A production process abnormal root cause tracing method and system based on a time sequence knowledge graph

Abstract

This invention relates to the field of anomaly diagnosis technology, specifically disclosing a method and system for tracing the root causes of production process anomalies based on a time-series knowledge graph. The method includes: acquiring time-series data from multi-source sensors on the production line, performing hierarchical dynamic updating graph modeling, and constructing a time-series production knowledge graph; capturing and classifying abnormal events according to preset anomaly identification rules, and using local abnormal events as precise trigger points; performing reverse causal backtracking on the time-series feature entities of the precise trigger points based on the time-series production knowledge graph to obtain a directed backtracking subgraph; performing dual-dimensional root cause quantification on the causal paths of the directed backtracking subgraph to obtain a suspected root cause sequence; performing multi-path hybrid weighted root cause quantification on the suspected root cause sequence to determine the final root cause location result; and outputting an intelligent traceability analysis report based on the root cause location result and relevant data across the entire chain. This invention improves the accuracy and efficiency of tracing the root causes of production process anomalies.

Description

Technical Field

[0001] This invention relates to the field of anomaly diagnosis technology, and in particular to a method and system for tracing the root causes of anomalies in a production process based on a time-series knowledge graph. Background Technology

[0002] In existing technologies, root cause tracing methods for production process anomalies typically rely on expert experience rules or threshold alarms based on statistical process control. When a production anomaly occurs, operators need to check relevant equipment parameters, process indicators, and material states one by one based on alarm information and personal experience. This process lacks the ability to systematically model the physical relationships such as material flow direction and energy transfer paths in the process flow, making it difficult to establish an effective connection between discrete alarm information and causal transmission relationships between workstations. For continuous process industries, an anomaly is often caused by the cumulative deviation of multiple upstream workstations propagating step by step along the material or energy flow. Traditional methods cannot construct anomaly propagation paths with clear causal orientations in the time and space dimensions, leading to the tracing results often misjudging accompanying phenomena as the root cause, making it difficult to accurately locate the source of the anomaly.

[0003] In existing technologies, some factories have introduced data analysis methods based on association rule mining, which infer the causes of anomalies by calculating the correlation coefficients between different parameters in historical data. However, these methods can only discover statistical correlations at the data level, failing to distinguish between correlation and causation, and neglecting the impact of time lag effects and energy attenuation characteristics on transmission intensity during anomaly propagation. Furthermore, traditional methods offer a single dimension for the quantitative assessment of anomalies, typically ranking them only based on the magnitude of parameter deviations, lacking a comprehensive consideration of the severity of the node's own anomalies and its ability to propagate anomalies downstream. This makes it difficult to identify key nodes with insignificant deviations but that transmit energy downstream. Therefore, there is an urgent need to develop an anomaly root cause tracing method that integrates physical correlations in the process flow, temporal logical constraints, and dual-dimensional quantitative assessment to address the problems of causal confusion, missing temporal sequences, and a single assessment dimension in existing technologies, thereby improving the accuracy of root cause localization and the interpretability of diagnostic results. This invention provides a method and system for tracing the root causes of production process anomalies based on time-series knowledge graphs, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, this invention provides a method for tracing the root causes of production process anomalies based on time-series knowledge graphs, comprising: A1: Acquire time-series data from multiple sources of sensors in the production line, and perform hierarchical dynamic update graph modeling on the time-series data to obtain the time-series production knowledge graph of the production line; A2: Based on preset anomaly identification rules, the abnormal events of the production line are captured and classified to obtain the local abnormal events of the production line, and the local abnormal events are used as the precise trigger points of the production line. A3: Based on the time-series production knowledge graph, reverse causal backtracking is performed on the time-series feature entities of the precise trigger point to obtain the directed backtracking subgraph of the production line; A4: Perform two-dimensional root cause quantification on the causal path of the directed backtracking subgraph to obtain the suspected root cause sequence of the production line; A5: Perform multi-path hybrid weighted root cause quantization localization on the temporal feature entities of the suspected root cause sequence to obtain the root cause localization result of the production line; A6: Based on the full-chain related data of the root cause localization results, output the intelligent traceability analysis report of the production line.

[0005] In a preferred embodiment, the step of acquiring time-series data from multiple sources in the production line and performing hierarchical dynamic update graph modeling on the time-series data to obtain a time-series production knowledge graph of the production line includes: Based on the process flow design document of the production line, a topological model of the material and energy flow relationship of the workstations in the production line is performed to obtain the static ecological production topology map of the production line. Based on the multi-source sensors in the production line, the time-series data of the multi-source sensors are labeled with workstation spatiotemporal tags to obtain the time-series data stream of the production line; Based on the order of the timestamps in the workstation spatiotemporal labels, the data in the time-series data stream is matched and entered into the workstation nodes of the static ecological production topology map to obtain the time-series data sequence of the workstation nodes. The time-series data sequence is dynamically updated using a sliding window to obtain the time-series production knowledge graph of the production line.

[0006] In a preferred embodiment, the step of capturing and classifying abnormal events on the production line based on preset anomaly identification rules to obtain local abnormal events on the production line, and using these local abnormal events as precise trigger points for the production line, includes: Based on preset anomaly identification rules, anomaly detection and judgment are performed on the workstation parameters in the time-series production knowledge graph to obtain the abnormal events of the production line. The deviation of parameters from the abnormal events is assessed to determine the level of abnormal events on the production line. The hierarchical events are filtered and located to obtain local abnormal events of the production line, and these local abnormal events are used as precise trigger points for the production line.

