Corrugated paper production quality monitoring method based on risk assessment map

Abstract

The invention discloses a corrugated paper production quality monitoring method based on a risk assessment map, and particularly relates to the field of corrugated paper production monitoring, comprising the following steps: collecting operation parameters of a plurality of sensing units deployed in gluing, pressing and drying links in a corrugated paper production line, the operation parameters comprise a pressure value, a thickness value, a temperature value, a humidity value and a bonding strength value, time alignment processing is carried out on the operation parameters through a unified time sequence, a technological parameter data set is formed, and atlas construction is carried out on the technological parameter data set. By constructing a primary mirror parameter group and executing direction consistency calculation, trend offset analysis and inter-node cooperative relationship judgment on an edge calculation side in a unified time window, a sensing collapse behavior of multiple sensing nodes under low-amplitude cooperative drift is identified, and cooperative offset characteristics are injected into a risk assessment map, so that a risk assessment result is obtained. And correcting the map structure and risk value distribution, and finally deducing an intervention parameter group to realize closed-loop adjustment control of the production process.

Description

Technical Field

[0001] This invention relates to the field of corrugated paper production monitoring technology, and more specifically, to a method for monitoring the quality of corrugated paper production based on risk assessment maps. Background Technology

[0002] In current corrugated paper manufacturing, quality monitoring generally relies on multiple sets of key sensors to collect parameters such as pressure, bonding strength, and thickness in real time. These sensors are often installed in space-constrained enclosed devices, lacking redundancy layout capabilities, resulting in a structurally dependent state where the smallest node supports the largest decision.

[0003] Risk assessment under normal operating conditions involves identifying process deviations and intervening early. However, existing technologies generally assume that sensor failures are single-point failures or abrupt drifts, thus relying on abnormal amplitude exceeding limits, abrupt time changes, or the prominence of errors at isolated nodes to trigger fault identification mechanisms.

[0004] However, in actual industrial environments, due to coupling micro-interferences such as mechanical aging, electrostatic accumulation, and temperature difference excitation, multiple sets of sensors may exhibit slight collaborative errors within the same time window, where the value range does not exceed the threshold but the overall direction shifts. These slight collaborative errors are more likely to manifest as multi-source drift behavior in real-time monitoring and rapid discrimination on the edge computing side. At the graph level, this type of multi-source drift behavior appears as an intact logical structure and a continuous and stable risk value, which is easily judged as a normal state. This causes the risk assessment to lose its basis in principle, mistakenly treating the systematic deviation as a local fluctuation and filtering it, ultimately forming a structural hidden danger. This phenomenon is essentially a kind of perception collapse under collaborative offset. It is highly concealed, conventional fault-tolerant masking mechanisms are completely ineffective, and it does not have a single threshold triggering condition, making it difficult to be covered by any existing coding detection or tolerance recognition mechanism.

[0005] Based on the current technology, the most critical technical problem in the quality monitoring of corrugated paper production is that when multiple sensing nodes encounter micro-disturbances at the same time and form a coordinated drift, even if data acquisition and time window alignment have been completed on the edge computing side, the system cannot identify the perception collapse at the map level in a timely manner based on the existing threshold mechanism, resulting in the risk assessment logic being continuously built on pseudo-data. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for monitoring the production quality of corrugated paper based on a risk assessment map. By constructing a master mirror parameter set and performing directional consistency calculation, trend offset analysis, and node collaboration relationship judgment on the edge computing side within a unified time window, the method identifies the perception collapse behavior of multiple sensing nodes under low-amplitude collaborative drift. The collaborative offset features are then injected into the risk assessment map, thereby correcting the map structure and risk value distribution. Finally, an intervention parameter set is derived to achieve closed-loop adjustment and control of the production process, thus solving the technical problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for quality monitoring of corrugated paper production based on risk assessment maps, comprising:

[0008] S1. Collect the operating parameters of multiple sensing units deployed in the sizing, pressing and drying stages of the corrugated paper production line. The operating parameters include pressure value, thickness value, temperature value, humidity value and bonding strength value. The operating parameters are time-aligned by a unified time sequence to form a process parameter dataset.

[0009] S2. Construct a graph from the process parameter dataset. Based on the rate of change of state of each sensing unit in a continuous time period and the correlation between parameters between nodes, construct an initial risk assessment graph. The nodes of the risk assessment graph represent process state points, and the edges of the risk assessment graph represent the evolution path between state points.

[0010] S3. Establish a corresponding digital mirror derivation unit for each sensing unit in the process parameter dataset. The digital mirror derivation unit generates a mirror parameter sequence based on the modeling results of its historical stable cycle operation trajectory and disturbance state, and combines the mirror parameter sequence with the currently measured operation parameters to form a master mirror parameter group.

