A paperboard printing machine electrical operation state remote management and control method

Abstract

The application provides a paperboard printing machine electrical operation state remote management and control method, comprising: acquiring the air temperature data of the workshop outside the electrical cabinet and the temperature difference data of the front and back sides of the filter, filtering out random noise in the sensor collection process, obtaining the denoised workshop air temperature record and filter temperature difference record; according to the workshop air temperature record and filter temperature difference record, identifying whether there is a same direction following relationship between the temperature difference increment and the air temperature increment at each time, if there is no same direction following relationship, identifying that there is a suspected heat dissipation deterioration sign in the filter temperature difference record, forming a target temperature difference record; analyzing the amplitude fluctuation characteristics in the time dimension and the intensity distribution characteristics in the frequency dimension of the target temperature difference record, extracting trend characteristic quantities and cumulative characteristic quantities; combining the real-time working current and shell temperature data of each power device inside the electrical cabinet, screening out the weakest device set most sensitive to the blockage.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for remotely controlling the operating status of electrical components in a cardboard printing machine. Background Technology

[0002] Cardboard printing machines handle large-volume printing operations in packaging manufacturing. The electrical cabinet houses control modules such as frequency converters and drive boards, continuously releasing heat during operation. If heat dissipation is obstructed, rapid temperature rise in components can trigger shutdowns or even burn out. Remote monitoring and early warning of the electrical cabinet's heat dissipation status are crucial for ensuring continuous production line operation. Current remote monitoring relies heavily on temperature threshold alarms within the cabinet, triggering alerts when the set value is exceeded. This approach treats abnormal heat dissipation as an isolated event, ignoring the fact that multiple disturbances within the workshop can overlap at the same measuring point, making it difficult to distinguish whether the temperature rise originates from obstructed equipment heat dissipation or external operating condition fluctuations. The most typical disturbance in the workshop comes from corrugated paper scraps scattered during printing. These scraps adhere to the filter surface with the incoming airflow, gradually accumulating into loose deposits, narrowing the airflow channel and slowly widening the temperature difference before and after the filter. Logically, a more significant temperature difference indicates a more severe blockage. However, the workshop is not a constant temperature environment. Day and night cycles, machine start-ups and shutdowns, and the opening and closing of doors and windows all cause fluctuations of several degrees Celsius in the outside temperature, which in turn affects the temperature difference between the inlet and outlet airflow. Therefore, a large temperature difference is no longer the sole indicator of dust blockage; it could simply be a sign of external temperature fluctuations. When the remote platform observes a persistently high temperature difference across the filter at night, on-duty personnel may struggle to determine whether paper scraps have accumulated to the point of needing cleaning, or whether it's merely a false alarm caused by a drop in external temperature. If an alarm is triggered in every instance due to a large temperature difference, false alarms will inevitably occur frequently during periods of high temperature fluctuation. Conversely, simply raising the threshold for detection could allow genuine blockages to accumulate slowly and go unnoticed until components overheat. Summary of the Invention

[0003] This invention provides a method for remotely controlling the electrical operating status of a cardboard printing machine, mainly including: Acquire the temperature data of the workshop outside the electrical cabinet and the temperature difference data of the front and back sides of the filter screen, filter out the random noise in the sensor acquisition process, and obtain the noise-reduced workshop temperature record and filter screen temperature difference record. Based on the workshop temperature record and the filter temperature difference record, identify whether there is a co-current relationship between the temperature difference increment and the temperature increment at each time step. If there is no co-current relationship, identify signs of suspected heat dissipation deterioration in the filter temperature difference record and form a target temperature difference record. Analyze the amplitude fluctuation characteristics of the target temperature difference record in the time dimension and the intensity distribution characteristics in the frequency dimension, and extract trend features and cumulative features; The trend feature and the cumulative feature are concatenated into a blockage identification vector. The blockage identification vector is then input into a support vector machine classification model that has been trained in advance with samples labeled with paper scrap accumulation level in the historical cleaning cycle. The output is the blockage level determination result of the current filter. By combining the real-time operating current and case temperature data of each power device inside the electrical cabinet, the set of weak devices most sensitive to blockage was selected. When the clogging level determination result reaches the preset cleaning level, the set of weak components and their corresponding safe operating limits, along with the filter cleaning warning command, are pushed to the remote control platform. Based on the safe operating limits, the printing speed and paper feeding frequency of the cardboard printing machine are reduced and adjusted until the filter cleaning operation is completed on-site.

[0004] Furthermore, the acquisition of air temperature data from the workshop outside the electrical cabinet and temperature difference data between the front and rear sides of the filter screen, filtering out random noise during the sensor acquisition process, and obtaining denoised workshop air temperature records and filter screen temperature difference records includes: deploying temperature sensors at the workshop air inlet, deploying temperature sensors on the air inlet and outlet sides of the filter screen respectively, taking the difference between the readings on the air inlet and outlet sides of the filter screen as the temperature difference between the front and rear sides of the filter screen, forming an original air temperature sequence and an original temperature difference sequence according to a unified sampling period; applying median filtering to the original air temperature sequence and the original temperature difference sequence to remove instantaneous jump points, and then using wavelet thresholding to denoise and smooth residual jitter components.

[0005] Furthermore, based on the workshop temperature record and the filter temperature difference record, the step of identifying whether there is a co-current relationship between the temperature difference increment and the temperature increment at each time step, and if there is no co-current relationship, identifying signs of suspected heat dissipation deterioration in the filter temperature difference record to form a target temperature difference record, includes: calculating the difference between two adjacent samples of the workshop temperature record and the filter temperature difference record point by point to obtain the temperature increment sequence and the temperature difference increment sequence; for each sampling time, if the temperature increment and the temperature difference increment have the same sign, it is marked as the time when the temperature disturbance is dominant; if the temperature increment and the temperature difference increment have opposite signs, or the temperature increment amplitude is within the static threshold while the temperature difference increment amplitude exceeds the rise threshold, it is marked as the time when heat dissipation deterioration is suspected; expanding the time window forward and backward with the suspected heat dissipation deterioration time as the center, extracting continuous segments of the filter temperature difference record within the time window, and filling the reading points of the temperature disturbance dominant time in the segment with linear interpolation of the readings of the suspected heat dissipation deterioration times on both sides to form the target temperature difference record.

[0006] Furthermore, the analysis of the amplitude fluctuation characteristics of the target temperature difference record in the time dimension and the intensity distribution characteristics in the frequency dimension, and the extraction of trend features and cumulative features, includes: segmenting continuous sub-segments of the target temperature difference record using a sliding window, arranging the differences between the upper and lower envelopes of each sub-segment in chronological order to form an amplitude fluctuation sequence; obtaining the energy spectrum by applying a fast Fourier transform to the amplitude fluctuation sequence, dividing it into low-frequency and high-frequency segments according to preset frequency points, and reading the energy proportions of the low-frequency and high-frequency segments; multiplying the overall increase of the amplitude fluctuation sequence by the energy proportion of the low-frequency segment to obtain the trend features, and multiplying the cumulative value of the amplitude fluctuation at each moment by the sum of the energy proportions of the low-frequency and high-frequency segments to obtain the cumulative features.

[0007] Furthermore, the analysis of the target temperature difference record's amplitude fluctuation characteristics in the time dimension and intensity distribution characteristics in the frequency dimension, extracting trend features and cumulative features, includes: calculating the difference between adjacent sampled values ​​for the target temperature difference record at each time step and accumulating them point by point to obtain the cumulative amount of paper scrap deposition; calculating the standard deviation of the workshop temperature record using a sliding window to obtain the temperature disturbance intensity; if the cumulative amount of paper scrap deposition exceeds the cumulative dominant threshold and the temperature disturbance intensity is lower than the disturbance residual threshold, it is marked as the deposition dominant time; otherwise, linear interpolation of the readings of the adjacent two deposition dominant times is used to fill in the gaps to obtain the deposition dominant temperature difference record; empirical mode decomposition of the deposition dominant temperature difference record, taking the overall increase amplitude of the lowest frequency intrinsic mode function component as the trend feature, and taking the point-by-point accumulation value of the absolute values ​​of the remaining intrinsic mode function components as the cumulative feature.

[0008] Furthermore, the process of concatenating trend features and cumulative features into a blockage identification vector, and inputting the blockage identification vector into a support vector machine classification model pre-trained with samples labeled with paper scrap accumulation levels within historical cleaning cycles, to output the current filter blockage level determination result, includes: concatenating trend features and cumulative features at the same time into a two-dimensional vector in a preset order as the blockage identification vector; extracting the blockage identification vectors before each cleaning from the historical cleaning cycle operation archives as samples, classifying them into three levels—light, moderate, and heavy—according to the thickness of the paper scrap deposition layer and adding labels, and using radial basis kernel functions in a one-to-one multi-class classification manner to obtain the support vector machine classification model on the training set; inputting the current blockage identification vector into the support vector machine classification model and summarizing it through voting to obtain the blockage level determination result.

[0009] Furthermore, by combining the real-time operating current and case temperature data of each power device inside the electrical cabinet, the set of weak devices most sensitive to blockage is screened out. This includes: installing current transformers and surface-mounted temperature sensors on the frequency converter, rectifier module, inverter module, and relay control module inside the electrical cabinet, and collecting the real-time operating current and case temperature readings of each power device at the same sampling time to record the operating status; for each power device, the larger of the case temperature rise within the observation period divided by the square of the average real-time operating current during the same period and the lower limit of the current is taken to obtain the temperature rise per unit current; the difference between the temperature rise per unit current and the benchmark temperature rise calibrated in the non-blockage state is taken and then divided by the benchmark temperature rise per unit current to obtain the temperature rise deviation ratio; the power devices are sorted in descending order of the temperature rise deviation ratio to obtain the sensitivity ranking; the mapping relationship between the blockage level and the interception ranking is queried according to the blockage level determination result, and the set of weak devices is extracted from the sensitivity ranking.