[0007] In a preferred embodiment, the step of performing reverse causal backtracking on the temporal feature entities of the precise trigger point based on the temporal production knowledge graph to obtain a directed backtracking subgraph of the production line includes: Queries the upstream associated entities of the workstation entity in the precise trigger point; Perform anomaly detection on the historical time series data of the upstream associated entity to obtain the anomaly detection result of the workstation entity itself; Based on the results of the self-abnormal event determination, the upstream associated entities are subjected to abnormal effective screening to obtain the set of effective upstream nodes of the workstation entity, and the associated edges of the effective upstream nodes in the set of effective upstream nodes are retained. The effective upstream nodes are iteratively backtracked until no effective upstream nodes are available, thus obtaining the abnormal link tracing result of the production line; Based on the abnormal link tracing results, the effective upstream nodes and the associated edges are constructed to obtain the directed backtracking subgraph of the production line.

[0008] In a preferred embodiment, the step of performing anomaly detection on the historical time-series data of the upstream associated entity to obtain the anomaly event detection result of the workstation entity includes: Based on the time-series production knowledge graph, obtain the time-series data of the upstream related entities; The time series data is subjected to time series anomaly detection to obtain the self-abnormal event determination result of the workstation entity; If the time series data has a period of parameter change, then the result of the self-abnormal event determination is that there is a self-abnormal event; If the time series data does not have a period of parameter change, then the result of the self-abnormal event determination is that the abnormality condition is not met.

[0009] In a preferred embodiment, the step of performing two-dimensional root cause quantification on the causal path of the directed backtracking subgraph to obtain the suspected root cause sequence of the production line includes: Obtain the maximum value of parameter deviation and the duration of anomalies of the abnormal event nodes in the directed backtracking subgraph, and use them as the first dimension quantization value; Obtain the propagation energy value pointing to the downstream node in the abnormal event node, and use it as the second dimension quantization value; The first dimension quantization value and the second dimension quantization value are coupled and quantized to obtain the comprehensive impact score of the production line; The formula for calculating the comprehensive impact score is as follows: ; in, The overall impact score is as follows. For nodes The maximum deviation of the parameters corresponding to the abnormal event. The maximum parameter deviation of all nodes in the directed backtracking subgraph. For nodes The duration corresponding to the abnormal event, The preset reference duration, It is the natural logarithm. For nodes The set of all downstream nodes pointed to in the directed backtracking subgraph, For the node To the node The energy value propagated on the directed edge, The maximum propagation energy value on all directed edges in the directed backtracking subgraph; The comprehensive impact scores are sorted and filtered to obtain the suspected root cause sequence of the production line.

[0010] In a preferred embodiment, obtaining the propagation energy value pointing to the downstream node from the abnormal event node includes: The workstation physical association and timing characteristics of the abnormal event nodes are quantitatively analyzed to obtain the transmission attenuation coefficient of the abnormal event nodes. The magnitude and duration of parameter deviation of the abnormal event node are quantitatively evaluated to obtain the abnormal excitation intensity of the abnormal event node; The propagation energy value pointing to the downstream node in the abnormal event node is obtained by weighted fusion of the abnormal excitation intensity and the propagation attenuation coefficient.

[0011] In a preferred embodiment, the step of performing multi-path hybrid weighted root cause quantization localization on the temporal feature entities of the suspected root cause sequence to obtain the root cause localization result of the production line includes: The candidate nodes in the suspected root cause sequence are traced and aggregated along the entire path to obtain the path set of the suspected root cause sequence. The path weights of the path set are obtained by accumulating the comprehensive influence scores of the nodes on the paths in the path set. The extreme value of the path weight is used as the multi-path mixing weight of the candidate node; The nodes in the suspected root cause sequence are reordered according to the multi-path mixed weight to obtain the root cause localization result of the production line.

[0012] In a preferred embodiment, the step of outputting an intelligent traceability analysis report for the production line based on the end-to-end related data of the root cause localization results includes: Based on the time-series production knowledge graph, multi-dimensional information parsing is performed on the workstation nodes of the root cause localization results to obtain the full-link related data of the workstation nodes; The propagation link of the directed backtracking subgraph is reconstructed to obtain the complete propagation path of the directed backtracking subgraph; Based on the relevant data across the entire chain, the root cause localization results and the relevant parameters of the complete propagation path are structurally integrated to obtain the traceability analysis report of the production line, and an intelligent traceability analysis report of the production line is output.