[0011] S4. Perform directional consistency calculation, trend offset calculation and node collaboration relationship judgment on the master mirror parameter group within a fixed time window, and statistically generate cross-node collaborative offset feature data to identify data collaborative drift patterns in the map.

[0012] S5. Inject the collaborative offset feature data into the risk assessment map, and form a corrected map containing a drift compensation mechanism by constructing an offset path structure and adjusting the risk value and propagation path weight of the corresponding nodes.

[0013] S6. Based on the corrected graph, perform risk path analysis to solve for the intervention parameter set used for feedback control, and output the intervention parameter set to the control system of the corrugated paper production line to complete the process adjustment.

[0014] In a preferred embodiment, in S2, the step of constructing the risk assessment map includes:

[0015] S201. Calculate the rate of change of state of each sensing unit in the process parameter dataset over a continuous time period. The rate of change of state is obtained by dividing the difference in operating parameters at adjacent time points by the time interval.

[0016] S202. Based on the state change rate, the state of each sensing unit at different time points is represented as a process state point, and numbered according to the time sequence.

[0017] S203. Based on the operating parameters of each sensing unit in the process parameter dataset, extract the change vectors corresponding to its pressure value, thickness value, and bonding strength value within a unified time period to construct a parameter matrix; perform two-dimensional covariance calculation on the parameter matrix to obtain the degree of coordinated fluctuation between different physical dimension parameters of each sensing unit, which is used to characterize the parameter correlation between nodes; identify the correlation strength between nodes at the physical process layer based on the degree of coordinated fluctuation, and use the correlation strength as the condition for establishing graph edges;

[0018] S204. The process state points are used as nodes in the risk assessment graph. When the parameter correlation between any two process state points is greater than a preset threshold, a graph edge connection is established between the corresponding nodes to represent the evolution path between the state points.

[0019] S205. Combine the resulting graph node set and edge set to construct the initial risk assessment graph structure.

[0020] In a preferred embodiment, in S3, the first derivative and standard deviation joint analysis operation is performed on the operating parameter sequence of each sensing unit in the process parameter dataset within a set observation period. The time period in which the change of the derivative value is less than a preset threshold and the standard deviation is lower than a preset stability limit is selected and defined as the historical stable period. A set of stable state reference sequences is constructed using the operating parameters at each time point within the historical stable period as the input of the operating trajectory model.

[0021] Based on the running trajectory model, a disturbance response modeling operation is performed to construct three disturbance functions: a drift amplitude function generated based on historical abnormal data, a response hysteresis function generated based on historical sensing delay, and a periodic fluctuation function generated based on external disturbance events. The three disturbance functions are applied to the parameter values ​​at each time point in the steady-state reference sequence. The original parameters are transformed item by item through amplitude superposition, time axis offset, and periodic disturbance correction to generate a mirror parameter sequence with the same structure as the original sequence, which is used to express the theoretical response trajectory of the sensing unit under the complex disturbance situation.

[0022] The mirror parameter sequence is aligned with the measured operating parameters collected at the current time point by index matching, and a corresponding array structure is established with the parameter field order as the index to generate the main mirror parameter group.

[0023] In a preferred embodiment, the drift amplitude function takes the sequence of operating parameters of similar sensing units in a set abnormal segment in historical abnormal data as input, calculates the difference between the operating parameters at each time point in the current time period and the parameter values ​​at the same position in the historical stable cycle, performs confidence interval statistical operation on all differences, extracts the upper confidence limit as the set value of the current disturbance amplitude, and uses it as the amplitude offset value used for the disturbance superposition at the current time point.

[0024] The response hysteresis function constructs a timing lookup table with the control input signal and the corresponding output response signal of the sensing unit in multiple consecutive sampling periods, calculates the response delay time difference between each pair of signals, and uses the average delay amount in the maximum density interval as the time axis shift step of the mirror parameter at the current time point.

[0025] The periodic fluctuation function takes the time series of external disturbance events recorded during the production process as input. The time series of external disturbance events consists of the start time and duration of various identified disturbance events such as equipment vibration, temperature and humidity fluctuations, or material feeding intervals. The time series is subjected to Fourier spectrum decomposition to extract the dominant frequency component and phase angle, construct the corresponding sinusoidal disturbance function, and then the sinusoidal disturbance function is mapped on the time axis of the steady-state reference sequence to generate mirror parameter values ​​with periodic disturbance characteristics.

[0026] In a preferred embodiment, in S4, the orientation consistency calculation is performed by performing a first-order difference operation on adjacent time points of each corresponding parameter sequence in the master mirror parameter group to obtain the orientation sign sequence, and the proportion of the orientation signs of the master sequence and the mirror sequence being consistent within the same time window is counted as the orientation consistency score of that parameter dimension.

