A carton production line equipment maintenance and health management intelligent diagnosis method and system

Abstract

The application discloses a carton production line equipment maintenance and health management intelligent diagnosis method and system, relates to the technical field of equipment health management and intelligent diagnosis, and comprises the following steps: collecting equipment state characteristics and production process parameters, carrying out normalization processing on the basis of a change amount ratio to obtain load decoupling characteristics. Degeneration evolution rules of similar equipment are extracted and matched with the load decoupling characteristics, and when a matching deviation exceeds a threshold value, cross-equipment correction characteristics are obtained by correction. The time sequence correlation between the cross-equipment correction characteristics and the production process parameters is calculated, process disturbances are removed to obtain a degradation characteristic sequence. The degradation characteristic sequence is subjected to segmented analysis to identify a degradation acceleration interval, the evolution rate is calculated, the time scale coefficient of life prediction is adjusted, the remaining service life is predicted, and a maintenance operation is performed. The application improves the degradation evaluation accuracy and the life prediction precision, reduces the maintenance cost, and improves the equipment operation reliability and the production efficiency.

Description

Technical Field

[0001] This invention relates to the field of equipment health management and intelligent diagnostics, specifically to an intelligent diagnostic method and system for the maintenance and health management of cardboard box production line equipment. Background Technology

[0002] With the increasing level of industrial automation, carton production line equipment is becoming increasingly complex, including multiple subsystems such as printing, die-cutting, and gluing units. These subsystems experience performance degradation during long-term operation, leading to equipment failure and production interruptions. Traditional equipment maintenance methods mainly rely on periodic inspections or post-failure repairs, making it difficult to accurately predict the remaining service life of the equipment. This increases maintenance costs and may also cause unexpected downtime losses.

[0003] Existing equipment health management technologies monitor equipment status and diagnose faults by collecting equipment operating data, but they face numerous challenges in practical applications. The operating status of carton production line equipment is significantly affected by production process parameters. Variations in process parameters such as carton specifications, cardboard thickness, and production speed for different orders cause fluctuations in equipment load, resulting in a mixture of normal fluctuations caused by process changes and abnormal changes caused by equipment degradation in the equipment's status characteristics, making it difficult to accurately identify the true degradation state of the equipment. Traditional methods often attribute all status changes to equipment degradation, leading to misjudgments and over-maintenance.

[0004] Equipment may experience alternating slow and rapid degradation phases during use, with accelerated degradation often occurring just before failure. Ignoring this non-linear degradation characteristic can lead to significant errors in remaining service life predictions, failing to provide accurate time windows for maintenance decisions. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent diagnostic method and system for the maintenance and health management of carton production line equipment, aiming to solve at least one of the technical problems existing in the prior art.

[0006] The technical solution of this invention is: an intelligent diagnostic method for equipment maintenance and health management in a cardboard box production line, comprising the following steps: Collect the equipment status characteristics and production process parameters of the carton production line equipment during operation, calculate the ratio between the change in equipment status characteristics and the change in production process parameters, and use the ratio to normalize the equipment status characteristics to obtain load decoupling characteristics. Extract the degradation and evolution patterns of similar carton production line equipment, match the load decoupling characteristics with the degradation and evolution patterns, and when the matching deviation exceeds the preset matching degree threshold, use the degradation and evolution patterns to correct the load decoupling characteristics to obtain cross-equipment correction characteristics. Calculate the time series correlation between cross-equipment correction features and production process parameters, and separate and remove process disturbances from the cross-equipment correction features based on the time series correlation to obtain the degradation feature sequence; The degradation feature sequence is segmented and analyzed to identify the degradation acceleration interval. The evolution rate of the degradation acceleration interval is calculated. The time scale coefficient of the life prediction is adjusted according to the evolution rate. The remaining life of the carton production line equipment is predicted according to the time scale coefficient and maintenance operations are performed.

[0007] The ratio between the change in equipment state characteristics and the change in production process parameters is calculated. This ratio is then used to normalize the equipment state characteristics to obtain load decoupling characteristics, including: Establish a correspondence between equipment status characteristics and production process parameters. Calculate the load response component in the equipment status characteristics affected by the production process parameters based on the correspondence. Remove the load response component from the equipment status characteristics to obtain residual characteristics. Calculate the ratio between the change in residual characteristics and the change in production process parameters, average the ratio using a sliding window, and use the average ratio to normalize the residual characteristics to obtain normalized residual characteristics. Calculate the variance of the normalized residual feature over a continuous time period. When the variance exceeds a preset variance threshold, the corresponding time period is determined to be a transient period. The normalized residual feature corresponding to the transient period is removed, and the normalized residual feature corresponding to the time period with variance below the preset variance threshold is retained as the load decoupling feature.

[0008] The degradation and evolution patterns of similar cardboard box production line equipment are extracted. Load decoupling characteristics are matched with these patterns. When the matching deviation exceeds a preset matching threshold, the load decoupling characteristics are corrected using the degradation and evolution patterns to obtain cross-equipment corrected characteristics, including: The load decoupling characteristics of multiple similar cardboard box production line equipment under the same production process parameters are collected. The difference between the load decoupling characteristics at adjacent time points is calculated. Data points with positive differences are extracted and connected according to the running time to form a degradation evolution law. Calculate the fluctuation range of the degradation evolution law at each runtime position, and determine the allowable deviation range at each runtime position based on the fluctuation range; Obtain the load decoupling characteristics of the device to be matched at the current runtime, and calculate the absolute value of the difference between the load decoupling characteristics and the regular characteristic value of the degradation evolution law at the current runtime position as the matching deviation; When the matching deviation exceeds the allowable deviation range corresponding to the current runtime position, the load decoupling feature and the regular feature value are weighted and fused using the ratio of the matching deviation to the allowable deviation range to obtain the cross-device correction feature.

[0009] The time-series correlation between cross-equipment correction features and production process parameters is calculated. Based on the time-series correlation, process disturbances are separated and removed from the cross-equipment correction features to obtain the degradation feature sequence, including: The cross-correlation coefficients between cross-equipment correction features and production process parameters at multiple lag positions are calculated as time series correlation, and the lag duration corresponding to the peak position of the time series correlation is extracted. After shifting the production process parameters by time based on the lag time, a regression model is established with the cross-equipment correction features. The process disturbance component is calculated, and the residual features are obtained by subtracting the process disturbance component from the cross-equipment correction features. Spectral analysis is performed on the residual features to obtain the spectral energy distribution. The energy proportion of the frequency band corresponding to the frequency of change of production process parameters is extracted from the spectral energy distribution. When the energy proportion exceeds the preset energy threshold, band-stop filtering is performed on the residual features to remove the frequency band corresponding to the frequency of change of production process parameters, thus obtaining the filtered features. The filter features are subjected to piecewise monotonicity testing to identify non-monotonic data segments and to calculate the local fluctuation gradient of the non-monotonic data segments and the global degradation gradient of the filter features. Calculate the deviation between the local fluctuation gradient and the global degradation gradient, remove non-monotonic data segments with deviations exceeding a preset deviation threshold, and retain monotonically increasing data segments to form a degradation feature sequence.

[0010] After shifting the production process parameters by time based on the lag duration, a regression model is established with the cross-equipment correction features. The process disturbance component is calculated, and the residual features are obtained by subtracting the process disturbance component from the cross-equipment correction features. The production process parameters are shifted according to the lag time to obtain the shifted production process parameters. The shifted production process parameters are used as independent variables and cross-equipment correction features are used as dependent variables to construct a training dataset. The coefficient parameters of the regression model are determined by minimizing the sum of squared errors between the actual values ​​of the dependent variables and the output values ​​of the regression model. The translated production process parameters are input into a regression model with coefficient parameters to obtain the regression output value. The fitting residual between the cross-equipment correction feature and the regression output value is calculated, and the variance of the fitting residual is calculated. When the variance value exceeds the preset variance threshold, the interaction term and higher-order term of the shifted production process parameters are added to the regression model. The update coefficient parameters of the regression model are determined again by minimizing the sum of squared errors. The shifted production process parameters are then input into the regression model with update coefficient parameters to obtain the updated regression output value. The updated regression output value is used as the process disturbance component, and the process disturbance component is subtracted from the cross-equipment correction feature to obtain the residual feature.