[0013] To address the aforementioned problems, this invention also provides a production process anomaly root cause tracing system based on time-series knowledge graphs, the system comprising: The time-series graph construction module is used to acquire time-series data from multiple sources of sensors in the production line, and to perform hierarchical dynamic updating graph modeling on the time-series data to obtain the time-series production knowledge graph of the production line. An anomaly capture and classification module is used to capture and classify abnormal events of the production line based on preset anomaly identification rules, obtain local anomaly events of the production line, and use the local anomaly events as precise trigger points of the production line. The reverse causal backtracking module is used to perform reverse causal backtracking on the temporal feature entities of the precise trigger point based on the temporal production knowledge graph, and obtain the directed backtracking subgraph of the production line; The dual-dimensional quantization judgment module is used to perform dual-dimensional root cause quantization judgment on the causal path of the directed backtracking subgraph to obtain the suspected root cause sequence of the production line. A multi-path hybrid localization module is used to perform multi-path hybrid weighted root cause quantization localization on the temporal feature entities of the suspected root cause sequence to obtain the root cause localization result. The traceability analysis report module is used to output an intelligent traceability analysis report for the production line based on the full-link related data of the root cause localization results.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs a temporal knowledge graph based on the dual constraints of physical correlation and temporal sequence in the process flow. It transforms the essential laws of the production process, such as material flow direction and energy transfer paths, into directed edges in the graph. Simultaneously, it uses the temporal sequence of abnormal events to match upstream and downstream workstations, ensuring that every causal relationship in the graph satisfies both physical and temporal logic. Compared to traditional traceability methods that rely on statistical correlation or expert experience, this invention fundamentally avoids the problem of causal confusion caused by misjudging accompanying phenomena as root causes, significantly improving the authenticity and traceability of abnormal propagation paths.

[0015] 2. This invention introduces a dual-dimensional root cause quantification mechanism, coupling the severity of anomalies within a node with its ability to propagate anomalies downstream, forming a comprehensive impact score. This mechanism not only considers the intensity of the anomalous event but also fully incorporates the transmission effect of the anomaly along the process chain, avoiding the problem of missing key nodes caused by traditional single-dimensional ranking. Furthermore, through multi-path hybrid weight localization technology, path weights are accumulated for all possible propagation paths, and the node with the highest multi-path hybrid weight is selected as the root cause, ensuring that the localization result takes into account both path diversity and the dominance of energy transmission. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for tracing the root causes of production process anomalies based on a time-series knowledge graph, as provided in an embodiment of the present invention. Figure 2 This is a functional module diagram of a production process anomaly root cause tracing system based on time-series knowledge graphs, provided in an embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] This application provides a method for tracing the root causes of production process anomalies based on a time-series knowledge graph. The execution entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cluster of cloud servers. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0019] Reference Figure 1 The diagram shown is a flowchart illustrating a method for tracing the root causes of production process anomalies based on a time-series knowledge graph, according to an embodiment of the present invention. In this embodiment, the method for generating reference information for tracing the root causes of production process anomalies based on a time-series knowledge graph includes: A1: Acquire time-series data from multiple sources of sensors in the production line, and perform hierarchical dynamic update graph modeling on the time-series data to obtain the time-series production knowledge graph of the production line; In this embodiment of the invention, the step of acquiring time-series data from multiple sources of sensors in the production line and performing hierarchical dynamic update graph modeling on the time-series data to obtain a time-series production knowledge graph of the production line includes: Based on the process flow design document of the production line, a topological model of the material and energy flow relationship of the workstations in the production line is performed to obtain the static ecological production topology map of the production line. Based on the multi-source sensors in the production line, the time-series data of the multi-source sensors are labeled with workstation spatiotemporal tags to obtain the time-series data stream of the production line; Based on the order of the timestamps in the workstation spatiotemporal labels, the data in the time-series data stream is matched and entered into the workstation nodes of the static ecological production topology map to obtain the time-series data sequence of the workstation nodes. The time-series data sequence is dynamically updated using a sliding window to obtain the time-series production knowledge graph of the production line.

[0020] The process flow design document details the entire production process from raw material inlet to finished product outlet, including the functional positioning of each workstation, the sequence of upstream and downstream connections, material transfer methods, energy medium types, and transfer directions. By analyzing this document, all workstation nodes and the directed connections between them are extracted. Using workstations as nodes and material flow or energy transfer as directed edges, a static ecological production topology map reflecting the physical structure of the production line is constructed. This map describes the inherent topology of the production line, does not contain temporal information, but provides a fixed framework for subsequent data integration.

[0021] The system utilizes multi-source sensors deployed at various workstations along the production line to collect data in real time. These sensors include temperature sensors, pressure sensors, flow meters, vibration sensors, position sensors, and material tracking systems. Each sensor automatically records its workstation number and the time of data collection, creating raw data with a workstation number and timestamp. The system then adds spatiotemporal tags to this raw data, clearly assigning each data point a specific workstation affiliation and precise collection time, transforming it into a structured data stream to obtain the time-series data stream of the production line. This process binds dispersed sensor data to specific workstations and time points, laying the foundation for subsequent map filling.

[0022] The system matches and inputs data from the time-series data stream into the corresponding workstation nodes in the static ecological production topology map, based on the timestamp order carried in the workstation spatiotemporal tags. Specifically, for each data point, the system first locates the corresponding workstation node in the static ecological production topology map according to its workstation number, and then appends the data to the pre-set time-series data storage area of ​​that node in timestamp order. In this way, each workstation node maintains a time-series data sequence that extends over time. Each data point in the sequence carries the collection time and parameter value, reflecting the changing operational status of the workstation over time. As new data is continuously added, the time-series data sequence of the workstation node continuously extends, forming a real-time data view reflecting the dynamic evolution of the production process.