[0010] Furthermore, when the blockage level determination result reaches the preset cleaning level, the vulnerable device set and its corresponding safe operating limit, along with the filter cleaning warning instruction, are pushed to the remote control platform. This includes: for each power device in the vulnerable device set, taking the difference between the upper limit of the withstand temperature and the current case temperature reading as the remaining temperature rise margin, taking the difference between the current case temperature reading and the case temperature reading before the backtracking time by dividing by the backtracking time as the measured temperature rise rate, dividing the remaining temperature rise margin by the measured temperature rise rate to obtain the single-machine withstand time, and taking the minimum value of all single-machine withstand times as the safe operating limit; packaging the vulnerable device set, the safe operating limit, and the filter cleaning warning instruction into a cleaning warning message, and pushing it to the remote control platform via the industrial communication link.

[0011] Furthermore, the set of weak components, safe operating limits, and filter cleaning warning instructions are packaged into a cleaning warning message, which is then pushed to the remote control platform via the industrial communication link. This includes: querying a pre-established speed reduction level mapping table using the blockage level determination result and the safe operating limit level as a two-dimensional index, and retrieving the printing speed reduction ratio and paper feeding frequency reduction ratio that match the current index.

[0012] Furthermore, after retrieving the printing speed reduction ratio and paper feeding frequency reduction ratio that match the current index, the process includes: issuing a graded speed reduction command to the main control circuit of the cardboard printing machine according to the printing speed reduction ratio and paper feeding frequency reduction ratio, so that the printing speed and paper feeding frequency are simultaneously reduced to the corresponding levels and maintained; and after receiving the job completion receipt transmitted from the site, issuing a parameter recovery command to restore the printing speed and paper feeding frequency to the normal operating parameters recorded before the filter cleaning.

[0013] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a remote control method for the electrical operation status of a cardboard printing machine. Addressing the unique scenario where paper scraps gradually accumulate on the surface of the electrical cabinet filter during cardboard printing, leading to deteriorated heat dissipation that is difficult to distinguish from temperature fluctuations caused by changes in workshop temperature, the method first denoises the workshop temperature and the temperature difference data before and after the filter. By judging whether the temperature difference increment and the temperature increment follow the same direction at each time step, environmental interference is removed to identify the target temperature difference record suspected of heat dissipation deterioration. Then, using the cumulative sum of temperature difference differences as the accumulated amount of paper scrap deposition and the sliding standard deviation of temperature as the disturbance intensity, it is confirmed that the temperature difference change is dominated by paper scrap accumulation. Empirical mode decomposition is performed on the target temperature difference record to extract low-frequency trend features representing slow deposition and high-frequency cumulative features representing local cluster disturbances. These are concatenated into a blockage identification vector, which is input into a support vector machine classification model to output the blockage level. Finally, by combining power device current and case temperature to screen weak device sets, when the cleaning level is reached, the safe operating limit and warning instructions are pushed to the remote control platform. Graded speed reduction is implemented for printing speed and paper feeding frequency, achieving accurate blockage identification and collaborative protection of the production line. Attached Figure Description

[0014] Figure 1 This is a flowchart of a method for remotely controlling the operating status of electrical components in a cardboard printing machine according to the present invention.

[0015] Figure 2 This is a schematic diagram of a method for remotely controlling the electrical operating status of a cardboard printing machine according to the present invention.

[0016] Figure 3 This is another schematic diagram of a remote control method for the electrical operating status of a cardboard printing machine according to the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0018] like Figures 1-3 This embodiment of a method for remotely controlling the electrical operating status of a cardboard printing machine may specifically include: Step S101: Obtain the temperature data of the workshop outside the electrical cabinet and the temperature difference data of the front and back sides of the filter screen, filter out the random noise in the sensor acquisition process, and obtain the noise-reduced workshop temperature record and filter screen temperature difference record.

[0019] Temperature sensors are deployed at the air inlet of the workshop outside the electrical cabinet to collect workshop air temperature. Two sets of temperature sensors are deployed on the air inlet and air outlet sides of the filter screen respectively. The temperature difference between the front and rear sides of the filter screen is obtained by subtracting the temperature reading on the air outlet side from the temperature reading on the air inlet side. The workshop air temperature and the temperature difference are used to form original air temperature sequences and original temperature difference sequences respectively according to a uniform sampling period. To remove the random noise mixed in the original air temperature sequences and the original temperature difference sequences, median filtering within a sliding window is used to remove instantaneous jump points that occur during the acquisition process. Then, wavelet threshold denoising is applied to the median-filtered sequences to smooth the residual jitter components, resulting in denoised workshop air temperature records and filter screen temperature difference records.

[0020] In one embodiment, the electrical cabinet in the cardboard printing machine workshop is usually installed near the printing host, and the air environment outside the cabinet is significantly affected by the scattering of corrugated paper scraps and the opening and closing of doors and windows.

[0021] For example, a temperature sensor is deployed in the free space of the workshop, more than 1.5 meters away from the air inlet of the electrical cabinet and avoiding local heat sources and air recirculation, to collect the overall air temperature of the workshop. This temperature serves as the baseline temperature of the workshop environment, reflecting the gradual change trend of the large-scale thermal environment. A temperature sensor is attached to the center of the filter's windward side, no more than 20 millimeters from the filter surface, as the inlet-side measuring point. The reading reflects the near-field temperature of the airflow that actually passes through the air inlet and is about to enter the filter, influenced by the local wind speed at the air inlet and the heat radiation from the cabinet. A temperature sensor is attached to the corresponding position on the leeward side of the filter as the outlet-side measuring point, reflecting the airflow temperature after passing through the filter. Subtracting the outlet-side reading from the inlet-side reading yields the temperature difference between the front and back sides of the filter, used to characterize the filter's blockage status. Subtracting the workshop air temperature reading from the inlet-side reading determines whether there are abnormalities such as backflow or obstruction at the air inlet. The two are not interchangeable. The three sensors sample synchronously according to a unified sampling period, forming the original workshop air temperature sequence, the original inlet air sequence, and the original temperature difference sequence, respectively.

[0022] Specifically, median filtering is used to remove isolated jump points caused by electromagnetic interference from components and instantaneous voltage fluctuations during the acquisition process.

[0023] In one possible implementation, a window of odd length is slid along the original temperature sequence and the original temperature difference sequence.

[0024] For example, a window length of 5 sampling points is used. The sampled values ​​within the window are sorted by size, and the value at the middle position after sorting is taken to replace the original reading at the center point of the window. This process is repeated point by point to complete the processing of the entire sequence, resulting in the median temperature sequence and the median temperature difference sequence after median filtering. Compared to mean processing, median processing is not skewed by individual outlier readings with large deviations and can preserve the true inflection points of temperature changes.

[0025] Preferably, the intermediate temperature sequence and the intermediate temperature difference sequence are further processed using wavelet thresholding to remove residual jitter components. The wavelet thresholding uses a db4 wavelet basis, decomposing the sequence into 3 to 5 layers of wavelet coefficients. For each layer's high-frequency coefficients, the noise standard deviation σ is estimated by dividing the median of the absolute value of the high-frequency coefficients by 0.6745. Then, the threshold for that layer is calculated using the general threshold formula T=σ*sqrt(2*ln(N)), where N is the length of the high-frequency coefficients in that layer, and ln represents the natural logarithm. High-frequency coefficients with amplitudes lower than T are set to zero, and high-frequency coefficients with amplitudes higher than T are contracted by the amplitude of T towards zero, while retaining the low-frequency approximation coefficients unchanged. The sequence is then reconstructed using inverse wavelet transform to obtain the denoised workshop temperature record and filter temperature difference record, respectively.

[0026] Step S102: Based on the workshop temperature record and the filter temperature difference record, identify whether there is a co-current relationship between the temperature difference increment and the temperature increment at each time step. If there is no co-current relationship, identify signs of suspected heat dissipation deterioration in the filter temperature difference record and form a target temperature difference record.

[0027] The workshop temperature record and the filter temperature difference record are sampled point by point at a uniform sampling time, and the difference between two adjacent samples is calculated to obtain the temperature increment sequence and the temperature difference increment sequence. For each sampling time, the signs of the temperature increment and the temperature difference increment are determined: if the signs are the same, the time is determined to be in a co-current relationship and marked as the temperature disturbance-dominant time; if the signs are opposite, or the amplitude of the temperature increment is within a preset static threshold while the amplitude of the temperature difference increment exceeds a preset rise threshold, the time is determined to be out of the co-current relationship and marked as a suspected heat dissipation deterioration time. A time window of a preset duration is extended forward and backward from the suspected heat dissipation deterioration time, and continuous segments of the filter temperature difference record within the time window are extracted. For the reading points corresponding to the temperature disturbance-dominant time still interspersed in the segment, the linear interpolation values ​​of the readings of the suspected heat dissipation deterioration time on both sides are used to fill the position. The filled segments are spliced ​​together according to the original sampling time sequence to form a target temperature difference record focusing on signs of heat dissipation deterioration.