[0011] Perform piecewise monotonicity testing on the filter features, identify non-monotonic data segments, and calculate the local fluctuation gradient of the non-monotonic data segments and the global degradation gradient of the filter features, including: Set the time length of the sliding window, use the sliding window to segment the filtering features to obtain multiple data segments, calculate the difference sequence between adjacent data points for each data segment, and use the proportion of the frequency of negative values ​​in the difference sequence to the total number of difference points as the monotonicity index. Extract the monotonicity index of data segments from similar cardboard box production line equipment within the same running time interval, calculate the statistical threshold of the monotonicity index of similar equipment, and mark the current data segment as a non-monotonic data segment when the monotonicity index exceeds the statistical threshold; For the marked non-monotonic data segment, the fluctuation amplitude of the data points in the non-monotonic data segment is extracted, the fluctuation coefficient of the production process parameters in the corresponding sliding window of the non-monotonic data segment is obtained, the fluctuation coefficient is used to normalize the fluctuation amplitude to obtain the corrected fluctuation amplitude, and the local fluctuation gradient is determined based on the corrected fluctuation amplitude and the time length of the sliding window. Extract the filter features at the start and end times, and determine the global degradation gradient based on the changes in the start and end times and the corresponding time span.

[0012] The degradation characteristic sequence is segmented and analyzed to identify accelerated degradation intervals. The evolution rate of each accelerated degradation interval is calculated. The time scale coefficient for lifetime prediction is adjusted based on the evolution rate. The remaining service life of the carton production line equipment is predicted based on the time scale coefficient, and maintenance operations are performed accordingly. The degradation feature sequence is segmented to obtain multiple degradation feature segments. The rate of change of degradation rate of adjacent degradation feature segments is calculated, and the rate of change distribution characteristics of similar carton production line equipment are extracted. When the rate of change deviates from the concentrated area of ​​the distribution characteristics, the corresponding degradation characteristic segment is identified as the degradation acceleration interval. The evolution rate of the degradation acceleration interval is calculated based on the initial degradation characteristic value, the final degradation characteristic value and the time span of the degradation acceleration interval. Obtain the evolution rate distribution of the historical degradation acceleration interval of the carton production line equipment, and locate the degradation stage type by analyzing the evolution rate of the degradation acceleration interval in the evolution rate distribution; Extract the degradation acceleration factor corresponding to the degradation stage type, amplify or reduce the evolution rate of the degradation acceleration interval based on the degradation acceleration factor to obtain the corrected evolution rate, and adjust the time scale coefficient of lifetime prediction according to the corrected evolution rate. The degradation feature sequence is scaled along the time axis according to the time scale coefficient to obtain the adjusted degradation feature sequence. The remaining service life of the carton production line equipment is predicted based on the adjusted degradation feature sequence. Extract the failure impact of each component and determine the maintenance time node based on the remaining service life, and control the carton production line equipment to perform maintenance operations according to the maintenance time node.

[0013] This invention provides an intelligent diagnostic system for equipment maintenance and health management in a cardboard box production line. The system includes: The data acquisition module is used to collect the equipment status characteristics and production process parameters of the carton production line equipment during operation, calculate the ratio between the change in equipment status characteristics and the change in production process parameters, and use the ratio to normalize the equipment status characteristics to obtain load decoupling characteristics. The cross-equipment correction module is used to extract the degradation and evolution patterns of similar carton production line equipment, match the load decoupling characteristics with the degradation and evolution patterns, and when the matching deviation exceeds the preset matching degree threshold, the load decoupling characteristics are corrected using the degradation and evolution patterns to obtain the cross-equipment correction characteristics. The process disturbance removal module is used to calculate the time series correlation between cross-equipment correction features and production process parameters, and to separate and remove process disturbances from the cross-equipment correction features based on the time series correlation to obtain the degradation feature sequence. The lifespan prediction module is used to perform segmented analysis on the degradation characteristic sequence, identify degradation acceleration intervals, calculate the evolution rate of the degradation acceleration intervals, adjust the time scale coefficient of lifespan prediction according to the evolution rate, predict the remaining lifespan of the carton production line equipment according to the time scale coefficient, and perform maintenance operations.

[0014] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0015] One technical solution provided in this embodiment of the invention is a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the steps in any of the aforementioned methods.

[0016] This invention eliminates the interference of production process parameter fluctuations on equipment status characteristics through load decoupling, accurately separating the true degradation signals of the equipment and avoiding misjudgments and over-maintenance caused by process changes. It utilizes the degradation evolution patterns of similar equipment for cross-equipment knowledge transfer, improving the accuracy of health assessments in the early stages of equipment operation and when data is insufficient. By eliminating process disturbances based on time series correlation, it effectively handles the time delay effect and complex nonlinear relationship between process parameter changes and equipment status. By identifying degradation acceleration intervals and dynamically adjusting the time scale coefficient, it accurately captures the nonlinear degradation characteristics of the equipment, significantly improving the accuracy of remaining service life prediction, providing a reliable time window for maintenance decisions, reducing maintenance costs, minimizing losses from sudden downtime, and improving equipment operational reliability and production efficiency. Attached Figure Description

[0017] Figure 1 A flowchart of an intelligent diagnostic method for equipment maintenance and health management in a carton production line, provided as an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an intelligent diagnostic system for equipment maintenance and health management of a carton production line according to an embodiment of the present invention. Detailed Implementation

[0018] like Figure 1 As shown, Figure 1 A flowchart of an intelligent diagnostic method for equipment maintenance and health management in a carton production line, provided by an embodiment of the present invention, is included in the following steps: Step 101: Collect the equipment status characteristics and production process parameters of the carton production line equipment during operation, calculate the ratio between the change in equipment status characteristics and the change in production process parameters, and use the ratio to normalize the equipment status characteristics to obtain the load decoupling characteristics.

[0019] In some embodiments of the present invention, step 101 may specifically include the following sub-steps: Sub-step 1011: Establish the correspondence between equipment status characteristics and production process parameters; calculate the load response component in the equipment status characteristics affected by the production process parameters based on the correspondence; and remove the load response component from the equipment status characteristics to obtain residual characteristics. Sub-step 1012: Calculate the ratio between the change in residual characteristics and the change in production process parameters, average the ratio using a sliding window to obtain the average ratio, and normalize the residual characteristics using the average ratio to obtain normalized residual characteristics. Sub-step 1013: Calculate the variance of the normalized residual feature in a continuous time period. When the variance exceeds a preset variance threshold, the corresponding time period is determined to be a transient period. The normalized residual feature corresponding to the transient period is removed, and the normalized residual feature corresponding to the time period with variance below the preset variance threshold is retained as the load decoupling feature.

[0020] The system collects equipment status characteristics and production process parameters during the operation of the cardboard box production line. By calculating the ratio between the changes in equipment status characteristics and the changes in production process parameters, the system normalizes the equipment status characteristics to obtain load decoupling characteristics. Equipment status characteristics include parameters such as the operating current, temperature, and vibration of the corrugated roller motor in the cardboard box production line. Production process parameters include production speed, paper weight, and pressure parameters. Based on historical operating data analysis, a mapping relationship between motor current and production speed, and a correspondence between vibration value and paper weight, are established. A linear regression method is used to fit the functional relationship between these two relationships to obtain the load response component. When the production speed increases from 120 m / min to 150 m / min, the motor current increases from 25 A to 28 A; the 3 A increase in current at this point is the load response component.

[0021] The residual characteristic is obtained by subtracting the calculated load response component from the actual measured motor current value. This residual characteristic primarily reflects changes in the equipment's own condition rather than changes caused by production load variations. Production load variations are usually normal fluctuations caused by adjustments to process parameters, while the residual characteristic reflects changes in the equipment's health status more closely. For example, when bearings wear, even under the same production process parameters, the motor current value will show slight changes; these changes can be effectively captured by the residual characteristic.