[0023] To control data storage scale and maintain data freshness, the system dynamically updates the time-series data sequence of each workstation using a sliding window. A fixed time window length is preset, such as the most recent 30 minutes. When the length of a node's time-series data sequence exceeds the number of data points that the window can hold, the system removes the oldest data in the sequence and adds the newly arrived data, ensuring that each workstation node always retains only the time-series data within the most recent time window. Through this sliding window mechanism, the system effectively controls memory usage while ensuring data timeliness, ultimately forming a time-series production knowledge graph that is dynamically updated over time and contains data with a fixed time span for each workstation node. This graph not only completely preserves the static physical structure of the production line but also reflects the operational status of each workstation in real time within the most recent time window, providing a solid data foundation for subsequent anomaly identification and root cause tracing.

[0024] The beneficial effects are that by constructing a static ecological production topology map that reflects the physical structure of the production line, and by labeling and sequentially matching the time-series data of multi-source sensors with spatiotemporal tags at each workstation, a time-series data sequence of each workstation node is formed. Then, the latest data within a fixed time window is dynamically updated and retained through a sliding window. Finally, a time-series production knowledge graph that integrates static physical structure and dynamic operating status is obtained, which provides a structured, traceable, and real-time updated data foundation for subsequent abnormal event identification, reverse causal backtracking, and root cause quantification.

[0025] A2: Based on preset anomaly identification rules, the abnormal events of the production line are captured and classified to obtain the local abnormal events of the production line, and the local abnormal events are used as the precise trigger points of the production line. In this embodiment of the invention, the step of capturing and classifying abnormal events of the production line based on preset anomaly identification rules to obtain local abnormal events of the production line, and using the local abnormal events as precise trigger points of the production line, includes: Based on preset anomaly identification rules, anomaly detection and judgment are performed on the workstation parameters in the time-series production knowledge graph to obtain the abnormal events of the production line. The deviation of parameters from the abnormal events is assessed to determine the level of abnormal events on the production line. The hierarchical events are filtered and located to obtain local abnormal events of the production line, and these local abnormal events are used as precise trigger points for the production line.

[0026] Based on preset anomaly identification rules, the system monitors and determines anomalies in the parameters of each workstation node within the time-series production knowledge graph. The preset anomaly identification rules pre-define upper and lower limits for process specifications and a counting threshold for the duration of anomalies for each parameter type at each workstation node. The system reads the parameter values ​​of each workstation node from the time-series production knowledge graph in real time, comparing the current parameter value with the preset upper and lower limits for process specifications point by point. When the parameter value of a workstation continuously exceeds the upper or lower limit for process specifications, and the number of consecutively exceeding sampling points reaches the preset counting threshold, the system marks the period of continuous exceedance as an anomaly event for that workstation. All marked anomaly events are aggregated to obtain the anomaly event set for the production line. This process, through real-time monitoring and rule comparison, accurately captures all abnormal periods exceeding the normal range from massive amounts of time-series data.

[0027] The system calculates the parameter deviation magnitude and classifies the anomalies detected. For each anomaly, the system extracts the maximum or minimum parameter value during the event period and calculates the parameter deviation magnitude based on the degree to which this value deviates from the upper or lower limit of the process specification. The parameter deviation magnitude reflects the severity of the anomaly. Subsequently, the system classifies each anomaly into different levels according to a preset deviation magnitude classification table; for example, anomalies with small deviation magnitudes are classified as Level 1, those with moderate deviation magnitudes as Level 2, and those with large deviation magnitudes as Level 3. Through the classification and assessment of anomalies, the system can distinguish anomalies of different severity, providing a basis for subsequent screening.

[0028] The system filters and locates all categorized events, selecting the most critical local anomaly from numerous abnormal events and using it as the precise trigger point. The specific filtering and location method is as follows: among all marked abnormal events, the highest-level anomaly is selected as a candidate; if multiple events of the same level exist, the one that occurred most recently is selected as the local anomaly. This filtering method ensures that the selected local anomaly is the most serious and recently occurring anomaly, and is most likely to become the final manifestation of a series of chain reactions. Determining the workstation node corresponding to this local anomaly and the time of its occurrence yields the precise trigger point of the production line. This precise trigger point will serve as the starting point for subsequent reverse causal backtracking to trace the source of the anomaly's propagation.

[0029] The beneficial effects are as follows: by using the anomaly identification rules of preset upper and lower limits of process specifications and continuous counting thresholds, the parameters of each workstation in the time-series production knowledge graph are compared in real time and continuously exceeded, accurately capturing all abnormal periods that exceed the normal range and forming a set of abnormal events on the production line; then, the parameter deviation of each abnormal event is graded and evaluated to distinguish different levels of anomaly severity, and the event with the highest level and latest occurrence time is selected as the local abnormal event and precise trigger point, thereby quickly locking the most serious and latest source of anomaly in complex production data, providing a clear and focused starting point for subsequent reverse causal backtracking, effectively avoiding invalid alarm interference, and significantly improving the pertinence and efficiency of anomaly root cause tracing.

[0030] A3: Based on the time-series production knowledge graph, reverse causal backtracking is performed on the time-series feature entities of the precise trigger point to obtain the directed backtracking subgraph of the production line; In this embodiment of the invention, the step of performing reverse causal backtracking on the temporal feature entities of the precise trigger point based on the temporal production knowledge graph to obtain the directed backtracking subgraph of the production line includes: Queries the upstream associated entities of the workstation entity in the precise trigger point; Perform anomaly detection on the historical time series data of the upstream associated entity to obtain the anomaly detection result of the workstation entity itself; Based on the results of the self-abnormal event determination, the upstream associated entities are subjected to abnormal effective screening to obtain the set of effective upstream nodes of the workstation entity, and the associated edges of the effective upstream nodes in the set of effective upstream nodes are retained. The effective upstream nodes are iteratively backtracked until no effective upstream nodes are available, thus obtaining the abnormal link tracing result of the production line; Based on the abnormal link tracing results, the effective upstream nodes and the associated edges are constructed to obtain the directed backtracking subgraph of the production line.

