warning signal of the same batch, with the value range normalized to 0 to 1. Then, the matrix is ​​weighted and fused column by column. The fusion weight is 0.6 for the heat accumulation feature component and 0.4 for the warning signal intensity. The output result is the batch characterization vector reflecting the current ink transfer efficiency of the anilox roller. The vector dimension is consistent with the number of feature components.

[0093] For example, the same batch distribution dataset is organized with the printing batch number as the primary key and the amplitude of the heat accumulation characteristic component and the hue drift warning signal trigger flag as field columns.

[0094] In one embodiment, the batch distribution dataset and the cell wall film thickness are summarized batch by batch along the printing batch time axis to obtain a current anilox roller ink transfer efficiency characterization that carries the triple indicators of cell wall film thickness, heat accumulation characteristics, and color drift warning batch, reflecting the current anilox roller's comprehensive efficiency level in transferring water-based ink to the porous printing surface of corrugated cardboard.

[0095] It should be noted that the current anilox roller ink transfer efficiency is incorporated into the printing equipment operation status monitoring data table along the time axis. The printing equipment operation status monitoring data table is organized with the printing batch number as the primary key and the three indicators as the field columns, to obtain the corrugated board printing equipment operation status monitoring results containing the current anilox roller ink transfer efficiency, reflecting the evolution of equipment operation status under the relationship between roller temperature rise, film accumulation and hue drift during continuous printing. Another implementation method involves encapsulating the hue drift warning signal, the corresponding cell wall film thickness, and heat accumulation characteristics into a time-series tag. Simultaneously, a long short-term memory network is triggered to update prior constraints using this time-series tag, outputting a monitoring result of the corrugated board printing equipment's operating status that includes the current anilox roller's ink transfer efficiency. This involves acquiring the hue drift warning signal, the corresponding cell wall film thickness, and the heat accumulation characteristics. The cell wall film thickness is derived from the film thickness value obtained by pre-press laser confocal scanning of the anilox roller's cell wall and conversion using a grayscale difference algorithm, with units in micrometers and a value range of 0.5 to 8.0. Using the printing batch number as the primary key, the three are aligned and linked. For the same printing batch number, a timestamp is added to the hue drift warning signal, cell wall film thickness, and heat accumulation characteristics according to the warning occurrence time, and these are arranged in ascending order along the printing batch time axis to obtain a time-series tag reflecting both batch operating status and hue drift event information. The initial ink transfer condition vector is composed of five components: roller surface temperature, ink viscosity, cell volume residual rate, ink supply pressure, and roller speed, spliced ​​in a fixed order. It is sent to the candidate state calculation path of the input gate as the initial hidden state value of the first time step of the long short-term memory network. Inside the input gate, it corresponds to a parallel prior path. This path enters the candidate memory generation node in parallel with the original input vector of the input gate in a splicing manner. Based on the time-series label, the combined values ​​of the thickness of the conjunctival membrane of the cavity wall and the heat accumulation characteristics within the time-series label are used as incremental samples to fill the prior path corresponding to the initial ink transfer condition vector in the Long Short-Term Memory network. The prior constraint weights are attached between the linear transformation output of the input gate and the tanh activation input of the candidate memory, and are introduced in the form of additional constraint terms multiplied element by element. The weight dimension is consistent with the hidden layer dimension of the candidate memory, taking 64 dimensions. The initialization adopts a normal distribution with a mean of 0 and a standard deviation of 0.01. The Long Short-Term Memory network performs backpropagation parameter updates on the prior constraint weights in the incremental training mode with a learning rate of 0.001 and a batch size of 32. The other native gating parameters remain frozen, thus obtaining the prior constraint conditions that reflect the latest ink transfer attenuation law of the roller.Based on the updated prior constraints, the batch distribution of the heat accumulation features of the long short-term memory network in subsequent printing batches is extracted and superimposed with the hue drift warning signal. This yields a characterization of the current anilox roller ink transfer efficiency, which carries the triple indicators of cell wall film thickness, heat accumulation features, and hue drift warning batch. This characterization is then incorporated into the printing equipment operation status monitoring data table along the time axis to obtain the corrugated board printing equipment operation status monitoring results that include the current anilox roller ink transfer efficiency.