[0022] Data within a 5-minute time window is selected, and the ratio of the average change in residual characteristics to the average change in production process parameters within this window is calculated. If the average change in residual characteristics within 5 minutes is 0.5A, while the average change in production speed within the same time period is 10 m / min, then the ratio is 0.05. A sliding window method is used to calculate the ratio over a continuous time period, with a window length of 5 minutes and a sliding step of 1 minute, yielding a series of ratio data. These ratios reflect the sensitivity of equipment condition characteristics to changes in production process parameters.

[0023] The obtained series of ratios are averaged to obtain the average ratio. The average ratio can more stably reflect the relationship between equipment state characteristics and production process parameters, reducing the impact of random fluctuations. This average ratio is then used to normalize the residual characteristics by dividing the residual characteristic by (average ratio × corresponding production process parameter). After normalization, the fluctuations in the residual characteristics mainly reflect changes in the equipment's own state, effectively reducing the interference caused by fluctuations in production process parameters.

[0024] For the normalized residual characteristics, the variance of their fluctuation within a 10-minute time window is calculated to evaluate the stability of the characteristics. A preset fluctuation threshold of 0.15 is set. When the variance of fluctuation exceeds this threshold, the corresponding time period is determined to be a transient period. Transient periods typically represent equipment operating in a non-steady-state phase, such as during start-up and shutdown processes or periods of sudden load changes. During the operation of a carton production line, when changing to different specifications of paper, the equipment undergoes an adjustment and adaptation process. During this time, the variance of the normalized residual characteristics may reach 0.2, exceeding the preset threshold of 0.15.

[0025] Data identified as transient periods are removed from the normalized residual features, retaining only data from time periods with variance below a preset fluctuation threshold. Removing transient data improves the stability and reliability of the features, allowing the extracted features to better reflect changes in equipment health. After transient period removal, the remaining normalized residual features become the load decoupling features, primarily reflecting changes in the equipment's own health, effectively eliminating the impact of production load variations on the features.

[0026] After obtaining the load decoupling characteristics, their long-term trends are further analyzed. By comparing them with the normal operating baseline of the equipment, potential abnormal states can be identified. In the application of blade wear monitoring for cutting machines in carton production lines, the load decoupling characteristics show a slow upward trend as the degree of blade wear increases. By monitoring this trend, the degree of blade wear can be predicted in advance, and maintenance and replacement can be arranged before the blade completely fails, avoiding losses caused by production interruptions.

[0027] This invention effectively separates the impact of production load changes from equipment status changes through load decoupling feature extraction technology, improving the accuracy and reliability of equipment status monitoring. It can adapt to complex environments with frequent changes in production process parameters, reducing false alarm rates and improving the accuracy of fault early warning. By decoupling equipment status from production load, equipment health status assessment is no longer affected by fluctuations in production conditions, providing reliable technical support for predictive maintenance, significantly reducing unplanned downtime and maintenance costs, and improving the overall operating efficiency and equipment utilization of the production line.

[0028] Step 102: Extract the degradation and evolution patterns of similar carton production line equipment, match the load decoupling features with the degradation and evolution patterns, and when the matching deviation exceeds the preset matching degree threshold, use the degradation and evolution patterns to correct the load decoupling features to obtain cross-equipment correction features.

[0029] In some embodiments of the present invention, step 102 may specifically include the following sub-steps: Sub-step 1021: Collect load decoupling characteristics of multiple similar carton production line equipment under the same production process parameters, calculate the difference in load decoupling characteristics at adjacent time points, extract data points with positive differences, and connect them according to runtime to form a degradation evolution law; Sub-step 1022: Calculate the fluctuation range of the degradation evolution law at each runtime position, and determine the allowable deviation range at each runtime position based on the fluctuation range; Sub-step 1023: Obtain the load decoupling characteristics of the device to be matched at the current runtime, and calculate the absolute value of the difference between the load decoupling characteristics and the regular characteristic value of the degradation evolution law at the current runtime position as the matching deviation; Sub-step 1024: When the matching deviation exceeds the allowable deviation range corresponding to the current runtime position, the load decoupling feature and the regular feature value are weighted and fused using the ratio of the matching deviation to the allowable deviation range to obtain the cross-device correction feature.

[0030] The load decoupling characteristics of multiple similar cardboard production line machines were collected under the same production process parameters. For example, five identical corrugated cardboard production machines operating in the same workshop, at a production speed of 150 m / min and a paper weight of 180 g / m³, were analyzed. 2 Under certain conditions, the load decoupling characteristics of the equipment are collected. The difference between the load decoupling characteristics at adjacent time points is calculated, and data points with positive differences are extracted. These positive differences indicate the trend of equipment performance degradation over time.

[0031] Connecting these positive difference points based on equipment operating time forms a degradation evolution curve. This curve reflects the general pattern of gradual performance degradation of equipment over time and can serve as a benchmark for assessing the health status of other similar equipment. Data collected from multiple devices shows a common characteristic: the degradation rate of corrugated roller motors is relatively slow in the initial stage of operation (0-500 hours), relatively stable in the middle stage (500-1500 hours), and gradually accelerates in the later stage (above 1500 hours). This degradation evolution pattern can be used to predict the remaining service life of equipment and schedule preventative maintenance.

[0032] The fluctuation amplitude of the degradation evolution law at each operating time position was calculated, and the allowable deviation range at each operating time position was determined based on the fluctuation amplitude. The fluctuation amplitude was obtained by statistically analyzing the dispersion of load decoupling characteristics of multiple similar devices under the same operating time. For a corrugated roller motor with an operating time of 1000 hours, the load decoupling characteristic values ​​of 5 similar devices were statistically analyzed to be 3.2, 3.3, 3.1, 3.25, and 3.15, respectively, with a calculated standard deviation of 0.08. Twice the standard deviation was taken as the allowable deviation range, i.e., ±0.16. For a device with an operating time of 2000 hours, the standard deviation of the load decoupling characteristic may increase to 0.12, with a corresponding allowable deviation range of ±0.24, indicating that the longer the operating time of the device, the more obvious the individual differences become.

[0033] Obtain the load decoupling characteristic of the device to be matched at the current runtime, and calculate the absolute value of the difference between this load decoupling characteristic and the characteristic value of the degradation evolution law at the current runtime position as the matching deviation. For example, a carton cutting machine that has run for 1200 hours has a load decoupling characteristic value of 3.8, while according to the degradation evolution law, the characteristic value corresponding to this runtime is 3.5, so the matching deviation is 0.3. This matching deviation reflects the gap between the current state of the device to be matched and the typical degradation law, which may indicate abnormal degradation of the device or differences in the usage environment.

[0034] Assuming the allowable deviation range for the aforementioned carton cutting machine at a runtime of 1200 hours is ±0.2, while the actual matching deviation is 0.3, exceeding the allowable range, the ratio of the matching deviation to the upper limit of the allowable deviation is calculated to be 0.3 / 0.2 = 1.5. Based on this ratio, the load decoupling characteristics and regularity characteristics are weighted and fused. Specifically, the fusion method is to substitute the ratio 1.5 into the weight calculation formula, obtaining a weight of 0.4 for the load decoupling characteristics and a weight of 0.6 for the regularity characteristics. The cross-device correction characteristic is then calculated as 3.5 × 0.6 + 3.8 × 0.4 = 3.62.

[0035] By calculating cross-equipment correction characteristics, the individual differences between different devices and the influence of the operating environment are effectively eliminated, making the equipment condition assessment results more universal. In a carton production line, after 1500 hours of use, the load decoupling characteristics of a certain creasing machine showed that its degradation level was close to the warning line. However, by comparing the degradation evolution patterns with similar equipment, it was found that its degradation rate was significantly faster than the average level. After cross-equipment correction, it was confirmed that the equipment was indeed in an abnormal degradation state and required targeted maintenance. Maintenance personnel found that the creasing wheel bearing had insufficient lubrication. After timely replenishment of lubricating oil, the equipment performance returned to normal, avoiding potential production interruptions.

