A method for processing digital factory monitoring data

Abstract

This invention provides a digital factory monitoring data processing method, which relates to the field of industrial process monitoring technology. The method acquires monitoring data from factory monitoring points, establishes a topological association diagram, and calculates the data transmission characteristics between monitoring points. This method then establishes a three-layer data transmission matrix and calculates the anomaly propagation credibility, which is used to generate a multi-level anomaly propagation topology diagram, ultimately outputting a comprehensive evaluation and analysis. This invention significantly improves the comprehensiveness and accuracy of anomaly propagation feature extraction, enhances the reliability of propagation path identification, and enables the timely detection of potential risks through an early warning mechanism. This method provides more effective technical support for anomaly propagation analysis in complex industrial systems, and has important practical value.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial process monitoring, and more particularly, to a digital factory monitoring data processing method. Background Art

[0002] Analyzing the propagation of anomalies in industrial production processes is crucial for ensuring production stability and product quality. Traditional anomaly analysis methods primarily rely on single-point monitoring and static threshold judgments, making it difficult to effectively characterize the dynamic propagation of anomalies in complex process systems. With the increasing automation and informatization of industry, massive amounts of equipment operating parameters, process parameters, and quality inspection data have accumulated in production processes. Anomaly propagation analysis based on this multi-source, heterogeneous data has become a research hotspot. Existing research primarily uses methods such as correlation analysis, causal inference, and graph theory to construct anomaly propagation models, extracting correlations between parameters to track the diffusion paths of anomalies. However, these methods often treat anomaly propagation as a static process, ignoring the dynamic evolution of parameter correlations in industrial production. This results in inaccurate and in-time identification of anomaly propagation paths.

[0003] Current anomaly propagation analysis technology faces the following major challenges: First, a systematic approach to characterizing multiscale data transmission characteristics is lacking, making it difficult to accurately capture anomaly propagation patterns at different time scales, such as rapid response, steady-state processes, and long-term evolution. Second, existing propagation feature extraction methods are mostly based on single-dimensional statistical indicators and fail to fully integrate multidimensional features such as time lag, amplitude correlation, and fluctuation consistency, thus affecting the reliability of propagation path identification. Third, traditional representations of anomaly propagation topology are overly simplistic, unable to effectively distinguish propagation paths of different credibility levels, and lack early warning mechanisms for dynamic transitions and sudden changes in intensity of propagation paths. These technical issues severely restrict the practical application of anomaly propagation analysis in complex industrial systems. Summary of the Invention

[0004] To address the above technical problems, the present invention provides a digital factory monitoring data processing method that can, to a certain extent, address the problems of incomplete feature extraction, unreliable propagation path identification, and insufficient early warning capabilities in industrial anomaly propagation analysis.

[0005] According to one aspect of the present invention, a method for processing digital factory monitoring data is provided, comprising:

[0006] Obtain monitoring data from factory monitoring points, establish a topological association diagram containing previous, current, and subsequent association points based on the process flow, and evaluate and eliminate abnormal data;

[0007] Based on the topological association graph, an adaptive time window method is used to calculate data transmission characteristics;

[0008] Establishing a three-layer data transfer matrix based on the transfer features, and calculating the abnormal propagation credibility based on the data transfer matrix when an abnormal monitoring point is detected;

[0009] A multi-level anomaly propagation topology diagram is generated based on the anomaly propagation credibility, and a comprehensive evaluation analysis is output in combination with the data transmission characteristics.

[0010] Furthermore, the monitoring data includes: equipment operating parameters, process parameters and quality inspection data, and a monitoring point topology association diagram including preceding, current and subsequent association points is established based on the process flow.

[0011] Furthermore, the data transmission characteristics include: time lag, amplitude correlation and fluctuation consistency;

[0012] The time lag is obtained by calculating the time difference between the peak values of the data at adjacent monitoring points;

[0013] The amplitude correlation is obtained by calculating the ratio of the data changes of adjacent monitoring points;

[0014] The fluctuation consistency is obtained by calculating the overlap of the fluctuation directions of the data at adjacent monitoring points.

[0015] Furthermore, data analysis is performed based on the adaptive time window method, and the parameter change period is identified through fast Fourier transform and wavelet analysis. The local steady-state characteristics and dynamic change characteristics of the data are captured in combination with the piecewise autocorrelation analysis and sliding entropy method, and the analysis window length is dynamically adjusted.

[0016] Furthermore, the determination of the window length adopts a multi-constraint adaptive mechanism. By setting a hard constraint range based on the process response time, the window length and sliding step are dynamically adjusted for different situations such as periodic fluctuations, sudden anomalies and working condition conversions, and a differentiated parameter adjustment strategy is established.

[0017] Furthermore, the data transmission characteristics between adjacent monitoring points are calculated based on the determined time window, including:

[0018] Calculate time lag based on peak identification;

[0019] Analyze the amplitude correlation by the ratio of the change amount;

[0020] Fluctuation consistency was assessed based on directional coincidence;

[0021] The nonlinear feature processing and the coupling effect of multiple transfer paths are considered at the same time, and the piecewise linearization and path decomposition methods are used to improve the calculation accuracy.