[0031] In this embodiment of the invention, the step of performing anomaly determination on the historical time-series data of the upstream associated entity to obtain the anomaly event determination result of the workstation entity includes: Based on the time-series production knowledge graph, obtain the time-series data of the upstream related entities; The time series data is subjected to time series anomaly detection to obtain the self-abnormal event determination result of the workstation entity; If the time series data has a period of parameter change, then the result of the self-abnormal event determination is that there is a self-abnormal event; If the time series data does not have a period of parameter change, then the result of the self-abnormal event determination is that the abnormality condition is not met.

[0032] Based on a time-series production knowledge graph, the system performs reverse causal backtracking on workstation entities within precise trigger points. The system locates the workstation entity corresponding to the precise trigger point from the time-series production knowledge graph and designates this workstation entity as the current node. Subsequently, the system queries all upstream related entities of this workstation entity within the time-series production knowledge graph, i.e., all workstation entities whose directed edges point to the current workstation entity. These upstream related entities represent workstations located before the current workstation in the process flow and with material flow or energy transfer relationships. The query operation is completed by traversing the directions of directed edges in the time-series production knowledge graph, retaining only those starting workstation entities whose directed edge endpoints coincide with the current workstation entity, thus obtaining the set of upstream related entities.

[0033] The system performs anomaly detection for each upstream associated entity. Based on a time-series production knowledge graph, the system obtains the time-series data of the upstream associated entity, which is the continuous parameter value sequence stored within the time window for that workstation. The system then scans this time-series data to determine if there are any periods of parameter change. The method for determining parameter change periods is as follows: traversing the sampling points in the time-series data, when a parameter value is detected to continuously deviate from the normal range and the number of deviating sampling points reaches a preset threshold, the continuous deviation period is marked as a parameter change period. If at least one such parameter change period exists in the time-series data, the upstream associated entity is determined to have its own anomaly event, and the anomaly event determination result is "existence of an anomaly event"; if no parameter change period exists in the time-series data, the anomaly event determination result is "not meeting the anomaly existence condition".

[0034] Based on the above self-abnormal event determination results, the system performs anomaly validity screening on upstream related entities. For an upstream related entity, if its self-abnormal event determination result indicates the existence of its own abnormal event, the node is retained as a valid upstream node, and the associated edge pointing from that node to the current workstation entity is also retained; if its self-abnormal event determination result indicates that the abnormality existence condition is not met, the node is removed from the candidate set. After screening, all retained valid upstream nodes and their corresponding associated edges constitute the set of valid upstream nodes for the current workstation entity.

[0035] The system uses a valid upstream node as the new current node and repeatedly performs the above operations of querying upstream related entities, determining abnormalities, and filtering valid exceptions. That is, it continues to trace upstream from the valid upstream node until no valid upstream node that meets the conditions can be found at the current node. This iterative process extends upstream continuously, recording all visited nodes and the associated edges between nodes, and finally forming a complete link from the upstream node from which it is impossible to trace back further to the precise trigger point.

[0036] When all branches can no longer be backtracked, the system constructs a directed acyclic graph (DAG) structure, consisting of all visited valid upstream nodes and the remaining edges between them. This DAG structure ends at the precise trigger point and begins at each node that ultimately cannot be backtracked. This structure is the directed backtracking subgraph of the production line. This DAG fully records all possible paths of the anomaly propagating downstream from its initial source along the process flow until the final anomaly, providing a clear causal structure foundation for subsequent root cause quantification.

[0037] The beneficial effect is that by starting from the precise trigger point, upstream related entities are queried level by level along the time-series production knowledge graph, and the effective upstream nodes with their own abnormal events are strictly screened using the super-differential anomaly judgment, forming a set of effective upstream nodes and their associated edges. Then, the effective upstream nodes are iteratively backtracked until they can no longer be traced. Finally, a directed backtracking subgraph is constructed with the precise trigger point as the end point and the node that can no longer be traced as the starting point. This fully restores all possible paths of the anomaly from the initial source to the downstream of the process flow until the final anomaly, providing a clear path structure with clear causal logic and physical correlation support for subsequent root cause quantification.

[0038] A4: Perform two-dimensional root cause quantification on the causal path of the directed backtracking subgraph to obtain the suspected root cause sequence of the production line; In this embodiment of the invention, the step of performing two-dimensional root cause quantification on the causal path of the directed backtracking subgraph to obtain the suspected root cause sequence of the production line includes: Obtain the maximum value of parameter deviation and the duration of anomalies of the abnormal event nodes in the directed backtracking subgraph, and use them as the first dimension quantization value; Obtain the propagation energy value pointing to the downstream node in the abnormal event node, and use it as the second dimension quantization value; The first dimension quantization value and the second dimension quantization value are coupled and quantized to obtain the comprehensive impact score of the production line; The formula for calculating the comprehensive impact score is as follows: ; in, The overall impact score is as follows. For nodes The maximum deviation of the parameters corresponding to the abnormal event. The maximum parameter deviation of all nodes in the directed backtracking subgraph. For nodes The duration corresponding to the abnormal event, The preset reference duration, It is the natural logarithm. For nodes The set of all downstream nodes pointed to in the directed backtracking subgraph, For the node To the node The energy value propagated on the directed edge, The maximum propagation energy value on all directed edges in the directed backtracking subgraph; The comprehensive impact scores are sorted and filtered to obtain the suspected root cause sequence of the production line.