[0036] Using cross-equipment correction characteristics for equipment health assessment can effectively identify abnormal degradation trends and provide early warnings. For gluing machines in carton production lines, a health status scoring mechanism is established by regularly monitoring their cross-equipment correction characteristics. When the cross-equipment correction characteristics show an upward trend for three consecutive monitoring cycles and the rate of increase exceeds 50% of the normal degradation rate, an early warning is triggered, reminding maintenance personnel to pay attention to the equipment status.

[0037] This invention overcomes the problems of insufficient data for single devices and large individual differences in equipment by extracting the degradation and evolution patterns of similar equipment and performing cross-device feature correction, thus achieving more accurate equipment status assessment and fault early warning. It effectively eliminates the influence of individual equipment differences and changes in the operating environment, making health status assessment results more consistent and comparable. By establishing general patterns of equipment degradation, it provides a scientific basis for equipment maintenance decisions, allowing for more precise scheduling of preventative maintenance and avoiding resource waste caused by premature maintenance and equipment damage caused by delayed maintenance.

[0038] Step 103: Calculate the time series correlation between cross-equipment correction features and production process parameters, and separate and remove process disturbances from the cross-equipment correction features based on the time series correlation to obtain the degradation feature sequence.

[0039] In some embodiments of the present invention, step 103 may specifically include the following sub-steps: Sub-step 1031: Calculate the cross-correlation coefficient between cross-equipment correction features and production process parameters at multiple lag positions as time series correlation, and extract the lag duration corresponding to the peak position of time series correlation. Sub-step 1032: After shifting the production process parameters by time based on the lag time, establish a regression model with the cross-equipment correction features, calculate the process disturbance component, and subtract the process disturbance component from the cross-equipment correction features to obtain the residual features. Sub-step 1033: Perform spectral analysis on the residual features to obtain the spectral energy distribution, extract the energy proportion of the frequency band corresponding to the frequency of change of production process parameters from the spectral energy distribution, and when the energy proportion exceeds the preset energy threshold, perform band-stop filtering on the residual features to remove the frequency band corresponding to the frequency of change of production process parameters to obtain the filtered features. Sub-step 1034: Perform a piecewise monotonicity test on the filter features, identify non-monotonic data segments, and calculate the local fluctuation gradient of the non-monotonic data segments and the global degradation gradient of the filter features. Sub-step 1035: Calculate the deviation between the local fluctuation gradient and the global degradation gradient, remove non-monotonic data segments with deviation exceeding a preset deviation threshold, and retain monotonically increasing data segments to form a degradation feature sequence.

[0040] Cross-correlation coefficients between cross-equipment correction features and production process parameters at different time lag positions were calculated using Pearson correlation analysis, with a lag step size of 1 sampling period and a maximum lag duration of 100 sampling periods. Cross-correlation analysis was performed on the carton indentation depth feature sequence and the indentation wheel pressure parameter sequence. The cross-correlation coefficient reached its maximum value of 0.82 when the lag duration was 15 sampling periods; this lag duration was taken as the delay in the influence of the indentation wheel pressure parameter on the indentation depth feature. By traversing all lag positions, the optimal lag duration corresponding to each production process parameter was extracted, and a lag parameter mapping table was constructed.

[0041] Based on the extracted lag time, the production process parameters were time-shifted, shifting the indentation wheel pressure parameter sequence forward by 15 sampling periods to align it with the indentation depth feature sequence on the time axis. A linear regression model was established using the least squares method, with the time-shifted indentation wheel pressure parameters as independent variables and cross-equipment correction features as dependent variables. The regression model had a slope coefficient of 0.0034, an intercept of 2.1, and a coefficient of determination of 0.67. Based on the established regression model, the process disturbance component was calculated. The corresponding process disturbance component was subtracted from each data point in the cross-equipment correction feature sequence to obtain the residual feature sequence. The residual feature sequence reflects the true characteristic changes after excluding the influence of process fluctuations during equipment degradation.

[0042] A fast Fourier transform (FFT) is performed on the residual feature sequence to obtain its frequency domain representation. The spectral energy distribution characteristics are analyzed, the power spectral density of each frequency component is calculated, and the frequency range where energy is concentrated is identified. The frequency of change in the pressure parameter of the indentation wheel is 0.05 Hz, corresponding to a frequency band of 0.04 Hz to 0.06 Hz. The proportion of energy within this frequency band to the total spectral energy is statistically analyzed. When the energy proportion exceeds 20%, it indicates that the residual features still contain significant process parameter variation components. A Butterworth band-stop filter is designed with a center frequency of 0.05 Hz, a stopband width of 0.02 Hz, and a filter order of 4. The residual feature sequence is processed through the band-stop filter to filter out the frequency components corresponding to the process parameter variation frequency, obtaining the filtered feature sequence.

[0043] The filtered feature sequence is segmented according to a fixed time window, with each segment containing 50 consecutive sampling points. The difference sequence between adjacent data points within each segment is calculated, and the proportion of positive and negative values ​​in the difference sequence is statistically analyzed. When the proportion of positive values ​​exceeds 80% or the proportion of negative values ​​exceeds 80%, the data segment is determined to be a monotonic data segment. For non-monotonic data segments, the local fluctuation gradient is calculated, which is the difference between the maximum and minimum values ​​within the segment divided by the time span. The global degradation gradient of the entire filtered feature sequence is calculated, which is the difference between the end value and the beginning value of the sequence divided by the total time span. The global degradation gradient of the carton indentation depth feature is 0.002 mm / hour, indicating that the equipment as a whole exhibits a degradation trend.

[0044] The deviation between the local fluctuation gradient and the global degradation gradient of each non-monotonic data segment is calculated. The deviation is defined as the absolute value of the difference between the local fluctuation gradient and the global degradation gradient divided by the absolute value of the global degradation gradient. When the deviation exceeds a set deviation threshold of 3.0, the non-monotonic data segment is considered to contain abnormal fluctuations and does not conform to the monotonicity characteristics of equipment degradation. Non-monotonic data segments with deviations exceeding the threshold are removed, while all monotonically increasing data segments and non-monotonic data segments with small deviations are retained. The time series gaps caused by the removed data segments are filled using a time interpolation method to form a continuous degradation feature sequence. This degradation feature sequence effectively eliminates the interference of changes in production process parameters and abnormal fluctuations, and truly reflects the healthy degradation process of the carton production equipment.

[0045] This invention effectively separates equipment degradation signals from process disturbance signals, improving the accuracy and reliability of degradation feature extraction. It solves the problem of process parameter interference in traditional degradation feature extraction, providing high-quality feature input for equipment health status assessment and fault prediction. Time series correlation analysis accurately identifies the lag effect of process parameters, improving the accuracy of disturbance separation. Spectrum analysis and band-stop filtering further eliminate process interference components in the frequency domain, enhancing the purity of degradation features. Segmented monotonicity testing and deviation analysis effectively eliminate abnormal fluctuation data, ensuring the monotonicity and continuity of the degradation feature sequence.

[0046] Sub-step 1032, which involves time-shifting the production process parameters based on the lag time and establishing a regression model with the cross-equipment correction features, calculating the process disturbance component, and subtracting the process disturbance component from the cross-equipment correction features to obtain the residual features, also includes: The production process parameters are shifted according to the lag time to obtain the shifted production process parameters. The shifted production process parameters are used as independent variables and cross-equipment correction features are used as dependent variables to construct a training dataset. The coefficient parameters of the regression model are determined by minimizing the sum of squared errors between the actual values ​​of the dependent variables and the output values ​​of the regression model. The translated production process parameters are input into a regression model with coefficient parameters to obtain the regression output value. The fitting residual between the cross-equipment correction feature and the regression output value is calculated, and the variance of the fitting residual is calculated. When the variance value exceeds the preset variance threshold, the interaction term and higher-order term of the shifted production process parameters are added to the regression model. The update coefficient parameters of the regression model are determined again by minimizing the sum of squared errors. The shifted production process parameters are then input into the regression model with update coefficient parameters to obtain the updated regression output value. The updated regression output value is used as the process disturbance component, and the process disturbance component is subtracted from the cross-equipment correction feature to obtain the residual feature.