[0022] Furthermore, based on the data transfer characteristics, the sliding window method is used to divide the original data into a rapid response layer, a steady-state process layer, and a long-term evolution layer according to the sampling period, and a three-layer data transfer matrix is constructed, including:

[0023] An asymmetric time lag matrix is constructed through cross-correlation analysis of high-frequency data;

[0024] A symmetrical amplitude correlation matrix is constructed using the dynamic correlation coefficients of the aligned data;

[0025] A symmetric fluctuation consistency matrix is constructed through the dynamic time warping algorithm of long-term data.

[0026] Furthermore, a dynamic linkage mechanism is established based on the three-layer data transmission matrix. The time lag change of the rapid response layer triggers the adjustment of the calculation parameters of the steady-state process layer. The change of the correlation intensity of the steady-state process layer triggers the re-evaluation of the long-term evolution layer. The new collaborative mode of the long-term evolution layer feedback optimizes the feature extraction of the first two layers; and the credibility of the abnormal propagation is calculated by comprehensively evaluating the time dimension, intensity dimension and collaborative dimension.

[0027] Furthermore, a three-layer propagation topology diagram is constructed based on the abnormal propagation credibility, and the propagation intensity and timing characteristics are expressed through graphical features to generate a comprehensive report including abnormal source location, impact range analysis and propagation feature evaluation.

[0028] Furthermore, based on the comprehensive report, an early warning mechanism is established for abnormal propagation characteristics, including:

[0029] Monitoring time-lag fluctuations to warn of transmission path conversion risks;

[0030] Track the intensity attenuation changes to warn of mutation spread trends;

[0031] Analyze the possible problems of coupling intensity enhancement warning path merging;

[0032] Assess abnormalities in key nodes and warn of the formation of new transmission hubs.

[0033] Compared with the existing technology, the present invention realizes the multi-dimensional characterization of abnormal propagation characteristics by constructing a three-layer data transmission matrix, significantly improving the comprehensiveness and accuracy of abnormal propagation analysis. At the same time, a multi-level propagation topology structure based on credibility is proposed, which intuitively displays the importance and reliability of the propagation path through different graphic features, and establishes an early warning mechanism for abnormal propagation characteristics, which can timely detect potential risks such as propagation path conversion, intensity mutation, and path merging. And through the dynamic linkage mechanism between levels, adaptive adjustment of feature updates is achieved, so that abnormal propagation analysis can better adapt to the dynamic characteristics of complex industrial systems. It significantly improves the practicality and reliability of industrial abnormal propagation analysis, and provides more effective technical support for industrial process monitoring and fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0035] Figure 1 is a flow chart of a digital factory monitoring data processing method according to an embodiment of the present invention;

[0036] Figure 2 4 is a flowchart of calculating the anomaly propagation credibility according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0038] As mentioned in the background technology above, existing industrial anomaly propagation analysis technologies suffer from three major issues: First, there is a lack of systematic characterization methods for multi-scale data transmission characteristics, making it difficult to accurately capture anomaly propagation patterns at different time scales, such as rapid response, steady-state processes, and long-term evolution. Second, existing propagation feature extraction methods are mostly based on single-dimensional statistical indicators and fail to fully integrate multi-dimensional feature information such as time lag, amplitude correlation, and fluctuation consistency, which affects the reliability of propagation path identification. Third, traditional anomaly propagation topological structure representations are too simple to effectively distinguish propagation paths with different credibility levels, and lack early warning mechanisms for dynamic propagation path transitions and intensity mutations. Our invention addresses these technical pain points and proposes a digital factory monitoring data processing method. By constructing a three-layer data transmission matrix to characterize the multi-scale transmission relationship between parameters, integrating time, intensity, and coordination dimension features to calculate anomaly propagation credibility, and establishing a multi-level propagation topology structure and early warning mechanism, this method effectively addresses the problems of incomplete feature extraction, unreliable propagation path identification, and weak early warning capabilities in industrial anomaly propagation analysis. This invention belongs to the field of industrial process monitoring technology.

[0039] Figure 1 FIG. 1 is a flow chart of a method for processing digital factory monitoring data according to an embodiment of the present invention. Figure 1 As shown, the digital factory monitoring data processing method includes:

[0040] S1: Obtain monitoring data from factory monitoring points, establish a topological association diagram containing previous, current, and subsequent association points based on the process flow, and evaluate and eliminate abnormal data.

[0041] Acquire data from monitoring points distributed throughout each production process. The collected real-time monitoring data covers three dimensions: equipment operating parameters, process parameters, and quality inspection data.

[0042] Among them, the equipment operating parameters reflect the real-time operating status of the equipment, including the operating status data of the speed parameters, temperature parameters, pressure parameters, and vibration parameters of the production equipment, which can directly reflect the health status of the equipment;

[0043] Process parameters reflect the operation status of the production process, including process control data such as flow parameters, concentration parameters, liquid level parameters, and output parameters, which are directly related to the production process control of the product;

[0044] Quality inspection data reflects the quality characteristics of the product, including the product's size parameters, weight parameters, density parameters, and composition parameters, which determines whether the final product meets the quality requirements.