[0039] In this embodiment of the invention, obtaining the propagation energy value pointing to the downstream node from the abnormal event node includes: The workstation physical association and timing characteristics of the abnormal event nodes are quantitatively analyzed to obtain the transmission attenuation coefficient of the abnormal event nodes. The magnitude and duration of parameter deviation of the abnormal event node are quantitatively evaluated to obtain the abnormal excitation intensity of the abnormal event node; The propagation energy value pointing to the downstream node in the abnormal event node is obtained by weighted fusion of the abnormal excitation intensity and the propagation attenuation coefficient.

[0040] The causal path of the directed backtracking subgraph is determined using two-dimensional root cause quantization. The system traverses the abnormal event nodes in the directed backtracking subgraph and performs quantization operations for each node.

[0041] For each abnormal event node, the system extracts the maximum parameter deviation and the duration of the abnormal event corresponding to that node, and uses these two values ​​as the first dimension of quantification. The maximum parameter deviation is determined by backtracking the parameter values ​​of all sampling points during the abnormal period of that node, and extracting the value with the most severe deviation from the process specification limit as the deviation magnitude of that node. The duration of the abnormality is obtained by calculating the time between the start and end of the abnormality at that node, reflecting the persistence of the abnormality's impact.

[0042] The system simultaneously extracts the propagation energy values ​​from all directed edges pointing to downstream nodes of the node, and summarizes these propagation energy values ​​as the second-dimensional quantification value. The process of obtaining the propagation energy value is as follows: The system performs a quantitative analysis of the workstation physical association and temporal characteristics of the abnormal event node. Based on the material transfer distance between the workstation where the node is located and its downstream workstations, the energy transfer path length, and the response delay time of the process parameters between them, a transmission attenuation coefficient is determined. The transmission attenuation coefficient reflects the physical law that the impact of upstream anomalies on downstream decreases with distance and delay. The system quantifies and evaluates the parameter deviation amplitude and duration of the abnormal event node, multiplying the parameter deviation amplitude by the duration to obtain the anomaly excitation intensity, which characterizes the total energy of the node's own anomaly. Subsequently, the system weightedly fuses the anomaly excitation intensity and the transmission attenuation coefficient, i.e., multiplies the two to obtain the propagation energy value from the node to the downstream node. For cases where the node has multiple downstream nodes, the system calculates the corresponding propagation energy value for each downstream node separately, and finally sums the propagation energy values ​​of all downstream nodes to obtain the second-dimensional quantification value of the node.

[0043] After obtaining the first and second dimension quantization values, the system performs coupled quantization on these two dimensions. Specifically, the first dimension quantization value is multiplied by the second dimension quantization value to obtain the node's overall impact score. This coupling process reflects both the severity of the node's own anomaly and its ability to propagate anomalies downstream in the same value, focusing not only on the severity of the node itself but also on whether the node plays a key role in anomaly propagation.

[0044] In the formula, The overall impact score is as follows. For nodes The maximum deviation of the parameters corresponding to the abnormal event. The maximum parameter deviation of all nodes in the directed backtracking subgraph. For nodes The duration corresponding to the abnormal event, The preset reference duration, It is the natural logarithm. For nodes The set of all downstream nodes pointed to in the directed backtracking subgraph, For the node To the node The energy value propagated on the directed edge, The first dimension is the maximum propagation energy value on all directed edges in the directed backtracking subgraph, where the first dimension is given by the formula. The first dimension quantifies the intensity of a node's own anomalous events; the greater the deviation in parameters and the longer the duration, the higher the weight of the anomalous event. Furthermore, a logarithmic function smooths out excessive amplification caused by prolonged durations. The second dimension is derived from the formula... The dimension represents the total energy of a node anomaly propagating to downstream workstations. The greater the propagation energy value and the more downstream nodes affected, the higher the propagation impact weight.

[0045] The maximum deviation is derived from extracting the parameter values ​​of all sampling points of the abnormal event node in the directed backtracking subgraph during its abnormal period, and selecting the difference with the most serious deviation from the process specification limit as the maximum deviation of the node.

[0046] The maximum parameter deviation is determined by comparing the maximum parameter deviation of all nodes in the subgraph and taking the maximum value as the normalization benchmark. The anomaly duration is the time between the start and end of the anomaly at the node. The preset reference duration is a fixed value used to make the duration dimensionless.

[0047] The propagation energy value is derived from a weighted fusion of the propagation attenuation coefficient and the anomalous excitation intensity, obtained through quantitative analysis of the physical correlation and temporal characteristics of the workstations at the anomalous event nodes. The propagation attenuation coefficient is determined based on the material transfer distance between the workstation where the node is located and its downstream workstations, the length of the energy transfer path, and the response delay time of the process parameters; the greater the distance and the greater the delay, the smaller the attenuation coefficient. The anomalous excitation intensity is obtained by multiplying the maximum deviation of the node's parameters by the duration of the anomalous event, reflecting the total energy of the anomalous event itself at that node. Multiplying the two values ​​yields the propagation energy value from the node to the downstream node, which quantifies the degree of impact of the upstream anomalous event on the downstream.