[0047] The indentation wheel pressure parameter sequence was time-shifted according to the lag time of the 15 sampling periods extracted above. The original indentation wheel pressure parameter sequence contained 1000 sampling points. After time shifting, the shifted production process parameter sequence was obtained. The first data point of this sequence corresponds to the 16th data point of the original sequence, the second data point corresponds to the 17th data point of the original sequence, and so on. The effective data length of the shifted sequence is 985 sampling points, which is consistent with the length of the cross-equipment correction feature sequence for the corresponding time period.

[0048] When constructing the training dataset, the translated production process parameters were used as independent variables, and the corresponding cross-equipment correction features were used as dependent variables. The training dataset contains 985 data pairs, each pair containing a translated indentation wheel pressure value and a corresponding indentation depth correction feature value. The numerical range of the translated indentation wheel pressure parameter is 12.5 kN to 15.8 kN, and the corresponding indentation depth correction feature value ranges from 2.1 mm to 2.8 mm. The least squares estimation method was used to determine the coefficient parameters of the linear regression, and the optimal coefficients were solved by minimizing the cumulative sum of squared errors between the actual value of the dependent variable and the regression output value.

[0049] Regression analysis yielded a slope coefficient of 0.0034 mm / kN and an intercept of 2.1 mm. The pressure value of each indentation wheel in the translated production process parameter sequence was input into the established regression relationship, and the corresponding regression output value was calculated. When the pressure of the indentation wheel after translation was 14.2 kN, the regression output value was 2.148 mm. The actual values ​​of the cross-equipment correction features were compared with the corresponding regression output values, and the fitted residual sequence was calculated. Each data point in the fitted residual sequence represents the deviation between the actual cross-equipment correction feature value and the regression prediction value.

[0050] The variance of the fitted residual sequence is calculated to be 0.025 mm. 2 This variance value reflects the accuracy of the linear regression fit. The preset variance threshold is 0.02 mm. 2 When the calculated variance value is 0.025mm 2 When this threshold is exceeded, it indicates that simple linear regression cannot adequately fit the complex relationship between cross-equipment modified features and the translated production process parameters. It is necessary to add interaction terms and higher-order terms for the translated production process parameters to the regression expression to improve the fitting accuracy.

[0051] To supplement the original linear term, quadratic and cubic terms of the pressure parameters of the embossing wheel after translation are added. The expanded regression expression contains four parameters: a constant term, a linear term, a quadratic term, and a cubic term. The least squares estimation method is reapplied, and the updated coefficient parameters are determined by minimizing the sum of squared errors. The updated coefficient parameters are as follows: constant term coefficient 1.95, linear term coefficient 0.0028, quadratic term coefficient 0.000015, and cubic term coefficient -0.0000008.

[0052] The translated production process parameter sequence is input into a multinomial regression relation with updated coefficient parameters to calculate the updated regression output value sequence. When the pressure of the indentation wheel after translation is 14.2 kN, the updated regression output value is 2.142 mm, which is closer to the actual cross-equipment corrected characteristic value of 2.140 mm than the linear regression output value of 2.148 mm. The calculated variance of the updated fitting residual is 0.018 mm. 2 The variance is 0.02 mm lower than the preset variance threshold. 2 This indicates that polynomial regression effectively improves the fitting accuracy.

[0053] The updated regression output sequence is used as the process disturbance component, reflecting the influence of changes in the indentation wheel pressure parameters on the cross-equipment correction feature. The corresponding process disturbance component is subtracted point-by-point from the cross-equipment correction feature sequence to obtain the residual feature sequence. Each data point in the residual feature sequence represents the difference between the cross-equipment correction feature value and the process disturbance component. The original cross-equipment correction feature value of 2.140 mm is subtracted from the process disturbance component of 2.142 mm to obtain a residual feature value of -0.002 mm.

[0054] The residual feature sequence effectively eliminates the interference of indentation wheel pressure parameter fluctuations on cross-equipment correction features, preserving characteristic information that truly reflects equipment health degradation. Multinomial regression fitting accurately captures the nonlinear relationship between process parameters and equipment characteristics, exhibiting higher disturbance separation accuracy compared to simple linear regression. The numerical fluctuation range of the residual feature sequence is significantly smaller than that of the original cross-equipment correction feature sequence, with the fluctuation amplitude decreasing from 0.7 mm to 0.1 mm, indicating effective removal of process disturbance components.

[0055] This invention achieves accurate modeling and effective separation of process disturbances through time lag compensation and nonlinear regression fitting. Multinomial regression captures the complex influence of process parameters on equipment characteristics, avoiding the limitations of linear assumptions. A variance threshold mechanism ensures that the regression fitting accuracy meets the requirements for disturbance separation, improving the quality of residual features. By subtracting process disturbance components point by point, the interference of production process parameter fluctuations on equipment health characteristics is effectively eliminated, providing pure feature signals for subsequent degradation feature extraction and health status assessment. It can adapt to the influence characteristics of different process parameters and has good versatility and scalability.

[0056] Sub-step 1034, which involves performing a piecewise monotonicity test on the filter features, identifying non-monotonic data segments, and calculating the local fluctuation gradient of the non-monotonic data segments and the global degradation gradient of the filter features, further includes: Set the time length of the sliding window, use the sliding window to segment the filtering features to obtain multiple data segments, calculate the difference sequence between adjacent data points for each data segment, and use the proportion of the frequency of negative values ​​in the difference sequence to the total number of difference points as the monotonicity index. Extract the monotonicity index of data segments from similar cardboard box production line equipment within the same running time interval, calculate the statistical threshold of the monotonicity index of similar equipment, and mark the current data segment as a non-monotonic data segment when the monotonicity index exceeds the statistical threshold; For the marked non-monotonic data segment, the fluctuation amplitude of the data points in the non-monotonic data segment is extracted, the fluctuation coefficient of the production process parameters in the corresponding sliding window of the non-monotonic data segment is obtained, the fluctuation coefficient is used to normalize the fluctuation amplitude to obtain the corrected fluctuation amplitude, and the local fluctuation gradient is determined based on the corrected fluctuation amplitude and the time length of the sliding window. Extract the filter features at the start and end times, and determine the global degradation gradient based on the changes in the start and end times and the corresponding time span.

[0057] The sliding window duration is set to 2 hours, corresponding to 120 consecutive sampling points, with a sampling interval of 1 minute. The sliding window starts from the beginning of the filtered feature sequence and moves forward one sampling point at a time until the end of the sequence is reached. The total length of the filtered feature sequence is 1000 sampling points. After segmentation by the sliding window, 881 data segments are obtained, each containing 120 consecutive filtered feature values.

[0058] Calculate the difference sequence between adjacent data points within each data segment. Taking the first data segment as an example, this segment contains 120 data points with filtered feature values ​​ranging from 2.142 mm to 2.158 mm. The difference sequence between adjacent data points contains 119 difference values, where the first difference value is the difference of 0.001 mm between the second data point value (2.143 mm) and the first data point value (2.142 mm). The difference sequence contains 73 positive values, 42 negative values, and 4 zero values. Negative values ​​occur 42 times, and with a total of 119 difference points, the frequency of negative values ​​accounts for 35.3%.

[0059] The proportion of negative values ​​in a statistical difference sequence relative to the total number of difference points is used as a monotonicity indicator for that data segment. A monotonicity indicator of 35.3% indicates that 35.3% of adjacent data point pairs within that segment exhibit a decreasing trend. Ideally, equipment degradation should exhibit a monotonically increasing characteristic; a smaller monotonicity indicator indicates better monotonicity of the data segment.