[0045] Based on the monitoring data obtained above, the relationship between each monitoring point is grasped. By analyzing the factory's production process, sorting out the material flow path and the process connection sequence, a topological relationship diagram of the monitoring points is established.

[0046] In this topological association diagram, the relationships between monitoring points are described at three levels: preceding, current, and subsequent. Predecessors are monitoring points located upstream of the monitoring point to be analyzed in the process flow. Data changes often transmit and affect the status of the current monitoring point. The current monitoring point is the target point for data analysis and is the core object of anomaly analysis. Subsequent points are monitoring points located downstream of the monitoring point to be analyzed in the process flow. Their status may be affected by data changes at the current monitoring point.

[0047] To ensure the accuracy of anomaly propagation analysis, a comprehensive quality assessment of the collected real-time monitoring data is required. First, focus on data integrity, checking whether the data series contains missing values or abnormal discontinuities to ensure data continuity. Second, assess data consistency, checking whether the data from different monitoring points are consistent in terms of unit dimensions to avoid analytical bias caused by inconsistent units. Also, verify the timeliness of the data, ensuring that the data collection time meets the requirements of real-time analysis and the timeliness of the analysis results. Finally, verify the validity of the data, determining whether the data values are within a reasonable range and eliminating outliers that clearly exceed the normal range.

[0048] The data quality assessment results are quantitatively evaluated using a preset quality threshold. If the data quality assessment result of a monitoring point falls below the preset quality threshold, it indicates that the data point may have an acquisition anomaly, sensor failure, or communication interruption, and is removed from the dataset.

[0049] Furthermore, a topological association diagram of monitoring points is established, including:

[0050] Obtain a complete process flow chart and process piping instrumentation diagram, and extract the material transfer network structure therein. Based on the material balance principle, identify the material transfer relationship between each process unit, including: material input node, material conversion node and material output node. For each material transfer path, determine the layout of its monitoring points, mainly including: process parameter measurement points such as flow, pressure, liquid level, and process parameter measurement points such as temperature and composition. In particular, for material diversion nodes and material confluence nodes, establish a multi-path transfer structure in the topological association diagram, use directed edges to represent the material flow direction, and mark the flow distribution ratio. For the material circulation loop, a closed-loop structure is used to represent it, and characteristic parameters such as the circulation ratio and material change cycle are marked.

[0051] Furthermore, for direct mechanical coupling systems, the corresponding relationship between the driving device and the driven device is identified, represented by solid lines in the topological association diagram, and parameters such as transmission ratio and efficiency are annotated. For indirect energy coupling systems, the energy transfer link between the heat source device, heat transfer medium, and heat receiving device is analyzed, represented by dashed lines in the topological association diagram, and characteristics such as heat transfer coefficient and delay time are annotated. For process control systems, the components of the control loop are extracted, including: measuring elements, controllers, actuators, and controlled objects. Special symbols are used to represent the control loop structure in the topological association diagram, and information such as control mode and control parameters are annotated.

[0052] Furthermore, based on process mechanisms and process dynamics, we identify causal relationships between key process parameters. For example, changes in feed flow rate can lead to changes in reactor temperature and pressure, and subsequently to changes in product quality parameters. By analyzing historical data, we extract characteristic parameters of parameter transfer: transfer time, transfer intensity, and transfer direction. In the topological association diagram, connecting lines of varying thickness represent transfer intensity, arrows indicate transfer direction, and characteristic parameters such as transfer time are annotated on the connecting lines.

[0053] Among them, the calculation of the transfer time adopts the cross-correlation analysis method to pre-process the time series data of the two monitoring points, including removing outliers, data standardization and eliminating trend items. When the sliding time window is used to calculate the cross-correlation function, the window length is set according to the process characteristics and is 3-5 times the parameter change period. During the calculation process, the cross-correlation function is obtained by evaluating the degree of correlation between the two parameter sequences under different time delays. When the cross-correlation function reaches its maximum value, the corresponding time delay is the parameter transfer time. If multiple local maxima are detected, a reasonable transfer time must be selected in combination with the process characteristics. When the parameter changes have periodic characteristics, the transfer time should be determined within a complete cycle.

[0054] Transfer strength is calculated using partial correlation analysis. When calculating the transfer strength between two parameters, we first identify and eliminate other potentially influencing parameter sets, then calculate the partial correlation coefficient between the target parameter pair. This coefficient reflects the true degree of correlation between the two parameters after eliminating the influence of other factors. When multiple influencing parameters exist, a recursive method is used to gradually calculate higher-order partial correlation coefficients. If the absolute value of the partial correlation coefficient is greater than a preset threshold, a significant transfer relationship is considered to exist between the parameters. When operating conditions change significantly, the transfer strength needs to be recalculated to adapt to the new operating conditions.