[0028] In the operational scenario of a cardboard printing machine workshop, the increase in temperature difference before and after the filter screen could originate from either the accumulation of corrugated paper scraps on the filter screen surface or from the drop in outside temperature caused by day-night cycles, machine start-ups and shutdowns, and the opening and closing of doors and windows. If alarms are only based on the absolute value of the temperature difference, the former type of genuine blockage and the latter type of temperature fluctuation cannot be distinguished. Therefore, it is necessary to utilize the dynamic relationship between the workshop temperature records and the filter screen temperature difference records to differentiate between the two types of temperature increases at their source.

[0029] Specifically, the temperature increment sequence is obtained by subtracting the readings at two adjacent sampling times in the workshop temperature record point by point. Similarly, the temperature difference sequence is obtained by subtracting the filter temperature difference record at the same sampling clock. The timestamps of the two increment sequences are kept aligned, forming the basis for subsequent determination of their unidirectional following relationship.

[0030] In one embodiment, the determination of the same-direction following relationship uses symbol consistency as the primary criterion.

[0031] For example, when the workshop temperature rises from 28 degrees Celsius to 30 degrees Celsius during the day, the temperature on the inlet side rises accordingly. The temperature difference across the filter widens synchronously due to the slightly delayed temperature rise on the outlet side. At this time, both the temperature increment and the temperature difference increment are positive, and their signs are consistent. Conversely, when the workshop temperature slowly declines from 26 degrees Celsius to 22 degrees Celsius at night, the temperature on the inlet side declines synchronously, and the temperature difference converges accordingly. Both the temperature increment and the temperature difference increment are negative, and their signs remain consistent. The sampling times with consistent signs in both cases are marked as the dominant time for temperature disturbance.

[0032] It should be noted that relying solely on sign consistency to determine the same-direction following relationship is insufficient to cover periods when the temperature is almost stable but paper scraps are slowly accumulating. During such periods, the magnitude of the temperature increment is within a very small range, and the sign direction no longer has a stable meaning. Therefore, a combination of a static threshold and a rising threshold is introduced: when the absolute value of the temperature increment is within a preset static threshold, while the absolute value of the temperature difference increment exceeds a preset rising threshold, the sampling moment is determined to have deviated from the same-direction following relationship and is marked as a suspected moment of deteriorating heat dissipation.

[0033] For example, the static threshold is set at 0.05 degrees Celsius per sampling period, and the rise threshold is set at 0.2 degrees Celsius per sampling period. The former corresponds to the judgment threshold of basically stable workshop temperature, and the latter corresponds to the judgment threshold of significant increase in temperature difference. In addition to the above combined judgment, sampling moments where the temperature increment and temperature difference increment have opposite signs are also classified as suspected heat dissipation deterioration moments. The reason is that a decrease in temperature should cause the air outlet side of the filter to decrease synchronously, and the temperature difference should converge. If the temperature difference increases instead, it indicates that the contribution of the inlet and outlet airflow to the temperature difference has significantly deviated from the normal response of temperature fluctuations. Thus, the dominant time of temperature disturbance and the suspected heat dissipation deterioration moment constitute mutually exclusive labels for the filter temperature difference record. Furthermore, with each suspected heat dissipation deterioration moment as the center, a time window of a preset duration is extended before and after that moment.

[0034] For example, the time window is extended by 15 minutes on one side, and a continuous segment covering this time window is extracted from the filter temperature difference record. If multiple adjacent time windows of suspected heat dissipation deterioration times overlap, the overlapping segments are merged into the same continuous segment to avoid the same temperature difference rise process being divided into multiple segments.

[0035] In one possible implementation, the extracted continuous segments will still contain a few readings corresponding to moments dominated by temperature disturbances. For these readings, linear interpolation is performed using the readings of the nearest preceding and following moments of suspected heat dissipation deterioration within the segment. The interpolated result replaces the original reading at that position, ensuring the segment remains continuously and evenly spaced in the sampling sequence. All interpolated and filled continuous segments are then stitched together sequentially according to the original sampling sequence to obtain the target temperature difference record focusing on signs of heat dissipation deterioration. This target temperature difference record retains only the periods where the co-current following relationship is broken, thus eliminating the spurious rise caused by temperature fluctuations and providing a clean input sequence for subsequent characterization of the amplitude and frequency features of paper scrap deposition.

[0036] Step S103: Analyze the amplitude fluctuation characteristics of the target temperature difference record in the time dimension and the intensity distribution characteristics in the frequency dimension, and extract trend features and cumulative features.

[0037] The target temperature difference record is segmented into continuous sub-segments along the time dimension using a sliding window. For each sub-segment, the difference between the upper and lower envelopes of the target temperature difference record is extracted as an instantaneous representation of amplitude fluctuation. The amplitude fluctuations of each sub-segment are arranged in chronological order to form an amplitude fluctuation sequence, thereby obtaining an amplitude fluctuation sequence reflecting the trend of amplitude change along the time dimension. For the amplitude fluctuation sequence, a fast Fourier transform is used to transform it from the time dimension to the frequency dimension to obtain an energy spectrum. The energy spectrum is divided into low-frequency and high-frequency bands by predefined frequency points. The energy proportions of the low-frequency and high-frequency bands are read as intensity distributions along the frequency dimension, thereby determining the dominant frequency band distribution of the target temperature difference record along the frequency dimension. The trend characteristic is obtained by multiplying the overall increase of the amplitude fluctuation sequence from the start time to the current time by the low-frequency energy ratio. The cumulative characteristic is obtained by multiplying the cumulative value of the amplitude fluctuation value at each time of the sequence by the sum of the low-frequency energy ratio and the high-frequency energy ratio. Thus, the trend characteristic and cumulative characteristic that characterize the amplitude fluctuation characteristics and frequency intensity distribution characteristics of the target temperature difference record are obtained.

[0038] The target temperature difference record still contains, in its temporal sequence, an overall upward trend caused by the slow accumulation of paper scraps, and short-term turbulence components caused by the localized shedding of paper scraps and instantaneous airflow turbulence. It is difficult to separate these two types of components solely from the magnitude of the original temperature difference value. It is necessary to characterize the fluctuations in amplitude over time and the distribution of energy over frequency, thereby converging them into trend features and cumulative features that can be used for subsequent identification.

[0039] In one embodiment, the length of the sliding window is exemplarily 60 sampling points, and the sliding step size is exemplarily 10 sampling points. Continuous segments are gradually truncated from left to right along the sampling time sequence of the target temperature difference record. Adjacent segments overlap at their start and end positions to avoid the jitter at the edge of the segment being truncated.

[0040] Specifically, for each sub-segment, the upper and lower envelopes of the target temperature difference record are extracted. A sliding neighborhood with a half-width of 5 sampling points is taken centered on the sampling point. If the value of the center point is strictly greater than the other values ​​within the range of 5 points to its left and right, it is recorded as a local maximum; if it is strictly less than, it is recorded as a local minimum. When equal values ​​appear in the neighborhood, the center sampling point of the parallel segment is taken as the representative extreme value point to avoid double counting. The first and last sampling points of the sub-segment are always forced to participate in the fitting as endpoint extreme values, and the values ​​of the first and last points are simultaneously included in the endpoint sets of the upper and lower envelopes to ensure that the envelope has a clear start and end at both ends of the sub-segment. If fewer than 2 local maxima are identified within a sub-segment, the upper envelope is supplemented by linear extrapolation of the values ​​of the first and last points of the sub-segment; the lower envelope is processed in the same way. The upper envelope is formed by connecting the local maxima of the temperature difference within a sub-segment in a piecewise linear manner in chronological order, and the lower envelope is formed by connecting the local minima in the same manner. Both envelopes have a unique value at each sampling time. Subtracting the lower envelope value from the upper envelope value at the same sampling time yields the instantaneous envelope difference. Averaging all instantaneous envelope differences within a sub-segment over time yields the amplitude fluctuation value for that sub-segment. Arranging the amplitude fluctuation values ​​of each sub-segment in chronological order of their center times constitutes an amplitude fluctuation sequence, which reflects the fluctuation trend of amplitude over time. When the paper scraps are still loosely stacked, airflow can pass through the gaps, and the amplitude fluctuation value is at a low level; as the stacking becomes denser and local clusters detach, the amplitude fluctuation value shows a significant jump. Furthermore, the amplitude fluctuation sequence needs to be characterized in the frequency dimension to distinguish whether the amplitude fluctuation originates from slow accumulation or short-term jitter.

[0041] In one possible implementation, a fast Fourier transform is applied to the amplitude fluctuation sequence to transform the time-dimensional sequence to the frequency-dimensional sequence, thereby obtaining an energy spectrum covering all frequency points. The horizontal axis of the energy spectrum represents the frequency point, and the vertical axis represents the energy magnitude of the component corresponding to that frequency point.

[0042] Preferably, the energy spectrum is divided into low-frequency band and high-frequency band by dividing frequency points according to a preset frequency band.

[0043] For example, the frequency band division point is taken as one-tenth of the sampling Nyquist frequency. Frequency points below the division point are classified as low-frequency bands, and frequency points above the division point are classified as high-frequency bands. The energy ratio of the low-frequency band is obtained by dividing the sum of the energy of each frequency point in the low-frequency band by the sum of the energy of all frequency points in the energy spectrum, denoted as p1; the energy ratio of the high-frequency band is obtained by dividing the sum of the energy of each frequency point in the high-frequency band by the sum of the energy of all frequency points in the energy spectrum, denoted as p2. The low-frequency band energy ratio characterizes the proportion of the slow trend component caused by the gradual accumulation of paper scraps in the amplitude fluctuations, and the high-frequency band energy ratio characterizes the proportion of the short-term jitter component caused by the airflow disturbance caused by the paper scrap clusters in the amplitude fluctuations.