[0060] The monotonicity distribution of historical data segments for similar cardboard box production line equipment within the same operating time range was extracted. Filtered characteristic data of 10 creasing machines of the same model were collected over a period of 100 to 150 hours of operation. The same sliding window parameters were used for segmented processing, and the monotonicity index of each data segment was calculated. The average monotonicity index of similar equipment was 28.7%, and the standard deviation was 8.4%. The statistical threshold was determined by adding twice the standard deviation to the average, resulting in a statistical threshold of 45.5%. The monotonicity index of the current data segment (35.3%) is less than the statistical threshold of 45.5%, therefore this data segment is marked as a monotonic data segment.

[0061] Data segments with a monotonicity index exceeding the statistical threshold of 45.5% were identified and marked as non-monotonic data segments. The monotonicity index of the 327th data segment was 52.1%, exceeding the statistical threshold, and was therefore marked as a non-monotonic data segment. The time window corresponding to this non-monotonic data segment was the runtime from 327 min to 446 min, containing fluctuations in the indentation depth filtering feature from 2.234 mm to 2.198 mm.

[0062] Extract the fluctuation amplitude of data points within the non-monotonic data segment. Calculate the difference between the maximum value (2.251 mm) and the minimum value (2.189 mm) of the filtered feature within this data segment, obtaining a fluctuation amplitude of 0.062 mm. The fluctuation amplitude reflects the magnitude of change in the indentation depth feature within this time window; a larger value indicates more severe feature fluctuation.

[0063] The fluctuation coefficient of the production process parameters within the sliding window corresponding to the non-monotonic data segment is obtained. The standard deviation of the indentation wheel pressure parameter within this time window is 0.8 kN, and the average value is 14.3 kN. The fluctuation coefficient is calculated to be 5.6%. The fluctuation coefficient reflects the relative degree of change of the process parameters within this time period and is used to quantify the potential impact of process fluctuations on equipment characteristics.

[0064] The corrected fluctuation amplitude was obtained by normalizing the fluctuation amplitude using the fluctuation coefficient. The ratio of the fluctuation amplitude of 0.062 mm to the fluctuation coefficient of 5.6% was used as the corrected fluctuation amplitude, which was calculated to be 1.11 mm. The corrected fluctuation amplitude eliminates the influence of process parameter fluctuations on equipment characteristic changes and more accurately reflects the characteristic fluctuation characteristics of the equipment itself.

[0065] The local fluctuation gradient is determined based on the corrected fluctuation amplitude and the sliding window time length. The ratio of the corrected fluctuation amplitude of 1.11 mm to the sliding window time length of 2 h is 0.555 mm / h, which is the local fluctuation gradient of this non-monotonic data segment. The local fluctuation gradient characterizes the average rate of change of the equipment characteristics within this time period; a positive value indicates an overall upward trend, and a negative value indicates an overall downward trend.

[0066] The values ​​of the filtered feature sequence at the start and end times were extracted. The value at the start time was 2.142 mm, corresponding to a device runtime of 0 hours. The value at the end time was 2.387 mm, corresponding to a device runtime of 16.67 hours. The change in value between the start and end times was 0.245 mm, corresponding to a time span of 16.67 hours.

[0067] The global degradation gradient is determined based on the change in values ​​between the start and end times and the time span. The ratio of the change of 0.245 mm to the time span of 16.67 h is 0.0147 mm / h, which is the global degradation gradient of the filtered feature sequence. The global degradation gradient reflects the average trend of equipment characteristics throughout the entire monitoring period; a positive value indicates that the equipment as a whole is showing a degradation trend.

[0068] This invention effectively identifies anomalous fluctuation data segments in equipment feature sequences through sliding window segmentation and monotonicity testing. The introduction of statistical thresholds for similar equipment improves the accuracy and adaptability of non-monotonic data segment identification. Fluctuation coefficient normalization eliminates the interference of process parameter changes on feature fluctuations, enabling local fluctuation gradients to accurately reflect changes in equipment health status. The global degradation gradient provides a quantitative indicator of the overall degradation trend of the equipment, offering a benchmark reference for subsequent deviation calculations and data segment selection. This method can adapt to the changing characteristics of equipment features under different operating conditions, improving the robustness and reliability of degradation feature extraction.

[0069] Step 104: Perform segmented analysis on the degradation feature sequence, identify the degradation acceleration interval, calculate the evolution rate of the degradation acceleration interval, adjust the time scale coefficient of the life prediction according to the evolution rate, predict the remaining life of the carton production line equipment according to the time scale coefficient, and perform maintenance operations.

[0070] In some embodiments of the present invention, step 104 may specifically include the following sub-steps: Sub-step 1041 involves performing segmented analysis on the degradation feature sequence to obtain multiple degradation feature segments, calculating the rate of change of degradation rate of adjacent degradation feature segments, and extracting the rate of change distribution characteristics of similar carton production line equipment. Sub-step 1042: When the rate of change deviates from the concentrated area of ​​the distribution characteristics, the corresponding degradation characteristic segment is identified as the degradation acceleration interval. The evolution rate of the degradation acceleration interval is calculated based on the initial degradation characteristic value, the final degradation characteristic value and the time span of the degradation acceleration interval. Sub-step 1043: Obtain the evolution rate distribution of the historical degradation acceleration interval of the carton production line equipment, and locate the degradation stage type in the evolution rate distribution of the degradation acceleration interval. Sub-step 1044: Extract the degradation acceleration factor corresponding to the degradation stage type, amplify or reduce the evolution rate of the degradation acceleration interval based on the degradation acceleration factor to obtain the corrected evolution rate, and adjust the time scale coefficient of lifetime prediction according to the corrected evolution rate. Sub-step 1045: Scaling the degradation feature sequence along the time axis according to the time scale coefficient to obtain the adjusted degradation feature sequence; and predicting the remaining service life of the carton production line equipment based on the adjusted degradation feature sequence. Sub-step 1046: Extract the failure impact of each component and determine the maintenance time node based on the remaining service life, and control the carton production line equipment to perform maintenance operations according to the maintenance time node.

[0071] The degradation feature sequence was segmented at fixed time intervals, with each segment containing feature data from a continuous 48-hour operation. The total length of the degradation feature sequence covered 1200 hours of continuous equipment operation, resulting in 25 degradation feature segments. Each segment contained 2880 sampling points at a 1-minute interval. The initial indentation depth feature value of the first degradation feature segment was 2.142 mm, the final feature value was 2.158 mm, and the corresponding degradation rate was 0.333 μm / h.

[0072] The rate of change of degradation rate between adjacent degradation feature segments was calculated to identify inflection points in the degradation trend. The degradation rate of the second degradation feature segment was 0.425 μm / h, a change of 27.6% compared to the degradation rate of 0.333 μm / h in the first segment. The degradation rate of the third degradation feature segment was 0.392 μm / h, a change of -7.8% compared to the second segment. Positive change rates indicate an accelerating degradation rate, while negative change rates indicate a slowing degradation rate.

[0073] The variation rate distribution characteristics of similar cardboard box production line equipment were extracted to establish a statistical benchmark for degradation rate changes. Historical degradation data of 12 identical creasing machines under similar operating conditions were collected, and the variation rate sequence of adjacent segments of each machine was calculated. The mean of the variation rate distribution was 8.3%, the standard deviation was 15.2%, and the concentrated region of the distribution was defined as the range of plus or minus one standard deviation of the mean, i.e., -6.9% to 23.5%.

[0074] When the rate of change of the eighth degradation feature segment reached 45.7%, this value significantly deviated from the upper limit of the concentrated distribution feature region by 23.5%, and the corresponding degradation feature segment was identified as the beginning of the degradation acceleration interval. The degradation acceleration interval started from the eighth segment and lasted until the end of the twelfth segment, with a total duration of 192 hours. The initial degradation feature value of the degradation acceleration interval was 2.267 mm, the final degradation feature value was 2.412 mm, and the feature value increment was 0.145 mm.