[0027] The trend offset is calculated based on the mean difference of each corresponding parameter in the master mirror parameter group within the same time window. The average deviation of each parameter dimension in the mirror sequence relative to the master sequence is calculated, and a trend offset vector is formed to measure the overall offset direction and offset intensity.

[0028] The determination of inter-node collaborative relationships is based on the trend offset vectors of multiple sensing units. After performing cosine similarity analysis with parameter dimension alignment, the trend collaborative strength between any two nodes is extracted. Node pairs with collaborative strength exceeding the relevant threshold are counted as cross-node collaborative offset feature units.

[0029] In a preferred embodiment, in S5, the collaborative migration feature data is indexed and matched with the node set in the risk assessment map to identify the map node corresponding to the collaborative migration feature, and an additional attribute field representing the perceived drift state is added to the corresponding node to form a node set with drift markers.

[0030] In the set of nodes with drift markers, a set of drift path structures is constructed based on the direction of the trend drift vector and the cooperative drift strength. The edges of the drift path are guided by the trend direction, and the connection weight of the edges is calculated with the cooperative strength value to represent the propagation path of drift in the graph structure.

[0031] The offset path structure is merged into the initial risk assessment map. For nodes with drift markers, risk value weighting correction is performed according to the cooperative offset magnitude and path connection density. The edge weights of the corresponding path edges are synchronously updated to generate a corrected map containing a drift compensation mechanism.

[0032] In a preferred embodiment, in S6, graph nodes with drift markers are identified in the modified graph and sorted according to the risk value of the graph nodes from high to low. Nodes with risk values ​​greater than a preset risk threshold are selected as starting nodes. A directed graph structure is constructed based on the edge weights in the modified graph. A set of risk propagation paths starting from the starting node and extending along the direction of decreasing edge weights is extracted to represent the dominant transmission path of perceived offset risk.

[0033] For each path in the risk propagation path, a joint weight calculation operation is performed, which accumulates the risk value of all graph nodes and the edge weight of all graph edges in the path, and constructs the corresponding joint weight index in the form of a weighted combination of the two.

[0034] In risk propagation paths where the joint weight index is greater than the pre-trigger threshold set by the intervention, the range of change of operating parameters corresponding to the starting node and the ending node of the path is extracted, and an intervention parameter group is constructed based on the range of change of operating parameters to limit the parameter adjustment boundary of feedback control.

[0035] The intervention parameter set is output to the control system of the corrugated paper production line, and feedback control commands are executed on the process links to which the corresponding nodes in the correction graph belong, according to the adjustment boundaries defined in the intervention parameter set.

[0036] The technical effects and advantages of this invention are as follows:

[0037] 1. This solution constructs a master mirror parameter set and performs judgments on directional consistency, trend shift, and collaborative relationship. It enables the identification of low-amplitude collaborative drift of multiple sensor nodes within the same time window on the edge computing side, solving the problem of the inability to identify perception collapse caused by sensor micro-perturbation collaborative shift in traditional risk assessment.

[0038] 2. The scheme adopts a risk assessment map based on the correlation between state change rate and parameters, with process state points as nodes and evolution paths as edges. It has a dual-dimensional mapping logic of time evolution and physical coupling, which improves the ability to express the causal relationship of process deviations in the structural hierarchy and explain the risk propagation mechanism.

[0039] 3. Based on the historical stable cycle, the running trajectory model is constructed and three types of disturbance functions are superimposed to generate a mirror parameter sequence. This simulates the theoretical response trajectory of the sensor under the compound interference scenario, which makes up for the modeling defects of scarce abnormal samples and unreproducible interference structure in the measured data, and improves the sensitivity and accuracy of offset identification.

[0040] 4. The solution constructs cosine similarity analysis aligned with the parameter dimensions by using trend offset vectors, thereby realizing the modeling and identification of collaborative relationships between different physical parameters. This breaks through the limitations of single-node anomaly detection, extends to the identification of collaborative behaviors across nodes, and enhances the perception coverage of complex quality hazards.

[0041] 5. After injecting collaborative offset features into the map, the risk value and path structure are corrected, and the dominant risk path is constructed based on the map edge weights. Then, the parameter boundary control group is derived, realizing a closed loop from anomaly identification, map correction to intervention control. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the overall framework of the present invention.

[0043] Figure 2 This is a flowchart of the process parameter acquisition and primary mirror parameter group generation process of the present invention.