[0044] In one embodiment, the arithmetic mean of five consecutive sampling points from the start time of the amplitude fluctuation sequence is taken as the baseline value b to avoid accidental jitter at the start time; then, the maximum value of the sequence from the start time to the current time is subtracted from b to obtain the overall rise amplitude h. The trend feature quantity F1 is calculated as F1 = h * p1, where p1 is the proportion of low-frequency energy, ranging from 0 to 1. When p1 is not less than 0.6, the energy distribution is considered to be concentrated in the low-frequency band, F1 is approximately equal to h, and the overall rise amplitude is preserved; when p1 is not greater than 0.3, the energy distribution is biased towards the high-frequency band, F1 does not exceed 0.3 * h, short-term jitter is suppressed, and false alarms of stacking trends are avoided. This calculation method relies on the energy proportion weighting obtained by fast Fourier transform as the main calculation process of F1 in this embodiment; the F1 calculation method based on empirical mode decomposition is presented later as an optional alternative embodiment, and either one can be used.

[0045] Specifically, the amplitude fluctuation values ​​of the amplitude fluctuation sequence from the start of the current cleaning cycle to the current time are accumulated point by point. The value of the accumulated sequence at the current time is taken as the amplitude accumulation, denoted as Sa, to distinguish it from the paper debris deposition accumulation C used in the dual threshold determination step later. The two respectively characterize the stacking intensity of the amplitude channel and the differential channel, and do not share signs or substitute for each other. The accumulated characteristic quantity is denoted as F2, which is obtained by F2=Sa*p1-Sa*p2, where p1 is the low-frequency energy proportion and p2 is the high-frequency energy proportion. Both values ​​are between 0 and 1 and have been normalized in the previous frequency domain analysis. Sa*p1 indicates that the amplitude accumulation is positively weighted by the low-frequency proportion, and Sa*p2 indicates that the same accumulation is negatively deducted by the high-frequency proportion. When the energy distribution concentrates in the low-frequency range, for example, when p1 reaches 0.7 and p2 drops to 0.15, F2 increases in the positive direction, indicating continuous stacking; when the energy distribution concentrates in the high-frequency range, for example, when p1 drops to 0.2 and p2 rises to 0.6, F2 is correspondingly suppressed to avoid overestimating the amplitude accumulation during the period dominated by cluster disturbances. The trend characteristic F1 and the cumulative characteristic F2 are used as outputs to characterize the amplitude fluctuation characteristics and frequency intensity distribution characteristics of the target temperature difference record, and are delivered together to the subsequent processing stages for determining the dominance of paper debris deposition and identifying the clogging level.

[0046] One method for determining trend and cumulative characteristics is as follows: calculate the cumulative sum of the time-by-time differences of the target temperature difference record as the cumulative amount of paper scrap deposition, calculate the standard deviation within the sliding window of the workshop temperature record as the intensity of temperature disturbance, and confirm that the temperature difference change is dominated by the gradual stacking of paper scraps on the filter surface rather than caused by residual temperature fluctuations based on the cumulative amount of paper scrap deposition and the intensity of temperature disturbance. Perform empirical mode decomposition on the target temperature difference record to obtain the low-frequency component characterizing the slow deposition trend of paper scraps and the high-frequency component characterizing the local cluster disturbance airflow of paper scraps, so as to determine the trend and cumulative characteristics.

[0047] For the target temperature difference record, the difference between two adjacent sampled values ​​is calculated moment by moment to obtain a difference sequence reflecting the small changes in temperature difference at each moment. This difference sequence is added point by point from the starting moment to the current moment to obtain a cumulative sum curve. The value of the cumulative sum curve at the current moment is taken as the cumulative amount of paper debris deposition. Simultaneously, the workshop temperature record is slid along the sampling time sequence by a preset length window. The standard deviation of the temperature reading in each window is calculated. The standard deviations of each window are arranged in chronological order according to the center moment of the window. The standard deviation corresponding to the current moment is taken as the temperature disturbance intensity. For the cumulative amount of paper debris deposition and the temperature disturbance intensity at the same moment: if the cumulative amount of paper debris deposition exceeds a preset cumulative dominance threshold and the temperature disturbance intensity is lower than a preset disturbance residual threshold, then it is confirmed that the change of the target temperature difference record at the current moment is dominated by the gradual accumulation of paper debris, and the current moment is marked as the deposition dominance moment; if neither condition is met, it is marked as the residual disturbance moment. The linear interpolation value of the readings of the deposition dominance moments on both sides is used to fill the position of this moment to obtain a time-series continuous deposition dominance temperature difference record. The deposition-dominant temperature difference record is subjected to empirical mode decomposition, which separates it into several intrinsic mode function components arranged from high to low frequency. The intrinsic mode function component with the lowest frequency is taken as the low-frequency component characterizing the slow deposition trend of paper scraps. The composite sequence obtained by adding the remaining intrinsic mode function components point by point at corresponding time is taken as the high-frequency component characterizing the local cluster disturbance airflow of paper scraps. The overall rise of the low-frequency component from the start time to the current time is taken as the trend characteristic quantity, and the point-by-point cumulative value of the absolute value of the high-frequency component from the start time to the current time is taken as the cumulative characteristic quantity.

[0048] In the cardboard printing machine workshop, the target temperature difference record is first preprocessed: using a 10-minute window, the workshop air temperature sequence and temperature difference sequence are taken at the same time, and the Pearson correlation coefficient between the two is calculated. When the coefficient is greater than 0.7 and the proportion of time when the two change in the same direction exceeds 60%, the window is marked as a co-current following period and removed from the temperature difference record, retaining only the remaining time period as input for subsequent judgment. After this separation, the large temperature drag component in the temperature difference record has been removed, but due to short-term factors such as the opening and closing of doors and windows, personnel movement, and air supply disturbances, there may still be residual traces of temperature fluctuations between 0.2 and 0.5 degrees Celsius. To further separate the actual temperature difference rise caused by the gradual accumulation of paper scraps on the filter surface from the residual temperature fluctuations, two independent observations need to be characterized simultaneously at the same time: one describes the trend after the accumulation of small temperature difference changes, and the other describes the dispersion intensity of the workshop air temperature near that time. The comparison results of the two observations with the threshold are used as the basis for the dominant judgment.

[0049] In one embodiment, the difference between two adjacent sampled values ​​is calculated time-by-time for the target temperature difference record, resulting in a difference sequence. Each element in this difference sequence reflects the direction and magnitude of the minute temperature difference variation between adjacent sampling times. If paper scraps are continuously accumulating on the filter surface, the filter ventilation cross-section gradually narrows, and the airflow temperature on the outlet side remains relatively lower than that on the inlet side, the elements in the difference sequence will mostly be positive. The difference sequence is then summed point-by-point from the start of the current cleaning cycle to the current time. Positive elements superimpose, and the cumulative sum curve shows a monotonically rising trend. If residual temperature fluctuations dominate, positive and negative elements cancel each other out, and the cumulative sum curve shows a slight oscillation. The value of the cumulative sum curve at the current time is taken as the accumulated amount of paper scrap deposition, denoted as C.

[0050] Specifically, the cumulative amount of paper scraps is physically equivalent to the total increase in temperature difference over the current cleaning cycle. Its ability to offset residual temperature fluctuations makes it more sensitive to the unidirectional process of paper scrap accumulation. Simultaneously with the calculation of the cumulative amount of paper scraps, the workshop temperature record is slid along the sampling time sequence by a window of a preset length.

[0051] For example, the window length is taken as the number of sampling points corresponding to 30 minutes. The standard deviation of the temperature reading in each window is calculated, and the standard deviations of each window are arranged in chronological order according to the center time of the window. The standard deviation corresponding to the current time is taken as the temperature disturbance intensity, denoted as D. The temperature disturbance intensity reflects the dispersion of the outside temperature in the period of time in the vicinity of that time. The larger the dispersion, the more active the external disturbances such as opening and closing of doors and windows, starting and stopping of the machine, etc., and the more likely the residual temperature fluctuations are to penetrate into the target temperature difference record.

[0052] Preferably, a dual threshold combination is set for determining the cumulative amount of paper scraps deposited and the intensity of temperature disturbance at the same time: if the cumulative amount of paper scraps deposited C exceeds the preset cumulative dominant threshold Ta, and the intensity of temperature disturbance D is lower than the preset disturbance residual threshold Tb, then it is confirmed that the temperature difference change at the current time is dominated by the gradual accumulation of paper scraps, and the current time is marked as the deposition dominant time.

[0053] For example, the cumulative dominant threshold Ta is 3.0 degrees Celsius, corresponding to the mid-range of the total temperature difference increment when the filter is newly installed to near the point where cleaning is required; the disturbance residual threshold Tb is 0.3 degrees Celsius, corresponding to the common magnitude of the sliding standard deviation of the workshop temperature during steady-state operation. The cumulative dominant threshold Ta corresponds to the judgment threshold of "a considerable increase in temperature difference has accumulated", and the disturbance residual threshold Tb corresponds to the judgment threshold of "sufficiently quiet external temperature". The two thresholds act independently.