[0048] The maximum propagation energy value on all directed edges in the directed backtracking subgraph is derived by comparing the propagation energy values ​​of all directed edges in the subgraph and taking the maximum value as the normalization benchmark.

[0049] The significance of the formula lies in integrating the severity of an anomaly of an event node itself with its ability to propagate anomalies downstream into a single comprehensive impact score through an integrated mathematical expression. The first part of the formula normalizes the maximum deviation of the parameters and then multiplies it by the logarithmic smoothing function of the duration to obtain the node's own anomaly weight. The role of the logarithmic function is to gradually reduce the influence of the duration on the weight as the duration increases, avoiding excessive weight amplification due to excessively long durations.

[0050] The second part of the formula normalizes and sums the propagation energy values ​​on all directed edges pointing downstream from the node to obtain the node's propagation influence weight. This weight reflects the total energy of the node's anomaly propagating to downstream workstations; the larger the propagation energy value and the more downstream nodes affected, the higher the propagation influence weight. The product of the two parts allows the comprehensive influence score to simultaneously consider the severity of the node's own anomaly and the node's crucial role in anomaly propagation.

[0051] The formula shows that the larger the deviation of a node's own parameters and the longer the duration of the anomaly, the larger the value of the first part, and the higher the node's own anomaly weight. Conversely, the larger the propagation energy value of a node pointing downstream and the more downstream nodes it affects, the larger the value of the second part, and the higher the node's propagation influence weight. The comprehensive influence score increases with the values ​​of both dimensions. When both the node's own anomaly weight and propagation influence weight are at high levels, the comprehensive influence score will significantly increase, making these nodes rank higher in the sorting. Conversely, when either dimension is low, the comprehensive influence score will be significantly suppressed, thus preventing nodes with prominent single-dimensional anomalies from being misjudged as root causes. This trend ensures that the finally selected suspected root cause nodes not only have significant self-abnormal characteristics but also play an important transmission role in the anomaly propagation path. After performing the above calculations on all nodes in the directed backtracking subgraph, the system obtains a comprehensive impact score for each node. Subsequently, the system sorts all nodes in descending order of their comprehensive impact scores and selects the top few nodes from the sorted results to form a sequence of suspected root causes for the production line. This sequence represents the set of candidate nodes most likely to be the source of the anomaly in the anomaly propagation path, providing a screening basis for subsequent precise localization.

[0052] The beneficial effect is that by constructing a two-dimensional root cause quantification judgment mechanism, the severity of the abnormality of the abnormal event node in the directed backtracking subgraph is integrated with the node's ability to propagate the abnormality downstream to form a comprehensive impact score. Suspicious root cause sequences are then selected based on the score ranking, thereby realizing a dual consideration of abnormal nodes. This effectively avoids the root cause misjudgment caused by traditional single-dimensional ranking and significantly improves the accuracy and robustness of root cause localization.

[0053] A5: Perform multi-path hybrid weighted root cause quantization localization on the temporal feature entities of the suspected root cause sequence to obtain the root cause localization result of the production line; In this embodiment of the invention, the step of performing multi-path hybrid weighted root cause quantization localization on the temporal feature entities of the suspected root cause sequence to obtain the root cause localization result of the production line includes: The candidate nodes in the suspected root cause sequence are traced and aggregated along the entire path to obtain the path set of the suspected root cause sequence. The path weights of the path set are obtained by accumulating the comprehensive influence scores of the nodes on the paths in the path set. The extreme value of the path weight is used as the multi-path mixing weight of the candidate node; The nodes in the suspected root cause sequence are reordered according to the multi-path mixed weight to obtain the root cause localization result of the production line.

[0054] The system performs full-path tracing and aggregation of candidate nodes in a suspected root cause sequence. Specifically, for each candidate node in the suspected root cause sequence, the system extracts all possible propagation paths from the time-series production knowledge graph, starting from the candidate node and ending at the precise trigger point. Each path consists of a series of nodes and directed edges connected sequentially, reflecting all possible routes of an anomaly propagating downstream along the process flow from the candidate node until the final anomaly. The system aggregates all propagation paths corresponding to all candidate nodes to form a path set for the suspected root cause sequence.

[0055] The system performs a comprehensive impact score accumulation operation on each path in the path set. For each path, the system iterates through the nodes on that path, retrieves the comprehensive impact score of that node calculated in previous steps, multiplies the comprehensive impact scores of all nodes on the path, and uses the product as the path weight. Through this accumulation method, the path weight reflects the strength of the combined effect of all abnormal nodes on the entire path; the longer the path and the higher the node scores, the greater the path weight.

[0056] The system uses the extreme values ​​of path weights as the multi-path mixed weights for candidate nodes. For each candidate node, the system identifies the path with the largest weight among all paths corresponding to that node and uses this largest path weight as the multi-path mixed weight for that candidate node. This operation ensures that the multi-path mixed weight for each candidate node represents the combined strength of the path that best reflects the dominant role of that node among all possible anomaly propagation paths.