[0044] Figure 3 This is a flowchart of the primary mirror parameter analysis and risk compensation control of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Refer to the instruction manual appendix Figure 1-3 An embodiment of the present invention provides a method for quality monitoring of corrugated paper production based on a risk assessment map, comprising:

[0047] S1. Collect the operating parameters of multiple sensing units deployed in the sizing, pressing and drying stages of the corrugated paper production line. The operating parameters include pressure value, thickness value, temperature value, humidity value and bonding strength value. The operating parameters are time-aligned by a unified time sequence to form a process parameter dataset.

[0048] S2. Construct a graph from the process parameter dataset. Based on the rate of change of state of each sensing unit in a continuous time period and the correlation between parameters between nodes, construct an initial risk assessment graph. The nodes of the risk assessment graph represent process state points, and the edges of the risk assessment graph represent the evolution path between state points.

[0049] S3. Establish a corresponding digital mirror derivation unit for each sensing unit in the process parameter dataset. The digital mirror derivation unit generates a mirror parameter sequence based on the modeling results of its historical stable cycle operation trajectory and disturbance state, and combines the mirror parameter sequence with the currently measured operation parameters to form a master mirror parameter group.

[0050] S4. Perform directional consistency calculation, trend offset calculation and node collaboration relationship judgment on the master mirror parameter group within a fixed time window, and statistically generate cross-node collaborative offset feature data to identify data collaborative drift patterns in the map.

[0051] S5. Inject the collaborative offset feature data into the risk assessment map, and form a corrected map containing a drift compensation mechanism by constructing an offset path structure and adjusting the risk value and propagation path weight of the corresponding nodes.

[0052] S6. Based on the corrected graph, perform risk path analysis to solve for the intervention parameter set used for feedback control, and output the intervention parameter set to the control system of the corrugated paper production line to complete the process adjustment.

[0053] In S2, the steps for constructing the risk assessment map include:

[0054] S201. Calculate the rate of change of state of each sensing unit in the process parameter dataset over a continuous time period. The rate of change of state is obtained by dividing the difference in operating parameters at adjacent time points by the time interval.

[0055] S202. Based on the state change rate, the state of each sensing unit at different time points is represented as a process state point, and numbered according to the time sequence.

[0056] S203. Based on the operating parameters of each sensing unit in the process parameter dataset, extract the change vectors corresponding to its pressure value, thickness value, and bonding strength value within a unified time period to construct a parameter matrix; perform two-dimensional covariance calculation on the parameter matrix to obtain the degree of coordinated fluctuation between different physical dimension parameters of each sensing unit, which is used to characterize the parameter correlation between nodes; identify the correlation strength between nodes at the physical process layer based on the degree of coordinated fluctuation, and use the correlation strength as the condition for establishing the graph edge; in addition, in S203, the reason why temperature and humidity values ​​are not introduced in the graph edge construction is that although temperature and humidity have an impact on corrugated paper production, they are environmental factors, and their changes are often indirect and lagging, unlike parameters such as pressure, thickness, and bonding strength, which can directly reflect the real-time changes in the structural state. If temperature and humidity are also used to calculate the correlation between parameters, it may mask the truly critical process connections, leading to inaccurate graph judgment. Therefore, only those parameters that can reflect the direct coupling relationship between process states are specifically selected here to ensure that the graph structure is clear, stable, and accurate.

[0057] S204. The process state points are used as nodes in the risk assessment graph. When the parameter correlation between any two process state points is greater than a preset threshold, a graph edge connection is established between the corresponding nodes to represent the evolution path between the state points.

[0058] S205. Combine the resulting graph node set and edge set to construct the initial risk assessment graph structure.

[0059] In S3, the first derivative and standard deviation joint analysis operation is performed on the operating parameter sequence of each sensing unit in the process parameter dataset within the set observation period. The time period in which the change of the derivative value is less than the preset threshold and the standard deviation is lower than the preset stability limit is selected and defined as the historical stable period. A set of stable state reference sequences is constructed using the operating parameters at each time point in the historical stable period as the input of the operating trajectory model.

[0060] Based on the running trajectory model, a disturbance response modeling operation is performed to construct three disturbance functions: a drift amplitude function generated based on historical abnormal data, a response hysteresis function generated based on historical sensing delay, and a periodic fluctuation function generated based on external disturbance events. The three disturbance functions are applied to the parameter values ​​at each time point in the steady-state reference sequence. The original parameters are transformed item by item through amplitude superposition, time axis offset, and periodic disturbance correction to generate a mirror parameter sequence with the same structure as the original sequence, which is used to express the theoretical response trajectory of the sensing unit under the complex disturbance situation.

[0061] The mirror parameter sequence is aligned with the measured operating parameters collected at the current time point by index matching, and a corresponding array structure is established with the parameter field order as the index to generate the master mirror parameter group, which is used for master mirror difference analysis in subsequent collaborative offset recognition calculation.