[0055] The direction of transmission is determined using the Granger causality test method. Two prediction models are constructed, one with and one without the parameter to be tested. The first model uses historical data for both parameters, while the second model uses historical data only for the parameter to be predicted. By comparing the predictions of these two models, if the model including the parameter to be tested significantly outperforms the model containing only a single parameter, it can be determined that the parameter to be tested has a significant causal influence on the target parameter. When performing a bidirectional causal test, the mutual influence of the two parameters needs to be tested simultaneously. If a bidirectional causal relationship is detected, the dominant transmission direction should be determined based on the transmission strength.

[0056] The resulting topological association diagram is a multi-layered network, with each monitoring point labeled with its location, functional attributes, and associated characteristics within the process flow. The connections between monitoring points not only reflect the material flow path but also encompass energy transfer channels and information transmission mechanisms.

[0057] S2: Calculate the data transmission characteristics between monitoring points based on the topological association graph, and use the adaptive time window method to calculate the transmission characteristics.

[0058] Based on the real-time data from monitoring points in the topological association graph, the data transmission characteristics between adjacent monitoring points are calculated, including time lag, amplitude correlation, and fluctuation consistency. Time lag is obtained by calculating the time difference between the peak values of the data at adjacent monitoring points. Amplitude correlation is calculated by calculating the ratio of the changes in the data at adjacent monitoring points. Fluctuation consistency is obtained by calculating the overlap in the fluctuation directions of the data at adjacent monitoring points.

[0059] To adapt to the dynamic nature of parameter changes in industrial production processes, an adaptive time window method is used for data analysis. When determining the length of the adaptive time window, spectral and autocorrelation analysis of historical data is performed to identify the primary cyclical components of parameter changes. When process parameters exhibit significant cyclical fluctuations, the time window length must encompass at least the entire cycle of variation. If parameter changes lack regularity, the appropriate window length is determined by calculating the statistical characteristics of the parameters.

[0060] Furthermore, when performing spectral analysis on historical data, a hybrid method combining fast Fourier transform and wavelet analysis is employed. While fast Fourier transform is used to identify the primary frequency components, wavelet transform is used to capture the sudden changes and local characteristics in the data. When the data contains multiple frequency components, not only is spectrum peak detection performed to determine the primary and secondary periods, but the energy contribution of each frequency component is also evaluated to select frequency components that have practical physical significance for parameter changes. When parameters exhibit non-stationary characteristics, wavelet packet analysis is employed to achieve a detailed characterization of the parameters in the time-frequency domain, with particular emphasis on identifying high-frequency transient characteristics caused by sudden anomalies.

[0061] When analyzing autocorrelation characteristics, a segmented autocorrelation analysis combined with sliding entropy is used. When calculating the autocorrelation function of a data sequence, local steady-state characteristics of the sequence are identified through segmented processing. For each segment, the autocorrelation function and sample entropy are calculated. Sudden changes in sample entropy often indicate changes in the system's dynamic characteristics. When the autocorrelation coefficient exhibits significant periodic fluctuations, this period can be used as a characteristic period of parameter variation, but the stability of this period must be verified through conditional autocorrelation analysis. When the autocorrelation coefficient decays below a preset threshold, the corresponding time delay can be used to determine the correlation length of the data. If multiple significant autocorrelation peaks are found, screening and confirmation based on typical periodic characteristics is necessary.

[0062] The time window length is determined using a multi-constraint adaptive mechanism, including:

[0063] Set a hard constraint range for the window length, with a lower limit of no less than 3 times the fastest process response time to ensure that the complete dynamic response process can be captured; the upper limit should not exceed 1.5 times the longest process cycle to avoid introducing irrelevant long-term fluctuations. Within the hard constraint range, the dynamic adjustment of the window length follows the following rules:

[0064] For periodic fluctuations under normal operating conditions, the window length is an integer multiple of the period;

[0065] When a sudden anomaly is detected, the dual time window mechanism is activated: the original long time window is maintained for steady-state feature analysis, while the short time window is opened for transient feature extraction. The analysis results of the two time windows are weighted fused;

[0066] During the working condition conversion process, the window length is dynamically adjusted with the parameter change rate: the faster the change rate, the shorter the window length, but it must meet the minimum length constraint; when the change tends to be stable, the window length gradually returns to the standard length;

[0067] Considering the computing resource constraints, the adaptive relationship between the window sliding step and the window length is set: when the window is short, the step size can be 1 / 10 of the window length; when the window is long, the step size can be appropriately increased, but not exceeding 1 / 5 of the window length.

[0068] For different types of process parameters, differentiated window length adjustment strategies are established.

[0069] The time lag between adjacent monitoring points within a defined time window is calculated. This lag reflects the time required for process parameters to propagate from upstream to downstream monitoring points. Peaks in the monitoring data are identified and the time difference between the peaks of adjacent monitoring points is calculated to determine the parameter propagation lag. To improve accuracy, the time difference between not only the primary peak but also the secondary peaks is calculated, resulting in a more reliable time lag through weighted averaging. Filtering is used during the calculation to eliminate the influence of false peaks.