[0054] It should be noted that if the accumulated amount of paper scraps, C, does not exceed Ta, or the intensity of the temperature disturbance, D, is not lower than Tb, then the current time is marked as the residual disturbance time. For such residual disturbance times, the readings of the nearest deposition-dominant times on both sides of the time sequence are used for linear interpolation to replace the original readings at that time position. This ensures that the target temperature difference record maintains the continuous and equally spaced sampling time sequence while eliminating residual disturbances, resulting in a deposition-dominant temperature difference record. Furthermore, the deposition-dominant temperature difference record still contains two components: one is the overall upward trend component caused by the slow accumulation of paper scraps, and the other is the short-term fluctuation component caused by the local clustering and shedding of paper scraps or slight disturbances in the airflow. Both are superimposed on the same temperature difference curve and cannot be separated by simple high-low division.

[0055] In one possible implementation, the deposition-dominant temperature difference record is subjected to empirical mode decomposition (EMD), adaptively separating it into several intrinsic mode function components arranged in descending order of frequency. The EMD process does not require pre-defined basis functions; it peels away high-frequency jitter components layer by layer based on the extreme value distribution of the signal itself. The remaining trend terms serve as input for the next round of decomposition until the remaining terms tend to monotonic. The resulting intrinsic mode function components are numbered from high to low frequency, with the last component with the largest number corresponding to the slowest change, and the preceding components with smaller numbers corresponding to faster jitter. The intrinsic mode function component with the lowest frequency is taken as the low-frequency component characterizing the slow deposition trend of paper scraps. The value of this low-frequency component at the start of the current cleaning cycle is used as the baseline. Subtracting the baseline from the current value of the low-frequency component yields the overall increase, which is used as the trend characteristic quantity F1. This trend characteristic quantity characterizes the steady-state contribution accumulated by the continuous stacking of paper scraps on the filter surface since the start of the current cleaning cycle. The remaining intrinsic mode function components are summed point-by-point at corresponding times to obtain a synthetic sequence, which serves as the high-frequency component characterizing the airflow disturbance caused by localized paper scrap clusters. The absolute values ​​of this high-frequency component are taken point-by-point from the start of the current cleaning cycle to the current time and accumulated to obtain the cumulative characteristic quantity F2. This cumulative characteristic quantity describes the disturbance intensity accumulated from the detachment of localized paper scrap clusters and the instantaneous airflow jitter within the same cleaning cycle. The trend characteristic quantity F1 and the cumulative characteristic quantity F2 together constitute a quantitative description of the slow deposition trend and localized cluster disturbance in the deposition-dominant temperature difference record.

[0056] Step S104: Concatenate the trend feature quantity and the cumulative feature quantity into a blockage identification vector. Input the blockage identification vector into a support vector machine classification model that has been trained in advance with sample data labeled with paper scrap accumulation level in the historical cleaning cycle, and output the blockage level determination result of the current filter.

[0057] The trend feature quantity and the cumulative feature quantity output at the same time are concatenated into a two-dimensional vector in a preset order. This two-dimensional vector is used as the blockage identification vector. Each dimension of the blockage identification vector corresponds to a feature component describing the current state of the filter. The blockage identification vector before each filter cleaning is extracted from the operation archives within the historical cleaning cycle as samples. Based on the thickness of the paper debris deposition layer measured on-site during cleaning, the samples are divided into three levels: light accumulation, moderate accumulation, and heavy accumulation according to preset thickness grading boundaries, and corresponding labels are attached. The labeled samples are aggregated into a training set. The radial basis function kernel function is used as the kernel function of the support vector machine classification model. The support vector and decision function coefficients of each binary hyperplane are solved on the training set in a one-to-one multi-class construction method to obtain the pre-trained support vector machine classification model. The input of the support vector machine classification model is the blockage identification vector, and the output is one of the three levels: light accumulation, moderate accumulation, and heavy accumulation. The current blockage identification vector is fed into the support vector machine classification model. The final category is obtained by summarizing the values ​​of each binary classification decision function through voting. The current blockage level of the filter is determined and the blockage level is output as the current filter blockage level determination result.

[0058] In the context of filter monitoring in a cardboard printing press workshop, the trend and cumulative characteristics alone are insufficient to directly determine clogging. Trend characteristics describe the steady-state contribution of paper scraps continuously accumulating on the filter surface, while cumulative characteristics describe the disturbance intensity accumulated from localized paper scrap clusters falling off and instantaneous airflow jitter within the same cleaning cycle. These two characteristics describe the current deposition state of the filter from different dimensions and need to be combined in a unified way for classification and determination. Then, a classification model based on historical cleaning experience can be used to determine the clogging level.

[0059] In one embodiment, the trend feature quantity and the cumulative feature quantity output at the same time are concatenated into a two-dimensional vector in the order of trend feature quantity first and cumulative feature quantity last. This two-dimensional vector is used as the blockage identification vector. The first dimension of the blockage identification vector carries the steady-state stacking contribution, and the second dimension carries the short-term disturbance cumulative contribution.

[0060] Specifically, the concatenation order of the blockage identification vectors is kept consistent throughout the training and inference processes to avoid the interchange of the meanings referred to by the first and second dimensions during the training and inference phases, which could lead to deviations in subsequent classification decisions. Furthermore, a training set for the support vector machine classification model needs to be constructed from the operational archives within the historical cleaning cycles. The operational archives maintain a record for each filter cleaning operation, containing the blockage identification vector at the moment preceding the cleaning and the thickness of the paper debris deposition layer measured on-site during that cleaning. The paper debris deposition layer thickness is measured on-site using calipers in the main deposition areas of the filter surface, and the average of multiple measurements is taken as the deposition thickness reading for that cleaning operation.

[0061] In one possible implementation, samples are divided into three levels based on the thickness of the paper debris deposition layer according to preset thickness grading boundaries: thickness less than 1 mm is classified as light accumulation, thickness between 1 mm and 3 mm as medium accumulation, and thickness greater than 3 mm as heavy accumulation. The preset thickness grading boundaries are selected based on the engineering margin allowing for narrowing of the filter ventilation cross-section; 1 mm corresponds to a basically unobstructed ventilation cross-section, and 3 mm corresponds to a ventilation cross-section narrowed to the point requiring immediate cleaning. The blockage identification vector of each sample is paired with its corresponding level label and aggregated into a training set.

[0062] It should be noted that the Support Vector Machine (SVM) classification model is inherently a binary classification model. When directly applied to the determination of level three congestion, it needs to be extended with a one-to-one multi-classification construction method. The one-to-one multi-classification construction method trains an independent binary sub-model for any pairwise combination of the three stacking levels, training a total of three binary sub-models: light stacking to medium stacking, light stacking to heavy stacking, and medium stacking to heavy stacking.

[0063] Specifically, when training each binary classification sub-model, samples belonging to the two levels corresponding to that sub-model are selected from the training set to form a sub-training set. Samples belonging to one level are denoted as the positive class, and samples from the other level are denoted as the negative class. The radial basis function takes the squared Euclidean distance between sample pairs as the independent variable and decays exponentially. Its value is expressed as exp(-g*d), where d represents the squared Euclidean distance between two blockage identification vectors, and g is the bandwidth parameter, appearing in the negative multiplication factor position of the exponent. The larger the value of g, the faster the function decays and the more sensitive the similarity between sample pairs is to distance. For example, g is taken as the reciprocal of the median of the squared Euclidean distances between any two samples in the training set. Using the radial basis function kernel as the kernel, the classification hyperplane that maximizes the classification margin is solved on the sub-training set. During the solution process, a penalty parameter C is introduced to penalize samples that cross the classification margin. The larger the value of C, the more likely it is to reduce the training classification error; the smaller the value of C, the more likely it is to expand the classification margin. The recommended value of C is a specific numerical range, that is, a continuous value between 1.0 and 10.0. The side closer to 1.0 corresponds to a wider classification margin and stronger noise resistance, and the side closer to 10.0 corresponds to a more rigorous training fit.

[0064] For example, we first take C=1.0 as the baseline, and then select the final value based on the classification accuracy on the validation set at the four candidate points of 1.0, 2.0, 5.0, and 10.0. C is the regularization coefficient that controls the misclassification tolerance when classifying soft margins. Its mathematical essence is a weighting coefficient that punishes training samples that cross the classification margin, that is, fall within the margin band or are misclassified to the opposite side.