[0062] The drift amplitude function takes the sequence of operating parameters of similar sensing units in the set abnormal segment in the historical abnormal data as input, calculates the difference between the operating parameters at each time point in the current time period and the parameter values ​​at the same position in the historical stable cycle, performs confidence interval statistical operation on all differences, extracts the upper confidence limit as the set value of the current disturbance amplitude, and uses it as the amplitude offset value used for the disturbance superposition at the current time point.

[0063] The response hysteresis function constructs a timing lookup table using the control input signal and the corresponding output response signal of the sensing unit in multiple consecutive sampling periods, calculates the response delay time difference between each pair of signals, and uses the average delay amount in the maximum density interval as the time axis shift step size of the mirror parameter at the current time point.

[0064] The periodic fluctuation function takes the time series of external disturbance events recorded during the production process as input. The time series of external disturbance events consists of the start time and duration of various identified disturbance events such as equipment vibration, temperature and humidity fluctuations, or material feeding intervals. The time series is subjected to Fourier spectrum decomposition to extract the dominant frequency component and phase angle, construct the corresponding sinusoidal disturbance function, and then the sinusoidal disturbance function is mapped on the time axis of the steady-state reference sequence to generate mirror parameter values ​​with periodic disturbance characteristics.

[0065] In S4, the direction consistency calculation is performed by performing a first-order difference operation on adjacent time points of each corresponding parameter sequence in the master mirror parameter group to obtain the direction sign sequence, and the proportion of the direction signs of the master sequence and the mirror sequence being consistent within the same time window is counted as the direction consistency score of that parameter dimension.

[0066] The trend offset is calculated based on the mean difference of each corresponding parameter in the master mirror parameter group within the same time window. The average deviation of each parameter dimension in the mirror sequence relative to the master sequence is calculated, and a trend offset vector is formed to measure the overall offset direction and offset intensity.

[0067] The judgment of inter-node collaboration relationship is based on the trend offset vector of multiple sensing units. After performing cosine similarity analysis after parameter dimension alignment, the trend collaboration strength between any two nodes is extracted. Node pairs with collaboration strength exceeding the relevant threshold are counted as cross-node collaboration offset feature units.

[0068] S4 also includes the construction of a direction consistency score calculation model, which is used to evaluate whether the direction of change of each parameter dimension in the master mirror parameter array is consistent over a continuous time. By performing direction sign matching on the first difference between the master sequence and the mirror sequence within a given time window, the degree of direction consistency is extracted and quantified.

[0069]

[0070] in Indicates within the time window within, no. Directional consistency score across all parameter dimensions; Indexing the parameter dimensions ( (), representing different types of physical parameters, such as pressure, thickness, strength, etc.; For time index ( ), representing discrete sampling points within the current time window; This represents the total number of time points included in the time window. This represents the total number of adjacent time point pairs; Indicates the first position in the main sequence The parameters in time The possible values ​​of ; For the first in the mirror sequence The parameters in time The possible values ​​of ; This is a sign function; if the input value is positive, the output is +1; if the input value is negative, the output is -1; if the input value is zero, the output is 0. For consistency function, when The value is 1 if it is true, and 0 otherwise.

[0071] It also includes constructing a trend offset vector calculation model, which is used to quantify the average trend deviation of each parameter dimension in the master mirror parameter group. By averaging the point-by-point difference between the mirror sequence and the master sequence, a trend offset vector reflecting the direction and magnitude of the numerical offset is extracted.

[0072]

[0073] in Indicates within the time window within, no. Trend offset in each parameter dimension.

[0074] It also includes constructing a node-to-node collaborative offset strength calculation model. The node-to-node collaborative offset strength calculation model is used to identify the trend offset synergy between multiple sensing nodes. By performing normalized cosine similarity calculation on the trend offset vectors of any two nodes, the consistency of their offset directions is extracted as the strength of the collaborative relationship.

[0075]

[0076] in Indicates within the time window within, no. The node and the first Cooperative offset strength between nodes; Number the sensor node ( Each node corresponds to a set of parameter sequences; This represents the total number of parameter dimensions. For the first In the nth node, the th Each parameter dimension in the time window Trend offset within; Indicates the first In the nth node, the th Each parameter dimension in the time window Trend offset within; The two square roots in the denominator represent the nodes respectively. With nodes The Euclidean norm of the trend offset vector is used for normalization.

[0077] In S5, the collaborative migration feature data is indexed and matched with the node set in the risk assessment map to identify the map node corresponding to the collaborative migration feature, and additional attribute fields representing the perceived drift state are added to the corresponding node to form a node set with drift markers.