[0096] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the concept of this application. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for real-time monitoring of the operating status of corrugated cardboard printing equipment, characterized in that, The method includes: Obtain the time sequence record of the surface temperature of the anilox roller in a continuous corrugated board printing batch and the color density of the first corrugated board printed in the same batch; The time-series recording of the surface temperature of the anilox roller is input into a long short-term memory network. Combined with the initial ink transfer conditions characterized by the color density of the first corrugated cardboard print as a priori constraint, the heat accumulation characteristics that match the thermal response differences of the current batch of water-based inks on the porous printing surface of the corrugated cardboard are extracted. Analyze the duration of the temperature rise in the heat accumulation characteristics to determine the thickness of the conjunctival membrane in the cavity wall; By combining the thickness of the cell wall film with the color density of the first corrugated cardboard print in the same batch, the ink transfer amount can be predicted, and printing batches with obstructed cell ink transfer and printing batches with smooth cell ink transfer can be distinguished. For printing batches where ink transfer to cells is obstructed, the current cell opening size of the anilox roller is compared with the cell opening size specified at the factory. Based on the comparison results, the color density is reduced to obtain the adjusted color density reading. For printing batches where ink transfer to cells is smooth, the measured color density of the first sheet is assigned as the reference color density reading. Obtain the color density reading of the current printing batch and compare it with the adjusted color density reading batch by batch. Filter out printing batches with color density readings lower than the adjusted color density readings and output a hue drift warning signal.

2. The method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to claim 1, characterized in that, The acquisition of the time sequence record of the surface temperature of the anilox roller in a continuous corrugated board printing batch and the color density of the first corrugated board printed in the same batch includes: The surface temperature of the anilox roller is collected, and the temperature values ​​at each sampling time are spliced ​​together in chronological order to form a temperature time sequence record and labeled with the printing batch number. A spectrophotometer is used to measure the color density of the first printed product of the batch, and the color density value is read as the color density of the first corrugated cardboard printed product, and stored in association with the temperature time sequence record according to the batch number.

3. The method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to claim 1, characterized in that, The process involves inputting the time-series recording of the surface temperature of the anilox roller into a long short-term memory network, combining it with the initial ink transfer conditions characterized by the color density of the first corrugated cardboard print as a priori constraints, and extracting the heat accumulation characteristics that match the thermal response differences of the current batch of water-based inks on the porous printing surface of the corrugated cardboard, including: The time-series record of the surface temperature of the anilox roller is divided according to a preset time step. The mean and slope of each time step are calculated to obtain the temperature state sequence. For the color density of the first corrugated cardboard print, the initial ink transfer condition vector is determined by combining the linear correspondence between the color block reflection density and the ink transfer amount. The temperature state sequence and the initial ink transfer condition vector are aligned with the printing batch number and sent to the memory unit frame by frame. The forget gate discards historical temperature components that are irrelevant to the current thermal state, and the input gate writes the roller thermal state components. The initial ink transfer condition vector is used to suppress candidate states that deviate from the initial transfer level, and the hidden state sequence is output. After the hidden state sequence is dimensionally reduced by a fully connected mapping layer, it is split into the ink-side heat-induced film formation tendency component and the printing surface-side heat-induced ink absorption fluctuation component according to the water-based ink solvent evaporation temperature threshold range. The heat accumulation feature is obtained by combining them.

4. The method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to claim 1, characterized in that, The analysis of the duration of the temperature rise in the heat accumulation characteristics to determine the thickness of the cell wall conjunctiva includes: For the heat accumulation characteristics, a solvent evaporation trigger threshold is set along the time axis for the ink-side heat-induced film formation tendency component. Continuous time steps where the amplitude of the time axis component of the ink-side heat-induced film formation tendency component exceeds the solvent evaporation trigger threshold are merged into a heating duration period. The time step span and the average thermal intensity of each heating duration period are statistically analyzed to obtain the heating duration characteristics.