[0065] Specifically, when solving for the optimal classification hyperplane, the objective function is to minimize 0.5*||w||²+C*Σξi, where w is the hyperplane normal vector, ξi is the slack variable for each sample (measuring the degree to which the sample violates the margin), and the summation term is the sum of the slack variables of all training samples. C is multiplied before this penalty term; therefore, the larger C is, the heavier the penalty on the slack variables in the objective function, and the optimization process will try to reduce ξi, thus tending to classify the training samples as correctly as possible and satisfy the margin constraint. However, this may lead to model overfitting, a smaller margin, and a decrease in generalization ability. The smaller C is, the lighter the penalty, and the optimization process focuses more on expanding the classification margin (i.e., reducing ||w||²), allowing more samples to fall within the margin band or even be misclassified, thus obtaining a smoother decision boundary, but the classification accuracy on the training set may decrease. In the actual training of each binary sub-model, C, as a pre-defined hyperparameter, directly participates in the calculation of the objective function in solving the quadratic programming problem. It is transformed into a constraint in the dual problem using the Lagrange multiplier method, meaning the range of the Lagrange multiplier αi is limited to 0 ≤ αi ≤ C, thus affecting the selection of support vectors and the coefficients of the decision function. In the example, C is taken to be on the order of 1 to 10, which is an empirical range; the specific optimal value needs to be determined through cross-validation or performance evaluation on the validation set. For each pair of levels, such as mild and moderate, corresponding samples are extracted from the training set and assigned positive and negative class labels. Then, the similarity between samples is calculated using the radial basis function kernel, constructing a kernel matrix K, where... , where g is taken as the reciprocal of the median of the squared pairwise Euclidean distances between all samples in the training set. This value ensures that the kernel function's response scale is moderate for different sample pairs, avoiding extreme saturation or flattening of the kernel value due to an excessively wide distance distribution. Subsequently, under constraints... and Next, maximize the dual objective function. Solving for the Lagrange multipliers αi, we obtain the samples where αi > 0 are the support vectors, and the corresponding decision function coefficients are αiyi. The bias term b is calculated from any support vector using the margin boundary conditions. x s Let SV represent the set of all support vectors, where SV is the selected support vector. Each of the three sub-models independently completes the above solution and stores its own set of support vectors, corresponding αiyi coefficients, and bias b. During inference, each sub-model's decision on a new sample x is determined through a decision function. The symbols are given, with a positive sign indicating a positive class (e.g., mild) and a negative sign indicating a negative class (e.g., moderate). The support vectors and decision function coefficients of this sub-model are obtained by solving the problem. The support vectors and decision function coefficients of the three sub-models are saved together to obtain a pre-trained support vector machine classification model. The input of the support vector machine classification model is the blockage identification vector, and the output is one of three levels: mild blockage, moderate blockage, or severe blockage.

[0066] Preferably, after training, a validation subset is randomly stratified from the historical cleanup cycle samples at an 8:2 ratio, with the proportion of samples for each congestion level remaining consistent with the full set. The classification accuracy of the trained model is checked on the validation subset. If the classification accuracy is below 90%, hyperparameter testing begins, with an initial g of 0.5 and an initial C of 1.0 as the baseline. During testing, C is first fixed, and g is recursively multiplied by 2 or divided by 2 in one direction. The search range of g is limited to 0.03125 to 8. Once the g with the highest accuracy in this direction is obtained, it is fixed. Then, C is adjusted in the same way between 0.125 and 32. If the accuracy is still below 90% after one round of alternation between g and C, the next round of alternation begins, with a maximum of 5 rounds. Alternatively, the process may terminate early if the accuracy improvement is less than 0.5% for two consecutive rounds, thus avoiding oscillations. During the inference phase, the current blockage identification vector is fed into the support vector machine (SVM) classification model. Each of the three binary classification sub-models provides a classification opinion based on its stored support vectors and decision function coefficients. The final category is determined according to the following two-level rules: First, a vote is taken based on the frequency of each category, and the category with the most frequency is selected as the final category. If two categories have the same number of votes, the absolute values ​​of the decision function values ​​in the relevant sub-models are compared, and the category with the larger sum is selected as the final category. `g` affects the local scope of the kernel function; if `g` is too large, the kernel value becomes extremely sensitive to distance, easily overfitting noise; if `g` is too small, the kernel value becomes too smooth, making it difficult to distinguish different categories. `C` affects the model's tolerance to training error. Through repeated trials on the validation set, a hyperparameter combination that achieves a good balance between bias and variance can be found, ensuring the model has reliable generalization ability.

[0067] In one embodiment, if two of the three decisions from the three binary sub-models point to moderate stacking and one to light stacking, the final category is determined to be moderate stacking. If the three decisions point to three different levels and no level receives a majority vote, the absolute values ​​|f(x)| of the decision functions of all binary sub-models involved in that level are summed for each level. This absolute value reflects the certainty of the sub-model's decision (the larger the value, the further the sample is from the sub-model's classification hyperplane, and the higher the confidence). Since each level involves two sub-models, the |f(x)| of these two sub-models are added to obtain the confidence accumulation value for that level. The value with the largest accumulation value is taken as the final category to avoid gaps in the classification conclusion. The final category is output as the current filter's clogging level determination result. The clogging level determination result is presented directly as light stacking, moderate stacking, or heavy stacking, for subsequent processing steps involving the screening and cleaning of weak components and the issuance of early warning instructions.

[0068] Step S105: Combining the real-time operating current and case temperature data of each power device inside the electrical cabinet, the set of weak devices most sensitive to blockage is selected.

[0069] Inside the electrical cabinet, current transformers and surface-mount temperature sensors are installed on each power device, consisting of the frequency converter, rectifier module, inverter module, and relay control module on the drive board. Real-time operating current and case temperature readings of each power device are collected at the same sampling time. The real-time operating current and case temperature readings of the same device are paired sequentially to obtain the device's operating status record, thus creating a set of operating status records covering all power devices in the electrical cabinet. For each power device in the operating status record set, the rise in case temperature reading from the beginning to the end within a preset observation period is taken. This rise is divided by the square of the average real-time operating current during the same period and a preset lower current limit, taking the larger of the two values ​​to obtain the device's unit current temperature rise. The difference between this unit current temperature rise and a pre-calibrated baseline unit current temperature rise for the device under unclogged filter conditions is then divided by the baseline unit current temperature rise to obtain the device's temperature rise deviation ratio. All power devices are arranged in descending order of the temperature rise deviation ratio to obtain a ranking of the power devices' sensitivity to heat dissipation obstruction. Based on the congestion level determination result, query the preset mapping relationship between congestion level and interception rank, extract the interception rank count value corresponding to the congestion level determination result, and extract the power devices ranked within the interception rank count value from the sensitivity sort from front to back. Then, aggregate the extracted power devices into a set of weak devices that are most sensitive to congestion.

[0070] Inside the electrical cabinet of a cardboard printing press, the frequency converter, rectifier module, inverter module, and relay control module on the drive board jointly undertake the task of driving and distributing power to the printing host. Their temperature rise responses under the same blockage condition are not consistent: devices with high load current generate more heat and are more prone to heat accumulation when heat dissipation is obstructed; devices near the filter outlet experience the first obstruction of cooling airflow, resulting in a shorter heat dissipation path. Simply applying a uniform approach to the entire cabinet based on the blockage level determination cannot address these individual differences. It is necessary to select the group of weakest devices with the strongest response to heat dissipation obstruction based on the real-time operating current and case temperature readings of each power device.

[0071] Specifically, within the electrical cabinet, current transformers and surface-mounted temperature sensors are installed on each power device, consisting of the frequency converter, rectifier module, inverter module, and relay control module. The current transformers are mounted on the main circuit input cable of the device to collect real-time operating current, while the surface-mounted temperature sensors are attached to the heat dissipation surface of the device's casing to collect casing temperature readings. Both readings are uploaded synchronously at a unified sampling period. The real-time operating current and casing temperature readings of the same device are paired one-to-one according to the sampling time to obtain the device's operating status record. This pairing operation is repeated for each power device within the electrical cabinet, resulting in a collection of operating status records covering all power devices within the electrical cabinet.

[0072] In one embodiment, a preset observation period is taken for each power device in the set of operating status records.

[0073] For example, a duration of 30 minutes is used. Within this observation period, the temperature rise of the device's casing at the end of the observation period is subtracted from the temperature at the beginning of the observation period to obtain the temperature rise of the device within this period, denoted as dT. The average value of the real-time operating current during the same period is denoted as I. The larger of the square of I and the square of the preset lower current limit Q0 is taken as the denominator base. Here, Q0 is exemplarily taken as 10% of the rated current of the device, and then squared. This is used to avoid the square of I tending to zero when the device is under no-load or light-load conditions, which would cause the denominator to become unstable.

[0074] Specifically, the temperature rise per unit current U of the device is obtained by U=dT / max(II,Q0), where max represents the larger of the two values, and II represents the square of the average value of the real-time operating current over a period of time. The temperature rise per unit current U represents the increase in case temperature corresponding to the square of the unit current, and its amplified value reflects the extension of the heat residence time in the device case under the same current conditions.

[0075] In one possible implementation, the pre-calibrated unit current temperature rise of the device under unclogged filter conditions is taken as the reference unit current temperature rise U0. The unit current temperature rise U during the current observation period is subtracted from the reference unit current temperature rise U0, and then divided by the reference unit current temperature rise U0 to obtain the temperature rise deviation ratio of the device. The reference unit current temperature rise U0 is calibrated according to the aforementioned U calculation process within the same preset observation period after the new filter is put into operation. The U0 value of each power device is used for a long time after the initial calibration. The temperature rise deviation ratio reflects the degree of deviation of the device from its normal heat dissipation conditions under the current heat dissipation conditions, and is comparable for power devices with different rated capacities under the same dimension. Furthermore, all power devices in the electrical cabinet are arranged in descending order according to their respective temperature rise deviation ratio values. The device ranked first has the most significant increase in unit current temperature rise relative to the reference and the strongest response to heat dissipation obstruction; the device ranked last has a unit current temperature rise close to the reference level and the weakest response to heat dissipation obstruction. This yields a ranking of the power devices' sensitivity to heat dissipation obstruction.