[0070] The calculation of amplitude correlation primarily examines the proportional relationship between data changes at adjacent monitoring points. By calculating the change in monitoring data within a time window and taking the ratio of the changes at adjacent monitoring points, this method reflects the attenuation or amplification characteristics of parameter transfer. To ensure the stability of the calculation results, data changes at multiple time points are considered, and the calculated ratios are statistically analyzed. When nonlinear characteristics are present, a piecewise linearization method is used to process the relationship between data changes to reflect amplitude correlation characteristics under different operating conditions.

[0071] Fluctuation consistency reflects the degree of coordination between the directions of change in data from adjacent monitoring points. This index is derived by comparing the direction of change in the data from adjacent monitoring points within the same time window and calculating the percentage of time that these directions overlap. The calculation involves determining the direction of data change, including whether it is rising, falling, or flat. The time periods during which the direction of change between adjacent monitoring points coincides are then counted, and the percentage of these periods in the total analysis time is calculated. To avoid interference from minor fluctuations, a corresponding change threshold is set; only changes exceeding the threshold are considered valid fluctuations.

[0072] In the actual calculation process, appropriate adjustments need to be made in combination with the specific process characteristics. For example, when the process exhibits strong nonlinear characteristics, a piecewise linearization method is used for data preprocessing, and the parameter variation range is divided into multiple intervals based on the process mechanism. When the parameter values are in different intervals, the corresponding linear transformation function is used for data conversion. If the parameter changes span multiple intervals, smoothing processing is required at the interval boundaries to ensure the continuity of the converted data. When the process has saturation characteristics, a compression nonlinear function should be used for data conversion near the saturation region. If an exponential or power function relationship is found between the parameters, the nonlinear relationship is converted into a linear relationship through logarithmic transformation.

[0073] When multiple transfer paths exist, eigenvalues are calculated using path decomposition and path integration. Based on the process flow, all possible transfer paths are identified, including both direct and indirect paths. When calculating the eigenvalue of a direct transfer path, the influence of other paths is temporarily ignored. If multiple transfer paths exist in parallel, the eigenvalue is calculated for each path separately, and weighting coefficients are set based on their importance. When paths intersect, the mutual influence between the paths is considered, and a path coupling correction mechanism is established.

[0074] S3: A three-layer data transfer matrix is established based on the data transfer characteristics, and when an abnormal monitoring point is detected, the abnormal propagation credibility is calculated based on the data transfer matrix.

[0075] Based on the calculated time lag, amplitude correlation, and fluctuation consistency data, a sliding window method is used to divide the raw data into a rapid response layer, a steady-state process layer, and a long-term evolution layer according to the sampling period. When dividing the data, overlap between the data windows of each layer is ensured to achieve data connectivity between layers. If data is missing at a layer, interpolation is performed by combining the data characteristics of adjacent layers.

[0076] In the fast response layer, an n×n time lag matrix T is constructed, where n is the number of monitoring points. For each element Tij in the matrix, the time lag characteristics from monitoring point i to monitoring point j are determined by performing cross-correlation analysis on the high-frequency data. Specifically, the data sequence from monitoring point i is used as the input signal, and the data sequence from monitoring point j is used as the output signal. By calculating the cross-correlation function Rij(τ) = E[xi(t)·xj(t+τ)], the time delay τ that maximizes the function is found. This value is the value of the matrix element Tij. The time lag matrix is not symmetric because the parameter transfer is directional. It is continuously updated through a sliding time window, dynamically reflecting changes in the response relationship between parameters.

[0077] Based on the time-lag characteristics of the rapid response layer, the steady-state process layer constructs the amplitude correlation matrix R. During this construction, the hourly data are first time-aligned according to the corresponding element values in the time-lag matrix T to ensure that truly corresponding data segments are compared. By calculating the dynamic correlation coefficient of the aligned data series, the element value Rij in the amplitude correlation matrix R is obtained. This value reflects the strength of the correlation between parameters i and j. Unlike the time-lag matrix, the amplitude correlation matrix R is symmetric because the correlation strength between parameters is mutual. The correlation coefficient is continuously calculated using a sliding time window, and the calculation window is dynamically adjusted based on the time-lag changes detected by the rapid response layer.

[0078] Based on the time lag and amplitude correlation features, the long-term evolution layer constructs the fluctuation consistency matrix W. Also n×n-dimensional, its elements Wij characterize the long-term synergistic relationship between parameters i and j. During the construction process, the daily or weekly data are aligned and preprocessed using the features of the first two layers. The long-term similarity of the parameter sequences is then evaluated using a dynamic time warping algorithm. The fluctuation consistency matrix W is symmetric, and its element values reflect the degree of synergy between the fluctuation patterns of the parameters. The feature calculations at this level continuously track the long-term trend of the parameters. When changes in the synergistic pattern are detected, the feature calculation parameters of the first two layers are adjusted in feedback, forming a complete feature update mechanism.