[0078] In the set of nodes with drift markers, a set of drift path structures is constructed based on the direction of the trend drift vector and the cooperative drift strength. The edges of the drift path are guided by the trend direction, and the connection weight of the edges is calculated with the cooperative strength value to represent the propagation path of drift in the graph structure.

[0079] The offset path structure is merged into the initial risk assessment map. For nodes with drift markers, risk value weighting correction is performed according to the cooperative offset magnitude and path connection density. The edge weights of the corresponding path edges are synchronously updated to generate a corrected map containing a drift compensation mechanism, which is used for subsequent risk path analysis and intervention parameter derivation.

[0080] In S6, the graph nodes with drift markers are identified in the modified graph and sorted from high to low according to the risk value of the graph nodes. The nodes with risk values ​​greater than the preset risk threshold are selected as the starting nodes. A directed graph structure is constructed based on the edge weights in the modified graph. The set of risk propagation paths starting from the starting node and extending along the direction of decreasing edge weights is extracted to represent the dominant transmission path of perceived offset risk.

[0081] For each path in the risk propagation path, a joint weight calculation operation is performed, accumulating the risk value of all graph nodes and the edge weight of all graph edges in the path, and constructing a corresponding joint weight index in the form of a weighted combination of the two, which is used to measure the risk intensity and structural sensitivity of the path.

[0082] In risk propagation paths where the joint weight index is greater than the pre-trigger threshold set by the intervention, the range of change of operating parameters corresponding to the starting node and the ending node of the path is extracted, and an intervention parameter group is constructed based on the range of change of operating parameters to limit the parameter adjustment boundary of feedback control.

[0083] The intervention parameter set is output to the control system of the corrugated paper production line, and feedback control commands are executed on the process links to which the corresponding nodes in the correction graph belong, based on the adjustment boundaries defined in the intervention parameter set, so as to realize closed-loop process adjustment based on the graph correction results.

[0084] It should be noted that this solution was developed based on the need for real-time data monitoring and identification of complex process state changes in multiple key stages of corrugated paper production. By uniformly aligning the operating parameters collected by multiple sensing units deployed in the sizing, pressing, and drying stages, a complete process parameter dataset was formed, laying a solid data foundation for subsequent full-process modeling. Furthermore, to address the problem of traditional methods failing to accurately represent the interrelationships between process states across nodes, this solution introduces a risk assessment graph as the core representation structure. By calculating the rate of state change of each sensing unit over a continuous time period and extracting the parameter correlations, a graph node composed of process state points and graph edges representing evolution paths are constructed. In the construction of the graph edges, only three types of parameters—pressure, thickness, and adhesive strength—are selected. The parameters directly reflect the coupling of structural processes, avoiding interference from indirect disturbances such as temperature and humidity on the associated structures, thus maintaining high structural accuracy and response sensitivity in the spectrum. To further enhance the ability to predict abnormal behavior, the scheme introduces a mirror derivation mechanism. Within the historical stable period of each sensing unit, a stable state reference sequence is extracted through dual constraints of derivative and standard deviation. This sequence is then used as input to construct an operational trajectory model. Three types of disturbance functions are then superimposed: a drift amplitude function based on historical abnormal data, a response hysteresis function based on sensing delay records, and a periodic fluctuation function based on external disturbance events. Through corresponding amplitude changes, time axis offsets, and periodic interference injections, a mirror parameter sequence with the same structure as the original parameter sequence is simulated and generated. This sequence is then aligned with the current measured parameters to form the master mirror parameter group.

[0085] The master mirror parameter set is used to perform three levels of feature calculations within a fixed time window: First, directional consistency calculation is used to identify the degree of consistency between the master sequence and the mirror sequence in the direction of change, thereby assessing the directional characteristics of the shift; second, trend shift calculation is used to quantify the overall deviation of the mirror parameters from the master parameters within the current window, reflecting the magnitude and trend of the shift; third, the inter-node collaborative relationship judgment is performed by analyzing the cosine similarity of the trend shift vectors between multiple sensing nodes to extract whether the trend change has cross-node collaborativeness; the above three processes together form cross-node collaborative shift feature data, which serves as an important input for subsequent map structure correction; in the map correction stage, the scheme injects the collaborative shift feature data into the map node set through index matching, marks the map nodes with drift, and constructs the shift path structure by combining the trend direction and collaborative strength. At the same time, weighted correction is performed on the node risk value and the path edge weight to form a corrected map with drift compensation capability, enabling the map structure to dynamically adapt to changes in interference;

[0086] In the process intervention stage, the solution identifies high-risk nodes with drift markers in the correction graph and constructs a directed graph structure guided by risk value sorting. It extracts the dominant risk propagation path along the direction of decreasing edge weight, and filters out the path set that meets the triggering conditions by weighting the joint risk of nodes and edges in the path. Then, it extracts the range of changes in operating parameters at the starting and ending points of the path to construct an intervention parameter group. This group of parameters is then used as the adjustment boundary of feedback control and sent to the production control system to achieve real-time closed-loop adjustment based on the graph structure and the deduction results.