[0076] It should be noted that the number of devices included in the vulnerable device set varies depending on the level of congestion: for mild congestion, only a few devices with the most prominent response need to be considered, while for severe congestion, the range of affected devices expands significantly. Therefore, a mapping relationship between congestion level and cutoff ranking is pre-established: mild congestion corresponds to cutoff ranking 1, moderate congestion corresponds to cutoff ranking 3, and severe congestion corresponds to cutoff ranking 6. The mapping relationship is established as follows: the temperature rise deviation ratio exceeding the warning threshold is used as the benchmark condition for determining whether a device is affected by congestion. The warning threshold is a temperature rise deviation ratio δ = 15%, meaning that when the measured temperature rise of a device deviates from its rated temperature rise by more than 15%, the device is considered to have an abnormal response under the current congestion level. Based on historical operational data from 12 consecutive cleaning cycles (each cycle lasting approximately 720 hours, covering about 200 components in the same type of unit), the number of components with temperature rise deviations exceeding the warning threshold under light, moderate, and severe accumulation conditions were counted in each cleaning cycle. The median value of the 12 cycles for each blockage level was taken: 1 for light accumulation, 3 for moderate accumulation, and 6 for severe accumulation. This directly determined the ranking for each level. The above mapping relationship and the warning threshold values ​​were written into the configuration items of the remote control platform for long-term use. If the statistical sample is subsequently expanded to 24 or more cleaning cycles, the median values ​​for each level should be recalculated and the configuration items updated.

[0077] Preferably, based on the congestion level determination result, the preset mapping relationship between congestion level and interception rank is queried, and the interception rank value corresponding to the current congestion level is retrieved. Power devices ranked within the interception rank value are then intercepted from the sensitivity sorting list, and these intercepted power devices are aggregated into a set of vulnerable devices most sensitive to congestion. This set of vulnerable devices is presented as a device identifier list for subsequent processing steps involving safe operating limit estimation and speed reduction control command issuance.

[0078] Step S106: When the clogging level determination result reaches the preset cleaning level, the set of weak components and their corresponding safe operating limits, along with the filter cleaning warning command, are pushed to the remote control platform. The remote control platform performs graded speed reduction control on the printing speed and paper feeding frequency of the cardboard printing machine according to the safe operating limits, until the filter cleaning operation is completed on-site and the normal operating parameters are restored.

[0079] The blockage level determination result is compared with the preset cleaning level. If the blockage level determination result reaches the cleaning level, then for each power device in the weak device set, the upper limit of the device's withstand temperature is subtracted from the current case temperature reading to obtain the remaining temperature rise margin. The difference between the current case temperature reading of the device and the case temperature reading of the device before the preset backtracking time is divided by the preset backtracking time to obtain the measured temperature rise rate. The remaining temperature rise margin is divided by the measured temperature rise rate to obtain the single-unit withstand time of the device. The minimum value of the single-unit withstand time of all devices in the weak device set is taken as the safe operating limit of the weak device set. The set of vulnerable components, the safe operating limit, and the filter cleaning early warning command are packaged into a cleaning early warning message according to a preset message format and pushed to the remote control platform via an industrial communication link. The remote control platform uses the blockage level determination result and the safe operating limit as a two-dimensional index to query a pre-established speed reduction level mapping table. Each entry in the speed reduction level mapping table stores the printing speed reduction ratio and paper feeding frequency reduction ratio under the corresponding index. The reduction ratio matching the current index is retrieved. The remote control platform issues graded speed reduction commands to the main control circuit of the cardboard printing machine according to the printing speed reduction ratio and the paper feeding frequency reduction ratio. The printing speed and paper feeding frequency of the cardboard printing machine are synchronously reduced to the corresponding level and maintained. After the filter cleaning operation is completed on-site, the on-site operation panel sends a work completion receipt back to the remote control platform. After receiving the work completion receipt, the remote control platform issues a parameter recovery command to the main control circuit to restore the printing speed and paper feeding frequency to the normal operating parameters recorded before the filter cleaning.

[0080] During continuous printing operations on a cardboard printing press, as filter screens gradually accumulate buildup from light to heavy, directly stopping the machine for cleaning would disrupt the entire packaging production line. Continuing to run at full speed with the original parameters would risk overheating the vulnerable components. Therefore, a controllable transition period is needed: on the one hand, the production line is notified to prepare for cleaning; on the other hand, the operating intensity of the printing press is reduced according to the temperature rise tolerance of the vulnerable components, thus providing a window of time for cleaning without exceeding the components' temperature limits.

[0081] In one embodiment, the blockage level determination result is compared with a preset cleaning level.

[0082] For example, the cleanup level is set to heavy buildup. If the blockage level determination result reaches the cleanup level, a safe operating time limit estimation process for the vulnerable device set is triggered. If the blockage level determination result does not reach the cleanup level, the normal monitoring cycle is maintained, and this step is not initiated.

[0083] Specifically, for each power device in the set of vulnerable devices, the upper limit of the withstand temperature Tmax given on the device's nameplate or in the device manual is taken, and the current case temperature reading Tc is taken. The remaining temperature rise margin M of the device is obtained by subtracting Tc from Tmax. To suppress single-sampling jitter, both Tc and Tp are substituted by the arithmetic mean of the three most recent sampling points. At the same time, the difference between the current case temperature reading Tc of the device and its case temperature reading Tp before the preset backtracking time L is taken. The difference is divided by the preset backtracking time L to obtain the measured temperature rise rate V of the device, where L is exemplarily taken as the sampling span corresponding to 5 minutes. To ensure computational stability under boundary conditions, an effectiveness threshold Vmin is set for V, which is 0.05 degrees Celsius per minute. When V is greater than or equal to Vmin, M is divided by V to obtain the single-machine tolerance time t of the device. When V is less than Vmin, including three cases: V is zero, V is negative, or V is a small positive value below the threshold, it is determined that the current case temperature of the device does not show an effective upward trend, and t is directly assigned the preset upper limit Tcap.

[0084] For example, 60 minutes is used, and the device is marked as a non-constraint item in this round of judgment, and is not included in the minimum value comparison of Tsafe. t represents the time required for the case temperature of the device to rise from its current value to the upper limit of its withstand temperature while maintaining the current current load and heat dissipation obstruction state. The single-machine withstand time t of all power devices in the weak device set is iterated. After removing those marked as non-constraint items, the smallest value among the remaining devices is taken as the safe operating limit of the weak device set, denoted as Tsafe. If all devices are marked as non-constraint items, Tsafe is directly set to Tcap, indicating that there is no temperature rise approach constraint under the current operating condition. The safe operating limit corresponds to the time margin determined by the device in the weak device set that first reaches the upper limit of its withstand temperature.

[0085] In one possible implementation, the list of device identifiers for the vulnerable device set, the safe operating time limit (Tsafe), and the filter cleaning warning instruction are packaged into a cleaning warning message according to a preset message format. The preset message format arranges the following fields sequentially: message header, congestion level, vulnerable device identifier, safe operating time limit, warning timestamp, and message tail. This message is then pushed to the remote management platform via the industrial communication link between the cardboard printing machine's field controller and the remote management platform.

[0086] Preferably, a speed reduction level mapping table is pre-established within the remote control platform. This speed reduction level mapping table uses a two-dimensional index based on the blockage level determination result and the corresponding safe operating time limit: the blockage level dimension is expanded along three levels: light blockage, moderate blockage, and heavy blockage, denoted as L1, L2, and L3 respectively; the safe operating time limit dimension is divided into three levels according to the left-closed, right-open principle, denoted as T1, T2, and T3 respectively. T1 corresponds to a safe operating time limit t satisfying 0 <= t < 15 minutes, T2 corresponds to 15 minutes <= t < 60 minutes, and T3 corresponds to t >= 60 minutes. That is, the endpoint of 15 minutes is assigned to the T2 level, and the endpoint of 60 minutes is assigned to the T3 level, avoiding boundary overlap or omissions. When looking up the table, a unique entry is located using the combination of Li and Tj. Each entry stores the printing speed reduction ratio and the paper feeding frequency reduction ratio. If the calculated safe operating time limit falls on the endpoint of a level, the reduction ratio of the corresponding entry is strictly used according to the above-mentioned allocation rules, ensuring that the control output under the same operating condition is unique and reproducible.

[0087] For example, heavy buildup with a safe operating time of less than 15 minutes corresponds to a 50% reduction in printing speed and a 50% reduction in paper feeding frequency; heavy buildup with a safe operating time between 15 and 60 minutes corresponds to a 30% reduction in printing speed and a 30% reduction in paper feeding frequency; and heavy buildup with a safe operating time greater than 60 minutes corresponds to a 10% reduction in printing speed and a 10% reduction in paper feeding frequency. Upon receiving the cleaning warning message, the remote control platform retrieves the blockage level determination result and the safe operating time Tsafe, uses the corresponding gear as an index to find the corresponding reduction ratio in printing speed and paper feeding frequency, and sends a graded speed reduction command to the main control circuit of the cardboard printing machine. The printing speed and paper feeding frequency of the cardboard printing machine are simultaneously reduced to the corresponding gear and maintained. During the speed reduction period, the blockage level determination and weak component screening process continues. If the blockage level or safe operating time crosses a new gear, the remote control platform re-queries the speed reduction gear mapping table and sends an updated graded speed reduction command.