[0075] The evolution rate was calculated based on the initial and final degradation characteristic values ​​and the time span of the accelerated degradation zone. Within the accelerated degradation zone, the indentation depth increased from 2.267 mm to 2.412 mm over a time span of 192 hours, with an evolution rate of 0.755 μm / h. This evolution rate is significantly higher than the average degradation rate of 0.387 μm / h in the previous segments, indicating that the equipment has entered a rapid degradation phase.

[0076] The evolution rate distribution of the accelerated degradation range in historical data was analyzed, and three degradation stages were defined: the initial degradation stage corresponds to an evolution rate of 0.2 μm / h to 0.6 μm / h, the intermediate degradation stage corresponds to an evolution rate of 0.6 μm / h to 1.2 μm / h, and the late degradation stage corresponds to an evolution rate of more than 1.2 μm / h. The current device's accelerated degradation range evolution rate of 0.755 μm / h falls within the intermediate degradation stage range.

[0077] A degradation acceleration factor of 1.35 was extracted for the intermediate degradation stage type. This factor reflects the degree of acceleration of the intermediate degradation stage relative to the normal degradation process. Multiplying the evolution rate of the accelerated degradation range (0.755 μm / h) by the degradation acceleration factor 1.35 yields a corrected evolution rate of 1.019 μm / h. The corrected evolution rate takes into account the impact of degradation stage characteristics on future degradation development, thus improving the accuracy of lifetime prediction.

[0078] The timescale factor for lifetime prediction is adjusted based on the modified evolution rate. The baseline timescale factor is 1.0, corresponding to the lifetime prediction time axis under normal degradation rate conditions. The ratio of the modified evolution rate of 1.019 μm / h to the baseline degradation rate of 0.387 μm / h is 2.63, and the adjusted timescale factor is 0.38, indicating that the time axis needs to be compressed to 38% of the original.

[0079] The degraded feature sequence is obtained by scaling the time axis of the original degraded feature sequence using a time scaling factor of 0.38. The time coordinates of the original degraded feature sequence are compressed according to the time scaling factor, so the data point at time 1200 corresponds to time 456 in the adjusted sequence. The adjusted degraded feature sequence maintains the same trend of feature value changes, but the time process is compressed to reflect the effect of accelerated degradation.

[0080] Based on the adjusted degradation feature sequence, the remaining service life of the carton production line equipment is predicted. A failure threshold of 3.2 mm is set for the indentation depth feature; when the feature value reaches this threshold, the equipment needs to be shut down for maintenance. In the adjusted degradation feature sequence, a linear extrapolation method is used to predict the time required for the feature value to increase from the current 2.412 mm to the failure threshold of 3.2 mm. The predicted remaining service life is 208 hours, significantly shorter than the 546 hours predicted without adjusting the time scale.

[0081] The failure impact of each component was extracted and combined with its remaining service life to determine the maintenance time node. The failure impact of the indentation wheel component was 0.85, indicating that the failure of this component would have a significant impact on the overall equipment function. The failure impact of the cutting blade component was 0.72, and the failure impact of the folding mechanism component was 0.63. Based on the failure impact, the remaining service life was weighted and adjusted. The effective remaining service life of the indentation wheel component was 177 hours, and the corresponding maintenance time node was set to perform preventive maintenance after 168 hours of operation.

[0082] The carton production line equipment is controlled to perform maintenance operations according to the maintenance time node. When the equipment operation time reaches the predetermined maintenance time node of 168 hours, the maintenance operation sequence is automatically triggered. The maintenance operations include surface inspection of the indentation wheel, measurement of cutting edge wear, lubrication oil replacement, and adjustment of the transmission mechanism. After the maintenance operations are completed, the equipment parameters are recalibrated, the equipment health status baseline value is updated, and the starting reference point of the degradation characteristic sequence is reset.

[0083] This invention achieves accurate identification and quantitative modeling of the accelerated degradation process of equipment through multi-level degradation feature analysis and dynamic timescale adjustment. Automatic identification of degradation acceleration intervals avoids the limitations of fixed degradation patterns in traditional methods and can adapt to changes in degradation characteristics under different equipment and operating conditions. Phased modeling and correction of evolution rates improves the accuracy and reliability of lifetime prediction during the accelerated degradation phase. The dynamic adjustment mechanism of the timescale coefficients enables lifetime prediction to reflect the current degradation state and development trend of the equipment in real time. The failure impact-weighted maintenance scheduling strategy optimizes maintenance resource allocation and reduces the risk of sudden equipment failures.

[0084] like Figure 2 As shown, Figure 2 This invention provides a schematic diagram of an intelligent diagnostic system for equipment maintenance and health management in a cardboard box production line, comprising: Data acquisition module 201 is used to collect equipment status characteristics and production process parameters during the operation of the carton production line equipment, calculate the ratio between the change in equipment status characteristics and the change in production process parameters, and use the ratio to normalize the equipment status characteristics to obtain load decoupling characteristics. The cross-equipment correction module 202 is used to extract the degradation and evolution patterns of similar carton production line equipment, match the load decoupling features with the degradation and evolution patterns, and when the matching deviation exceeds the preset matching degree threshold, the load decoupling features are corrected using the degradation and evolution patterns to obtain the cross-equipment correction features. The process disturbance removal module 203 is used to calculate the time series correlation between cross-equipment correction features and production process parameters, and to separate and remove process disturbances from the cross-equipment correction features based on the time series correlation to obtain a degradation feature sequence. The life prediction module 204 is used to perform segmented analysis on the degradation characteristic sequence, identify the degradation acceleration interval, calculate the evolution rate of the degradation acceleration interval, adjust the time scale coefficient of life prediction according to the evolution rate, predict the remaining service life of the carton production line equipment according to the time scale coefficient, and perform maintenance operations.

[0085] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0086] One technical solution provided in this embodiment of the invention is a computer-readable storage medium storing a computer program, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0087] The specific embodiments described above are preferred embodiments of the present invention and are not intended to limit the specific scope of the present invention. The scope of the present invention includes, but is not limited to, these specific embodiments. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A method for intelligent diagnosis of equipment maintenance and health management in a cardboard box production line, characterized in that, Includes the following steps: Collect the equipment status characteristics and production process parameters of the carton production line equipment during operation, calculate the ratio between the change in equipment status characteristics and the change in production process parameters, and use the ratio to normalize the equipment status characteristics to obtain load decoupling characteristics. Extract the degradation and evolution patterns of similar carton production line equipment, match the load decoupling characteristics with the degradation and evolution patterns, and when the matching deviation exceeds the preset matching degree threshold, use the degradation and evolution patterns to correct the load decoupling characteristics to obtain cross-equipment correction characteristics. Calculate the time series correlation between cross-equipment correction features and production process parameters, and separate and remove process disturbances from the cross-equipment correction features based on the time series correlation to obtain the degradation feature sequence; The degradation feature sequence is segmented and analyzed to identify the degradation acceleration interval. The evolution rate of the degradation acceleration interval is calculated. The time scale coefficient of the life prediction is adjusted according to the evolution rate. The remaining life of the carton production line equipment is predicted according to the time scale coefficient and maintenance operations are performed.

2. The method according to claim 1, characterized in that, The ratio between the change in equipment state characteristics and the change in production process parameters is calculated. This ratio is then used to normalize the equipment state characteristics to obtain load decoupling characteristics, including: Establish a correspondence between equipment status characteristics and production process parameters. Calculate the load response component in the equipment status characteristics affected by the production process parameters based on the correspondence. Remove the load response component from the equipment status characteristics to obtain residual characteristics. Calculate the ratio between the change in residual characteristics and the change in production process parameters, average the ratio using a sliding window, and use the average ratio to normalize the residual characteristics to obtain normalized residual characteristics. Calculate the variance of the normalized residual feature over a continuous time period. When the variance exceeds a preset variance threshold, the corresponding time period is determined to be a transient period. The normalized residual feature corresponding to the transient period is removed, and the normalized residual feature corresponding to the time period with variance below the preset variance threshold is retained as the load decoupling feature.