[0087] The overall process not only achieves the identification and attribution of data anomalies, but also forms a process optimization mechanism from the perception layer to the execution layer. The reason why this solution adopts the dual mechanism of graph modeling and mirror derivation is based on the characteristics of strong nonlinear coupling between various parameters in the corrugated paper production process, hidden disturbance transmission, and delayed impact. Traditional threshold alarm or single-point monitoring methods are difficult to meet the real-time and accuracy requirements of quality monitoring. Therefore, every step of this solution, from mechanism construction and indicator extraction to risk reasoning and intervention control, is based on physical logic and parameter linkage, which has technical rationality and engineering feasibility.

[0088] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for quality monitoring in corrugated paper production based on risk assessment maps, characterized in that, include: S1. Collect the operating parameters of multiple sensing units deployed in the sizing, pressing and drying stages of the corrugated paper production line. The operating parameters include pressure value, thickness value, temperature value, humidity value and bonding strength value. The operating parameters are time-aligned by a unified time sequence to form a process parameter dataset. S2. Construct a graph from the process parameter dataset. Based on the rate of change of state of each sensing unit in a continuous time period and the correlation between parameters between nodes, construct an initial risk assessment graph. The nodes of the risk assessment graph represent process state points, and the edges of the risk assessment graph represent the evolution path between state points. S3. Establish a corresponding digital mirror derivation unit for each sensing unit in the process parameter dataset. The digital mirror derivation unit generates a mirror parameter sequence based on the modeling results of its historical stable cycle operation trajectory and disturbance state, and combines the mirror parameter sequence with the currently measured operation parameters to form a master mirror parameter group. S4. Perform directional consistency calculation, trend offset calculation and node collaboration relationship judgment on the master mirror parameter group within a fixed time window, and statistically generate cross-node collaborative offset feature data to identify data collaborative drift patterns in the map. S5. Inject the collaborative offset feature data into the risk assessment map, and form a corrected map containing a drift compensation mechanism by constructing an offset path structure and adjusting the risk value and propagation path weight of the corresponding nodes. S6. Based on the corrected graph, perform risk path analysis to solve for the intervention parameter set used for feedback control, and output the intervention parameter set to the control system of the corrugated paper production line to complete the process adjustment.

2. The method for quality monitoring of corrugated paper production based on risk assessment maps according to claim 1, characterized in that: In S2, the steps for constructing the risk assessment map include: S201. Calculate the rate of change of state of each sensing unit in the process parameter dataset over a continuous time period. The rate of change of state is obtained by dividing the difference in operating parameters at adjacent time points by the time interval. S202. Based on the state change rate, the state of each sensing unit at different time points is represented as a process state point, and numbered according to the time sequence. S203. Based on the operating parameters of each sensing unit in the process parameter dataset, extract the change vectors corresponding to its pressure value, thickness value, and bonding strength value within a unified time period to construct a parameter matrix; perform two-dimensional covariance calculation on the parameter matrix to obtain the degree of coordinated fluctuation between different physical dimension parameters of each sensing unit, which is used to characterize the parameter correlation between nodes; identify the correlation strength between nodes at the physical process layer based on the degree of coordinated fluctuation, and use the correlation strength as the condition for establishing graph edges; S204. The process state points are used as nodes in the risk assessment graph. When the parameter correlation between any two process state points is greater than a preset threshold, a graph edge connection is established between the corresponding nodes to represent the evolution path between the state points. S205. Combine the resulting graph node set and edge set to construct the initial risk assessment graph structure.

3. The method for quality monitoring of corrugated paper production based on risk assessment maps according to claim 2, characterized in that: In S3, the first derivative and standard deviation joint analysis operation is performed on the operating parameter sequence of each sensing unit in the process parameter dataset within the set observation period. The time period in which the change of the derivative value is less than the preset threshold and the standard deviation is lower than the preset stability limit is selected and defined as the historical stable period. A set of stable state reference sequences is constructed using the operating parameters at each time point in the historical stable period as the input of the operating trajectory model. Based on the running trajectory model, a disturbance response modeling operation is performed to construct three disturbance functions: a drift amplitude function generated based on historical abnormal data, a response hysteresis function generated based on historical sensing delay, and a periodic fluctuation function generated based on external disturbance events. The three disturbance functions are applied to the parameter values ​​at each time point in the steady-state reference sequence. The original parameters are transformed item by item through amplitude superposition, time axis offset, and periodic disturbance correction to generate a mirror parameter sequence with the same structure as the original sequence, which is used to express the theoretical response trajectory of the sensing unit under the complex disturbance situation. The mirror parameter sequence is aligned with the measured operating parameters collected at the current time point by index matching, and a corresponding array structure is established with the parameter field order as the index to generate the main mirror parameter group.