[0088] It should be noted that after the on-site maintenance personnel complete the filter cleaning operation, they trigger a work completion receipt via the on-site operation panel. This receipt is then transmitted back to the remote control platform via the industrial communication link. Upon receiving the work completion receipt, the remote control platform sends a parameter restoration command to the main control loop, restoring the printing speed and paper feeding frequency to the normal operating parameters recorded before the filter cleaning, thus returning the printing operation of the cardboard printing machine to its normal cycle time.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for remotely controlling the operating status of electrical components in a cardboard printing machine, characterized in that, The method includes: Acquire the temperature data of the workshop outside the electrical cabinet and the temperature difference data of the front and back sides of the filter screen, filter out the random noise in the sensor acquisition process, and obtain the noise-reduced workshop temperature record and filter screen temperature difference record. Based on the workshop temperature record and the filter temperature difference record, identify whether there is a co-current relationship between the temperature difference increment and the temperature increment at each time step. If there is no co-current relationship, identify signs of suspected heat dissipation deterioration in the filter temperature difference record and form a target temperature difference record. Analyze the amplitude fluctuation characteristics of the target temperature difference record in the time dimension and the intensity distribution characteristics in the frequency dimension, and extract trend features and cumulative features; The trend feature and the cumulative feature are concatenated into a blockage identification vector. The blockage identification vector is then input into a support vector machine classification model that has been trained in advance with samples labeled with paper scrap accumulation level in the historical cleaning cycle. The output is the blockage level determination result of the current filter. By combining the real-time operating current and case temperature data of each power device inside the electrical cabinet, the set of weak devices most sensitive to blockage was selected. When the clogging level determination result reaches the preset cleaning level, the set of weak components and their corresponding safe operating limits, along with the filter cleaning warning command, are pushed to the remote control platform. Based on the safe operating limits, the printing speed and paper feeding frequency of the cardboard printing machine are reduced and adjusted until the filter cleaning operation is completed on-site.

2. The method for remote control of the electrical operating status of a cardboard printing machine according to claim 1, characterized in that, The process of acquiring air temperature data from the workshop outside the electrical cabinet and temperature difference data from the front and back sides of the filter screen, filtering out random noise during sensor acquisition, and obtaining denoised workshop air temperature records and filter screen temperature difference records includes: deploying temperature sensors at the workshop air inlet, deploying temperature sensors on the air inlet and outlet sides of the filter screen respectively, taking the difference between the readings on the air inlet and outlet sides of the filter screen as the temperature difference between the front and back sides of the filter screen, forming an original air temperature sequence and an original temperature difference sequence according to a unified sampling period; applying median filtering to the original air temperature sequence and the original temperature difference sequence to remove instantaneous jump points, and then using wavelet thresholding to denoise and smooth residual jitter components.

3. The method for remote control of the electrical operating status of a cardboard printing machine according to claim 1, characterized in that, The process involves identifying whether there is a co-current relationship between the temperature difference increment and the temperature increment based on the workshop temperature record and the filter temperature difference record. If no co-current relationship exists, the filter temperature difference record is identified as showing signs of suspected heat dissipation deterioration, forming a target temperature difference record. This includes: calculating the difference between two adjacent samples of the workshop temperature record and the filter temperature difference record point by point to obtain the temperature increment sequence and the temperature difference increment sequence; for each sampling moment, if the temperature increment and the temperature difference increment have the same sign, it is marked as the temperature disturbance-dominant moment; if the temperature increment and the temperature difference increment have opposite signs, or the temperature increment amplitude is within the static threshold while the temperature difference increment amplitude exceeds the rise threshold, it is marked as a suspected heat dissipation deterioration moment; expanding the time window forward and backward with the suspected heat dissipation deterioration moment as the center, extracting continuous segments of the filter temperature difference record within the time window, and filling the reading points of the temperature disturbance-dominant moment within the segment with linear interpolation of the readings of the suspected heat dissipation deterioration moments on both sides to form the target temperature difference record.

4. The method for remote control of the electrical operating status of a cardboard printing machine according to claim 1, characterized in that, The analysis of the target temperature difference record's amplitude fluctuation characteristics in the time dimension and intensity distribution characteristics in the frequency dimension, extracting trend features and cumulative features, includes: segmenting continuous sub-segments of the target temperature difference record using a sliding window, arranging the differences between the upper and lower envelopes of each sub-segment in chronological order to form an amplitude fluctuation sequence; using a fast Fourier transform on the amplitude fluctuation sequence to obtain the energy spectrum, dividing it into low-frequency and high-frequency segments according to preset frequency points, and reading the energy proportions of the low-frequency and high-frequency segments; multiplying the overall increase of the amplitude fluctuation sequence by the energy proportion of the low-frequency segment to obtain the trend feature, and multiplying the cumulative amplitude fluctuation value at each moment by the sum of the energy proportions of the low-frequency and high-frequency segments to obtain the cumulative feature.

5. The method for remote control of the electrical operating status of a cardboard printing machine according to claim 1, characterized in that, The analysis examines the amplitude fluctuation characteristics of the target temperature difference record in the time dimension and the intensity distribution characteristics in the frequency dimension, extracting trend features and cumulative features. This includes: calculating the difference between adjacent sampled values ​​for each time step of the target temperature difference record and accumulating them point by point to obtain the cumulative amount of paper debris deposition; calculating the standard deviation of the workshop temperature record using a sliding window to obtain the temperature disturbance intensity; if the cumulative amount of paper debris deposition exceeds the cumulative dominant threshold and the temperature disturbance intensity is lower than the disturbance residual threshold, it is marked as the deposition dominant moment; otherwise, linear interpolation of the readings of the adjacent two deposition dominant moments is used to fill in the gaps, obtaining the deposition dominant temperature difference record; empirical mode decomposition of the deposition dominant temperature difference record is performed, taking the overall increase amplitude of the lowest frequency intrinsic mode function component as the trend feature, and taking the point-by-point accumulation value of the absolute values ​​of the remaining intrinsic mode function components as the cumulative feature.

6. The method for remote control of the electrical operating status of a cardboard printing machine according to claim 1, characterized in that, The process of concatenating trend features and cumulative features into a blockage identification vector, inputting the blockage identification vector into a support vector machine classification model pre-trained with samples labeled with paper scrap accumulation levels within historical cleaning cycles, and outputting the current filter blockage level determination result includes: concatenating trend features and cumulative features at the same time into a two-dimensional vector in a preset order as the blockage identification vector; extracting the blockage identification vectors before each cleaning from the historical cleaning cycle operation files as samples, classifying them into three levels—light, moderate, and heavy—according to the thickness of the paper scrap deposition layer and adding labels, and using radial basis kernel functions in a one-to-one multi-class classification manner to obtain the support vector machine classification model on the training set; inputting the current blockage identification vector into the support vector machine classification model and summarizing through voting to obtain the blockage level determination result.

7. The method for remote control of the electrical operating status of a cardboard printing machine according to claim 1, characterized in that, The process involves combining real-time operating current and case temperature data of each power device inside the electrical cabinet to screen out the set of weak devices most sensitive to blockage. This includes: installing current transformers and surface-mounted temperature sensors on the inverter, rectifier module, inverter module, and relay control module inside the electrical cabinet to collect real-time operating current and case temperature readings of each power device at the same sampling time and pairing them as an operating status record; for each power device, taking the larger of the case temperature rise within the observation period divided by the square of the average real-time operating current during the same period and the current lower limit value to obtain the temperature rise per unit current; taking the difference between the temperature rise per unit current and the benchmark temperature rise calibrated in the non-blockage state and dividing it by the benchmark temperature rise per unit current to obtain the temperature rise deviation ratio; and sorting each power device in descending order of the temperature rise deviation ratio to obtain the sensitivity ranking of each device; and querying the mapping relationship from the blockage level to the extracted ranking based on the blockage level determination result to extract the set of weak devices from the sensitivity ranking.

8. The method for remote control of the electrical operating status of a cardboard printing machine according to claim 1, characterized in that, When the blockage level determination result reaches the preset cleaning level, the vulnerable device set, its corresponding safe operating limit, and the filter cleaning warning instruction are pushed to the remote control platform together. This includes: for each power device in the vulnerable device set, taking the difference between the upper limit of the withstand temperature and the current case temperature reading as the remaining temperature rise margin, taking the difference between the current case temperature reading and the case temperature reading before the backtracking time by dividing by the backtracking time as the measured temperature rise rate, dividing the remaining temperature rise margin by the measured temperature rise rate to obtain the single-unit withstand time, and taking the minimum value of all single-unit withstand times as the safe operating limit; packaging the vulnerable device set, the safe operating limit, and the filter cleaning warning instruction into a cleaning warning message, and pushing it to the remote control platform via the industrial communication link.

9. A method for remotely controlling the electrical operating status of a cardboard printing machine according to claim 8, characterized in that, The process of packaging the weak component set, safe operating limit, and filter cleaning early warning instruction into a cleaning early warning message, and then pushing it to the remote control platform via the industrial communication link, includes: querying a pre-established speed reduction level mapping table using the blockage level determination result and the safe operating limit level as a two-dimensional index, and retrieving the printing speed reduction ratio and paper feeding frequency reduction ratio that match the current index.

10. A method for remotely controlling the electrical operating status of a cardboard printing machine according to claim 9, characterized in that, After retrieving the printing speed reduction ratio and paper feeding frequency reduction ratio that match the current index, the process includes: issuing a graded speed reduction command to the main control circuit of the cardboard printing machine according to the printing speed reduction ratio and paper feeding frequency reduction ratio, so that the printing speed and paper feeding frequency are reduced to the corresponding levels and maintained; and after receiving the job completion receipt transmitted from the site, issuing a parameter recovery command to restore the printing speed and paper feeding frequency to the normal operating parameters recorded before the filter cleaning.

Images