3. The method according to claim 1, characterized in that, The degradation and evolution patterns of similar cardboard box production line equipment are extracted. Load decoupling characteristics are matched with these patterns. When the matching deviation exceeds a preset matching threshold, the load decoupling characteristics are corrected using the degradation and evolution patterns to obtain cross-equipment corrected characteristics, including: The load decoupling characteristics of multiple similar cardboard box production line equipment under the same production process parameters are collected. The difference between the load decoupling characteristics at adjacent time points is calculated. Data points with positive differences are extracted and connected according to the running time to form a degradation evolution law. Calculate the fluctuation range of the degradation evolution law at each runtime position, and determine the allowable deviation range at each runtime position based on the fluctuation range; Obtain the load decoupling characteristics of the device to be matched at the current runtime, and calculate the absolute value of the difference between the load decoupling characteristics and the regular characteristic value of the degradation evolution law at the current runtime position as the matching deviation; When the matching deviation exceeds the allowable deviation range corresponding to the current runtime position, the load decoupling feature and the regular feature value are weighted and fused using the ratio of the matching deviation to the allowable deviation range to obtain the cross-device correction feature.

4. The method according to claim 1, characterized in that, The time-series correlation between cross-equipment correction features and production process parameters is calculated. Based on the time-series correlation, process disturbances are separated and removed from the cross-equipment correction features to obtain the degradation feature sequence, including: The cross-correlation coefficients between cross-equipment correction features and production process parameters at multiple lag positions are calculated as time series correlation, and the lag durations corresponding to the peak positions of time series correlation are extracted. After shifting the production process parameters by time based on the lag time, a regression model is established with the cross-equipment correction features. The process disturbance component is calculated, and the residual features are obtained by subtracting the process disturbance component from the cross-equipment correction features. Spectral analysis is performed on the residual features to obtain the spectral energy distribution. The energy proportion of the frequency band corresponding to the frequency of change of production process parameters is extracted from the spectral energy distribution. When the energy proportion exceeds the preset energy threshold, band-stop filtering is performed on the residual features to remove the frequency band corresponding to the frequency of change of production process parameters, thus obtaining the filtered features. The filter features are subjected to piecewise monotonicity testing to identify non-monotonic data segments and to calculate the local fluctuation gradient of the non-monotonic data segments and the global degradation gradient of the filter features. Calculate the deviation between the local fluctuation gradient and the global degradation gradient, remove non-monotonic data segments with deviations exceeding a preset deviation threshold, and retain monotonically increasing data segments to form a degradation feature sequence.

5. The method according to claim 4, characterized in that, After shifting the production process parameters by time based on the lag duration, a regression model is established with the cross-equipment correction features. The process disturbance component is calculated, and the residual features are obtained by subtracting the process disturbance component from the cross-equipment correction features. The production process parameters are shifted according to the lag time to obtain the shifted production process parameters. The shifted production process parameters are used as independent variables and cross-equipment correction features are used as dependent variables to construct a training dataset. The coefficient parameters of the regression model are determined by minimizing the sum of squared errors between the actual values ​​of the dependent variables and the output values ​​of the regression model. The translated production process parameters are input into a regression model with coefficient parameters to obtain the regression output value. The fitting residual between the cross-equipment correction feature and the regression output value is calculated, and the variance of the fitting residual is calculated. When the variance value exceeds the preset variance threshold, the interaction term and higher-order term of the shifted production process parameters are added to the regression model. The update coefficient parameters of the regression model are determined again by minimizing the sum of squared errors. The shifted production process parameters are then input into the regression model with update coefficient parameters to obtain the updated regression output value. The updated regression output value is used as the process disturbance component, and the process disturbance component is subtracted from the cross-equipment correction feature to obtain the residual feature.

6. The method according to claim 4, characterized in that, Perform piecewise monotonicity testing on the filter features, identify non-monotonic data segments, and calculate the local fluctuation gradient of the non-monotonic data segments and the global degradation gradient of the filter features, including: Set the time length of the sliding window, use the sliding window to segment the filtering features to obtain multiple data segments, calculate the difference sequence between adjacent data points for each data segment, and use the proportion of the frequency of negative values ​​in the difference sequence to the total number of difference points as the monotonicity index. Extract the monotonicity index of data segments from similar cardboard box production line equipment within the same running time interval, calculate the statistical threshold of the monotonicity index of similar equipment, and mark the current data segment as a non-monotonic data segment when the monotonicity index exceeds the statistical threshold; For the marked non-monotonic data segment, the fluctuation amplitude of the data points in the non-monotonic data segment is extracted, the fluctuation coefficient of the production process parameters in the corresponding sliding window of the non-monotonic data segment is obtained, the fluctuation coefficient is used to normalize the fluctuation amplitude to obtain the corrected fluctuation amplitude, and the local fluctuation gradient is determined based on the corrected fluctuation amplitude and the time length of the sliding window. Extract the filter features at the start and end times, and determine the global degradation gradient based on the changes in the start and end times and the corresponding time span.

7. The method according to claim 1, characterized in that, The degradation characteristic sequence is segmented and analyzed to identify accelerated degradation intervals. The evolution rate of each accelerated degradation interval is calculated. The time scale coefficient for lifetime prediction is adjusted based on the evolution rate. The remaining service life of the carton production line equipment is predicted based on the time scale coefficient, and maintenance operations are performed accordingly. The degradation feature sequence is segmented to obtain multiple degradation feature segments. The rate of change of degradation rate of adjacent degradation feature segments is calculated, and the rate of change distribution characteristics of similar carton production line equipment are extracted. When the rate of change deviates from the concentrated area of ​​the distribution characteristics, the corresponding degradation characteristic segment is identified as the degradation acceleration interval. The evolution rate of the degradation acceleration interval is calculated based on the initial degradation characteristic value, the final degradation characteristic value and the time span of the degradation acceleration interval. Obtain the evolution rate distribution of the historical degradation acceleration interval of the carton production line equipment, and locate the degradation stage type by analyzing the evolution rate of the degradation acceleration interval in the evolution rate distribution; Extract the degradation acceleration factor corresponding to the degradation stage type, amplify or reduce the evolution rate of the degradation acceleration interval based on the degradation acceleration factor to obtain the corrected evolution rate, and adjust the time scale coefficient of lifetime prediction according to the corrected evolution rate. The degradation feature sequence is scaled along the time axis according to the time scale coefficient to obtain the adjusted degradation feature sequence. The remaining service life of the carton production line equipment is predicted based on the adjusted degradation feature sequence. Extract the failure impact of each component and determine the maintenance time node based on the remaining service life, and control the carton production line equipment to perform maintenance operations according to the maintenance time node.

8. An intelligent diagnostic system for equipment maintenance and health management of a cardboard box production line, used to implement the method described in any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to collect the equipment status characteristics and production process parameters of the carton production line equipment during operation, calculate the ratio between the change in equipment status characteristics and the change in production process parameters, and use the ratio to normalize the equipment status characteristics to obtain load decoupling characteristics. The cross-equipment correction module is used to extract the degradation and evolution patterns of similar carton production line equipment, match the load decoupling characteristics with the degradation and evolution patterns, and when the matching deviation exceeds the preset matching degree threshold, the load decoupling characteristics are corrected using the degradation and evolution patterns to obtain the cross-equipment correction characteristics. The process disturbance removal module is used to calculate the time series correlation between cross-equipment correction features and production process parameters, and to separate and remove process disturbances from the cross-equipment correction features based on the time series correlation to obtain the degradation feature sequence. The lifespan prediction module is used to perform segmented analysis on the degradation characteristic sequence, identify degradation acceleration intervals, calculate the evolution rate of the degradation acceleration intervals, adjust the time scale coefficient of lifespan prediction according to the evolution rate, predict the remaining lifespan of the carton production line equipment according to the time scale coefficient, and perform maintenance operations.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 7.

Images