4. The method for quality monitoring of corrugated paper production based on risk assessment maps according to claim 3, characterized in that: The drift amplitude function takes the sequence of operating parameters of similar sensing units in the set abnormal segment in the historical abnormal data as input, calculates the difference between the operating parameters at each time point in the current time period and the parameter values ​​at the same position in the historical stable cycle, performs confidence interval statistical operation on all differences, extracts the upper confidence limit as the set value of the current disturbance amplitude, and uses it as the amplitude offset value used for the disturbance superposition at the current time point. The response hysteresis function constructs a timing lookup table with the control input signal and the corresponding output response signal of the sensing unit in multiple consecutive sampling periods, calculates the response delay time difference between each pair of signals, and uses the average delay amount in the maximum density interval as the time axis shift step of the mirror parameter at the current time point. The periodic fluctuation function takes the time series of external disturbance events recorded during the production process as input. The time series of external disturbance events consists of the start time and duration of various identified disturbance events such as equipment vibration, temperature and humidity fluctuations, or material feeding intervals. The time series is subjected to Fourier spectrum decomposition to extract the dominant frequency component and phase angle, construct the corresponding sinusoidal disturbance function, and then the sinusoidal disturbance function is mapped on the time axis of the steady-state reference sequence to generate mirror parameter values ​​with periodic disturbance characteristics.

5. The method for quality monitoring of corrugated paper production based on risk assessment maps according to claim 4, characterized in that: In S4, the orientation consistency calculation is performed by performing a first-order difference operation on adjacent time points of each corresponding parameter sequence in the master mirror parameter group to obtain the orientation sign sequence, and the proportion of the orientation signs of the master sequence and the mirror sequence being consistent within the same time window is counted as the orientation consistency score of that parameter dimension. The trend offset is calculated based on the mean difference of each corresponding parameter in the master mirror parameter group within the same time window. The average deviation of each parameter dimension in the mirror sequence relative to the master sequence is calculated, and a trend offset vector is formed to measure the overall offset direction and offset intensity. The determination of inter-node collaborative relationships is based on the trend offset vectors of multiple sensing units. After performing cosine similarity analysis with parameter dimension alignment, the trend collaborative strength between any two nodes is extracted. Node pairs with collaborative strength exceeding the relevant threshold are counted as cross-node collaborative offset feature units.

6. The method for quality monitoring of corrugated paper production based on risk assessment maps according to claim 5, characterized in that: In S5, the collaborative migration feature data is indexed and matched with the node set in the risk assessment map to identify the map node corresponding to the collaborative migration feature, and additional attribute fields representing the perceived drift state are added to the corresponding node to form a node set with drift markers. In the set of nodes with drift markers, a set of drift path structures is constructed based on the direction of the trend drift vector and the cooperative drift strength. The edges of the drift path are guided by the trend direction, and the connection weight of the edges is calculated with the cooperative strength value to represent the propagation path of drift in the graph structure. The offset path structure is merged into the initial risk assessment map. For nodes with drift markers, risk value weighting correction is performed according to the cooperative offset magnitude and path connection density. The edge weights of the corresponding path edges are synchronously updated to generate a corrected map containing a drift compensation mechanism.

7. The method for quality monitoring of corrugated paper production based on risk assessment maps according to claim 6, characterized in that: In S6, the graph nodes with drift markers are identified in the modified graph and sorted from high to low according to the risk value of the graph nodes. The nodes with risk values ​​greater than the preset risk threshold are selected as the starting nodes. A directed graph structure is constructed based on the edge weights in the modified graph. The set of risk propagation paths starting from the starting node and extending along the direction of decreasing edge weights is extracted to represent the dominant transmission path of perceived offset risk. For each path in the risk propagation path, a joint weight calculation operation is performed, which accumulates the risk value of all graph nodes and the edge weight of all graph edges in the path, and constructs the corresponding joint weight index in the form of a weighted combination of the two. In risk propagation paths where the joint weight index is greater than the pre-trigger threshold set by the intervention, the range of change of operating parameters corresponding to the starting node and the ending node of the path is extracted, and an intervention parameter group is constructed based on the range of change of operating parameters to limit the parameter adjustment boundary of feedback control. The intervention parameter set is output to the control system of the corrugated paper production line, and feedback control commands are executed on the process links to which the corresponding nodes in the correction graph belong, according to the adjustment boundaries defined in the intervention parameter set.

Images