[0079] Furthermore, within the dynamic linkage mechanism of the three-layer matrix, when the rapid response layer detects a change in the time lag characteristic, manifested as a significant change in the value of an element Tij in the time lag matrix T, it first assesses the reliability of this change and determines whether to trigger an adjustment in the steady-state process layer by calculating the ratio of the change amplitude to the historical fluctuation range. If adjustment is confirmed to be necessary, the steady-state process layer will accordingly change the calculation method of the amplitude correlation, including adjusting the time offset of the data alignment to match the new time lag characteristic. It also changes the calculation window length of the correlation coefficient. When the time lag becomes shorter, the calculation window is narrowed to increase sensitivity, and when the time lag becomes longer, the window is appropriately expanded to enhance stability.

[0080] When the steady-state process layer detects a significant change in correlation strength, that is, a persistent deviation in the element values of the amplitude correlation matrix R, the potential impact of this change on the long-term fluctuation characteristics is evaluated. First, the duration and trend of the correlation strength change are analyzed. If the duration of the change exceeds the preset threshold or shows a clear unidirectional evolution trend, the reassessment mechanism of the long-term evolution layer is triggered. The long-term evolution layer adjusts the calculation parameters of the fluctuation consistency, including expanding the time span of trend analysis, increasing the weight coefficient for emerging patterns, and updating the baseline value of the synergy assessment. At the same time, the synergy characteristics in the historical data are re-examined to evaluate whether the new correlation pattern represents a fundamental change in the system behavior.

[0081] If the long-term evolution layer identifies a new synergistic pattern, manifested as a new stable feature in the fluctuation consistency matrix W, this change is fed back into the feature extraction process of the first two layers. For the rapid response layer, the system optimizes the calculation parameters of time lags, including adjusting the time window size of the cross-correlation analysis, updating the dynamic threshold for lag determination, and changing the signal preprocessing method to better adapt to the new synergistic pattern. For the steady-state process layer, the criteria for amplitude correlation are adjusted, including updating the calculation method of the correlation coefficient, changing the criteria for significance testing, and optimizing the data alignment strategy.

[0082] Based on the three-layer data transfer matrix, the credibility of anomaly propagation is calculated by comprehensively evaluating the time lag characteristics, amplitude correlation characteristics, and fluctuation consistency characteristics, including:

[0083] Feature extraction is performed on the detected anomaly signal to identify the initial location and time of the anomaly. Based on the time lag matrix of the rapid response layer, the theoretical propagation time of the anomaly signal from the initial location to each relevant monitoring point is calculated. These times are used as baseline features of the anomaly propagation. When the actual observed anomaly propagation time matches the theoretical time, the propagation path is considered to have a high degree of temporal credibility.

[0084] On the basis of time feature verification, the amplitude correlation matrix of the steady-state process layer is further used to evaluate the intensity characteristics of anomaly propagation. By analyzing the changes in the correlation strength between parameters before and after the anomaly occurs, the amplitude attenuation law of the anomaly signal during the propagation process is calculated. If the observed amplitude change trend is consistent with the historical correlation pattern and the change amplitude is within a reasonable range, the credibility of the intensity dimension of the propagation path is improved.

[0085] The consistency matrix of the fluctuations in the long-term evolution layer is then combined to assess the degree of match between the anomaly propagation and the long-term synergistic characteristics. By comparing the current anomaly propagation pattern with the historically accumulated synergistic patterns, it is determined whether this propagation conforms to the long-term coupling laws between parameters. If the anomaly propagation path is found to be highly consistent with the known synergistic relationship and the fluctuation characteristics during the propagation process remain consistent, the credibility of the synergistic dimension of the propagation path is further confirmed.

[0086] The time dimension refers to the temporal characteristics of anomaly propagation in the system, which is reflected in the response delay and transmission order between parameters. It is described by the time lag matrix and reflects the causal relationship and dynamic response characteristics between different monitoring points.

[0087] The intensity dimension is used to describe the amplitude variation characteristics of the anomaly during the propagation process. It is represented by the amplitude correlation matrix, reflecting the coupling strength between parameters and the signal attenuation law, including the amplitude attenuation, amplification or conversion characteristics of the abnormal signal during the propagation process.

[0088] The synergy dimension is used to characterize the long-term synergistic relationship between parameters and the consistency of fluctuation patterns. It is described by the fluctuation consistency matrix, reflecting the stable characteristics and coupling laws between parameters on a longer time scale, including the synchronization, periodicity and trend consistency of parameter changes.

[0089] A weighted fusion approach is used to combine the credibility of the time, intensity, and coordination dimensions into an anomaly propagation credibility indicator. The weight distribution considers the reliability and importance of the features in different dimensions and is dynamically adjusted according to actual working conditions.

[0090] As the anomaly spreads, the credibility assessment results are adjusted in real time. When a significant change in the credibility of a particular propagation path is found, the credibility distribution of the entire propagation network is reassessed.

[0091] S4: Generate a multi-level anomaly propagation topology map based on the anomaly propagation credibility, combine the transmission characteristics, and output a comprehensive evaluation analysis.