5. The method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to claim 4, characterized in that, After statistically analyzing the time step span and average thermal intensity of each heating period to obtain the characteristics of the heating period, the following is included: The time span in the heating period is divided into grades according to a preset segment interval and a corresponding reference conjunctival rate value is given. The reference conjunctival rate value is then corrected according to the average thermal intensity to obtain the graded conjunctival rate. The graded conjunctival rate is multiplied by the heating period feature segment by segment and then accumulated in chronological order. The membrane retreat amount calculated during the low temperature intermittent period is then deducted to obtain the cell wall conjunctival thickness.

6. The method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to claim 1, characterized in that, The method of predicting ink transfer amount by combining the cell wall film thickness with the color density of the first corrugated cardboard print in the same batch, distinguishing between printing batches with obstructed cell ink transfer and printing batches with smooth cell ink transfer, includes: The cell wall film thickness is normalized according to the original opening size calibrated by the anilox roller at the factory to obtain the cell opening blockage rate. Based on the color density of the first corrugated board print in the same batch, and combined with the linear correlation between color block reflection density and ink transfer amount, the baseline transfer amount per unit area is calculated. The baseline transfer amount per unit area is adjusted downwards using the blockage attenuation coefficient corresponding to the cell opening blockage rate to obtain the predicted ink transfer amount. The difference between the predicted ink transfer amount and the rated ink transfer amount calibrated by the anilox roller at the factory is used to determine the batch ink transfer amount deviation. Printing batches with a batch ink transfer amount deviation lower than the ink transfer judgment threshold are judged as printing batches with smooth cell ink transfer, while printing batches with a deviation not lower than the ink transfer judgment threshold are judged as printing batches with obstructed cell ink transfer.

7. The method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to claim 1, characterized in that, For printing batches where ink transfer to the cells is obstructed, the current cell opening size of the anilox roller is compared with the factory-calibrated cell opening size. Based on the comparison result, the color density is reduced to obtain an adjusted color density reading, including: Non-contact measurement of multiple cell openings along the roller axis is performed, and the arithmetic mean of the multi-point measurement values ​​is taken as the current cell opening size of the anilox roller. The difference between the current cell opening size and the factory-calibrated cell opening size is calculated and divided by the factory-calibrated cell opening size to obtain the cell opening shrinkage rate. When the cell opening shrinkage rate exceeds the shrinkage response threshold, the color density is reduced by a preset color density reduction coefficient to obtain the adjusted color density reading.

8. The method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to claim 1, characterized in that, The process of acquiring the color density reading of the current printing batch and comparing it batch by batch with the adjusted color density reading, filtering out printing batches with color density readings lower than the adjusted color density reading, and outputting a hue drift warning signal includes: Simultaneous measurement of solid color blocks of the current printing batch is performed to obtain color density readings; the color density readings are compared with the adjusted color density readings batch by batch number, and batches with color density readings lower than the adjusted color density readings are selected as hue drift candidate batches; for the hue drift candidate batches, a hue drift warning signal carrying the difference between the printing batch number and the color density is output.

9. A method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to claim 8, characterized in that, After outputting a hue drift warning signal carrying the difference between the printing batch number and the color density for the hue drift candidate batch, the process includes: The hue drift warning signal is transmitted to the printing equipment control unit along the fieldbus. The printing equipment control unit reads the color density difference. When the color density difference exceeds the cleaning trigger threshold, the anilox roller cleaning program is activated. When the color density difference does not exceed the cleaning trigger threshold, the ink supply adaptive recalibration operation is activated.

10. The method for real-time monitoring of the operating status of corrugated cardboard printing equipment according to claim 1, characterized in that, The method further includes: Using the printing batch number as the primary key, the hue drift warning signal, cell wall film thickness, and heat accumulation characteristics are aligned, linked, and timestamps are added. These are then arranged in ascending order of timestamps to obtain time-series labels. The combined values ​​of cell wall film thickness and heat accumulation characteristics within the time-series labels are used as incremental samples to backfill the prior constraints. The long short-term memory network updates the prior constraints in incremental training mode. Based on the updated prior constraints, the heat accumulation characteristics of subsequent printing batches are superimposed with the hue drift warning batch distribution to obtain the operating status monitoring results, which include the current anilox roller ink transfer efficiency.