[0092] Based on the calculated anomaly propagation credibility, a multi-level anomaly propagation topology structure is constructed. Specifically: a grading threshold for anomaly propagation credibility is set, and the propagation path with the highest credibility is classified into the main propagation path layer, the path with the second highest credibility is divided into the secondary propagation path layer, and the paths with relatively low credibility but still having reference value constitute the potential propagation path layer.

[0093] In the primary propagation path layer, the propagation links with the highest credibility are highlighted. By analyzing the time lag matrix of the rapid response layer, the propagation timing of the anomaly along the primary propagation path is calculated, including the propagation delay from the source to each affected node. Simultaneously, based on the amplitude correlation matrix of the steady-state process layer, the intensity changes of the anomaly signal during propagation are quantified, and the attenuation or amplification effect of each propagation link is calculated. Combined with the fluctuation consistency matrix of the long-term evolution layer, the stability of the primary propagation paths is evaluated to determine whether these propagation paths are consistent with the long-term operating characteristics of the system.

[0094] Similarly, within the secondary transmission path layer, identify transmission paths with slightly lower credibility but still significant. These paths may have some uncertainty or be confounded by other factors, but they still require special attention. Analyze the temporal, intensity, and stability characteristics of these paths, but use a different representation than for primary paths during visualization and risk assessment to quickly identify different levels of transmission risk.

[0095] At the potential transmission path layer, focus on those that are currently less reliable but may be activated under certain conditions. These typically exhibit weaker transmission characteristics or have a poor match with historical data, and are therefore included in the monitoring scope. Continue to track the evolution of these potential paths to promptly identify new transmission patterns.

[0096] Specifically, for the primary propagation path layer, the line width of the propagation path is proportional to the propagation strength; stronger propagation relationships have thicker lines. Solid arrows are used at node connections to indicate the propagation direction, and the size of the arrows reflects the time lag of propagation, with larger arrows indicating shorter time lags.

[0097] Secondary propagation paths are represented by dashed lines. The density of these lines indicates the confidence level of the propagation; higher confidence levels lead to denser dashed lines. Connections between nodes are represented by hollow arrows, whose transparency varies with the strength of the propagation; stronger propagation relationships have lower transparency.

[0098] At the potential propagation path level, dotted lines represent propagation links. The spacing between dotted lines reflects the uncertainty of propagation, with larger spacing indicating higher uncertainty. Nodes are connected by dotted arrows, with arrow length proportional to the degree of time lag, with longer arrows indicating longer time lags. Furthermore, the flashing frequency of the arrows indicates the activity of the propagation path; a higher frequency indicates a greater likelihood of the path transitioning to a higher level.

[0099] After constructing a complete multi-level topology, a comprehensive analysis report is generated. By analyzing the monitoring point where the anomaly first appears, combined with time lag characteristics and the credibility of the propagation path, the source of the anomaly is identified. The report also details the propagation of the anomaly at different levels, including the number of monitoring points involved at each level, the scope of process units covered, and the relationships between levels.

[0100] The report focuses on analyzing the anomaly's impact, including not only the areas already affected but also forecasting the potential impact. For propagation analysis, a detailed timeline is provided, noting the anomaly's occurrence and expected arrival time at each key node.

[0101] In terms of propagation intensity, the intensity variation law of abnormal signals on each propagation path is quantitatively described, and the impact of cumulative effect and coupling effect is evaluated.

[0102] It also further marks weak links and potential risk points that require special attention, providing targeted suggestions for subsequent monitoring and control.

[0103] For areas where the transmission path may change or the intensity may suddenly change, special early warning information will be provided, including:

[0104] When fluctuations in the time lag characteristics of a certain propagation node are detected, for example, the time lag that was originally stable at 30 seconds is suddenly shortened to 15 seconds, and at the same time, the node begins to show enhanced amplitude correlation characteristics with the nodes on other potential propagation paths, an early warning of a possible propagation path conversion is issued, indicating the downward trend in the stability of the current main propagation path, and the possible activation of new propagation paths, and an estimated time window for the conversion to occur is given.

[0105] If a change in the intensity attenuation pattern of a certain transmission link is observed, for example, the originally stable linear attenuation suddenly changes to exponential attenuation, and this change is accompanied by an increase in the fluctuation consistency of adjacent nodes, an intensity mutation warning will be generated, including a quantitative analysis of the intensity change, the boundary conditions for the mutation, and an assessment of the possible scope of impact. When the intensity mutation may lead to a significant increase in the propagation diffusion rate, the warning level will be raised to the highest level.

[0106] When it is found that the coupling strength of multiple propagation paths is enhanced at the same time, the focus is on analyzing the possible merging of propagation paths. For example, two originally relatively independent secondary propagation paths may merge into one main propagation path due to changes in certain process conditions. This will warn of the risk of reorganization of the topological structure and evaluate the possible amplification effect of the propagation intensity after the merger.

[0107] If the fluctuation consistency index of a certain propagation node is abnormal, and the node has a key position on multiple potential propagation paths, it will warn that the node may become a new propagation hub, including the evolution trend of node characteristics, the possible activated propagation path combination, and the cascade effect analysis that may be caused by this.

[0108] In summary, a digital factory monitoring data processing system based on an embodiment of the present invention is illustrated, which realizes a multi-dimensional characterization of abnormal propagation characteristics by constructing a three-layer data transfer matrix, significantly improving the comprehensiveness and accuracy of abnormal propagation analysis. At the same time, a multi-level propagation topology structure based on credibility is proposed, which intuitively displays the importance and reliability of the propagation path through different graphic features, and establishes an early warning mechanism for abnormal propagation characteristics, which can timely detect potential risks such as propagation path conversion, intensity mutation, and path merging. And through the dynamic linkage mechanism between levels, adaptive adjustment of feature updates is achieved, so that abnormal propagation analysis can better adapt to the dynamic characteristics of complex industrial systems. It significantly improves the practicality and reliability of industrial abnormal propagation analysis, and provides more effective technical support for industrial process monitoring and fault diagnosis.

Claims

1. A digital factory monitoring data processing method, characterized in that: include: Obtain monitoring data from factory monitoring points, establish a topological association diagram containing previous, current, and subsequent association points based on the process flow, and evaluate and eliminate abnormal data; Based on the topological association graph, an adaptive time window method is used to calculate data transmission characteristics; Establishing a three-layer data transfer matrix based on the transfer features, and calculating the abnormal propagation credibility based on the data transfer matrix when an abnormal monitoring point is detected; Generate a multi-level anomaly propagation topology map based on the anomaly propagation credibility, and output a comprehensive evaluation analysis based on the data transmission characteristics; Based on the data transfer characteristics, the sliding window method is used to divide the original data into a rapid response layer, a steady-state process layer, and a long-term evolution layer according to the sampling period, and a three-layer data transfer matrix is constructed, including: An asymmetric time lag matrix is constructed through cross-correlation analysis of high-frequency data; A symmetrical amplitude correlation matrix is constructed using the dynamic correlation coefficients of the aligned data; A symmetrical volatility consistency matrix is constructed using a dynamic time warping algorithm for long-term data; A dynamic linkage mechanism is established based on the three-layer data transmission matrix. The time lag change of the rapid response layer triggers the adjustment of the calculation parameters of the steady-state process layer. The change of the correlation intensity of the steady-state process layer triggers the re-evaluation of the long-term evolution layer. The new collaborative mode of the long-term evolution layer is used to feedback and optimize the feature extraction of the first two layers. The credibility of the anomaly propagation is calculated by comprehensively evaluating the time dimension, intensity dimension and collaborative dimension.

2. The digital factory monitoring data processing method according to claim 1, characterized in that: The monitoring data includes: equipment operating parameters, process parameters and quality inspection data, and a monitoring point topology association diagram including preceding, current and subsequent association points is established based on the process flow.

3. The digital factory monitoring data processing method according to claim 2, characterized in that: The data transmission characteristics include: time lag, amplitude correlation and fluctuation consistency; The time lag is obtained by calculating the time difference between the peak values of the data at adjacent monitoring points; The amplitude correlation is obtained by calculating the ratio of the data changes of adjacent monitoring points; The fluctuation consistency is obtained by calculating the overlap of the fluctuation directions of the data at adjacent monitoring points.

4. The digital factory monitoring data processing method according to claim 3, characterized in that: Data analysis is performed based on the adaptive time window method, and the parameter change cycle is identified through fast Fourier transform and wavelet analysis. The local steady-state characteristics and dynamic change characteristics of the data are captured in combination with piecewise autocorrelation analysis and sliding entropy method, and the analysis window length is dynamically adjusted.

5. The digital factory monitoring data processing method according to claim 4, characterized in that: The determination of the window length adopts a multi-constraint adaptive mechanism. By setting a hard constraint range based on the process response time, the window length and sliding step are dynamically adjusted according to different situations such as periodic fluctuations, sudden abnormalities and working condition conversions, and a differentiated parameter adjustment strategy is established.

6. The digital factory monitoring data processing method according to claim 5, characterized in that: Calculate the data transmission characteristics between adjacent monitoring points based on a certain time window, including: Calculate time lag based on peak identification; Analyze the amplitude correlation by the ratio of the change amount; Evaluate fluctuation consistency based on directional coincidence; The nonlinear feature processing and the coupling effect of multiple transfer paths are considered at the same time, and the piecewise linearization and path decomposition methods are used to improve the calculation accuracy.

7. The digital factory monitoring data processing method according to claim 1, characterized in that: A three-layer propagation topology diagram is constructed based on the anomaly propagation credibility, and the propagation intensity and timing characteristics are expressed through graphical features to generate a comprehensive report including anomaly source location, impact range analysis and propagation feature evaluation.

8. The digital factory monitoring data processing method according to claim 7, characterized in that: Based on the comprehensive report, an early warning mechanism is established for abnormal propagation characteristics, including: Monitoring time-lag fluctuations to warn of transmission path conversion risks; Track the intensity attenuation changes to warn of mutation spread trends; Analyze the possible problems of coupling intensity enhancement warning path merging; Assess abnormalities in key nodes and warn of the formation of new transmission hubs.

Images