[0057] The system reorders all nodes in the suspected root cause sequence according to their multi-path mixed weights from largest to smallest, meaning nodes with higher weights rank higher. After reordering, the node ranked first in the sequence is the node with the highest multi-path mixed weight, and the system identifies this node as the final root cause localization result. This result represents the anomaly source node with the strongest overall influence and the dominant role in multiple propagation paths within the entire anomaly propagation network.

[0058] The beneficial effect is that by performing full-path source tracing and aggregation on each candidate node in the suspected root cause sequence, all possible propagation paths from the node to the precise trigger point are extracted, and the product of the comprehensive influence scores of the nodes on each path is calculated as the path weight. Then, the maximum value of the path weight is used as the multi-path mixed weight of the node. Finally, the nodes are reordered according to the weight and the one with the highest weight is selected as the root cause localization result. In this way, the source node with the strongest comprehensive influence and the dominant role is identified among the multiple possible paths of abnormal propagation, which significantly improves the comprehensiveness and accuracy of root cause localization.

[0059] A6: Based on the full-chain related data of the root cause localization results, output the intelligent traceability analysis report of the production line.

[0060] In this embodiment of the invention, the step of outputting an intelligent traceability analysis report for the production line based on the end-to-end related data of the root cause localization result includes: Based on the time-series production knowledge graph, multi-dimensional information parsing is performed on the workstation nodes of the root cause localization results to obtain the full-link related data of the workstation nodes; The propagation link of the directed backtracking subgraph is reconstructed to obtain the complete propagation path of the directed backtracking subgraph; Based on the relevant data across the entire chain, the root cause localization results and the relevant parameters of the complete propagation path are structurally integrated to obtain the traceability analysis report of the production line, and an intelligent traceability analysis report of the production line is output.

[0061] Based on the root cause localization results and related data across the entire supply chain, the system outputs an intelligent traceability analysis report for the production line.

[0062] Based on a time-series production knowledge graph, multi-dimensional information parsing is performed on the workstation nodes corresponding to the root cause localization results to obtain the full-link related data of that workstation node. After locating the workstation node corresponding to the root cause localization result, the system extracts all information stored in that node from the time-series production knowledge graph, including the workstation node number, all parameter values ​​of the node during the abnormal event, the maximum parameter deviation, the duration of the abnormality, and the start and end times of the abnormality. Simultaneously, the system obtains all relevant operational parameters of that workstation during the entire abnormality propagation period from the node's time-series data sequence, and summarizes this information into the full-link related data of that workstation node.

[0063] The system reconstructs the propagation chain of the directed backtracking subgraph to obtain the complete propagation path. The system traverses all directed edges in the directed backtracking subgraph from the root cause localization node to the precise trigger point node. It sequentially reads each node and the directed edges connecting them, following an upstream-to-downstream order, and connects these nodes in chronological order to form a complete propagation path starting from the root cause node, passing through intermediate anomaly nodes, and finally reaching the precise trigger point. This path fully demonstrates each step of the anomaly's transmission process from its source to its final manifestation.

[0064] Based on end-to-end relevant data, the system structurally integrates root cause localization results and relevant parameters of the complete propagation path to generate a traceability analysis report for the production line. The system extracts core information corresponding to the root cause localization results, such as workstation number, anomaly occurrence time, parameter deviation magnitude, and anomaly duration, and associates this information with nodes and their anomaly characteristics along the complete propagation path. This data is then organized according to a preset report format to form a traceability analysis report containing the root cause location, propagation chain, anomaly intensity description, and relevant time information. After the report is generated, it is output to the production management system's display terminal for operators to view, and can also be stored in a database for subsequent statistical analysis, thus completing the output of an intelligent traceability analysis report for the production line.

[0065] The beneficial effects are as follows: by performing multi-dimensional information analysis on the workstation nodes corresponding to the root cause localization results, extracting relevant data such as their number, abnormal time period, and deviation magnitude, and reconstructing the propagation link of the directed backtracking subgraph, a complete propagation path is formed from the root cause node through the intermediate abnormal node to the precise trigger point. Finally, the root cause localization results, the complete propagation path, and related parameters are structurally integrated to generate and output a traceability analysis report containing root cause location, propagation chain, abnormality intensity, and time information. This transforms complex technical analysis results into intuitive, understandable, and traceable decision support information, significantly improving the interpretability and engineering applicability of abnormality diagnosis results.

[0066] like Figure 2The diagram shown is a functional block diagram of a production process anomaly root cause tracing system based on time-series knowledge graphs provided in an embodiment of the present invention.

[0067] The production process anomaly root cause tracing system 100 based on time-series knowledge graphs described in this invention can be installed in an electronic device. Depending on the functions implemented, the production process anomaly root cause tracing system 100 may include a time-series knowledge graph construction module 101, an anomaly capture and classification module 102, a reverse causal backtracking module 103, a dual-dimensional quantification judgment module 104, a multi-path hybrid positioning module 105, and a tracing analysis report module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0068] In this embodiment, the functions of each module / unit are as follows: The time-series graph construction module 101 is used to acquire time-series data from multiple sources of sensors in the production line, and to perform hierarchical dynamic update graph modeling on the time-series data to obtain the time-series production knowledge graph of the production line. The anomaly capture and classification module 102 is used to capture and classify the abnormal events of the production line based on preset anomaly identification rules, obtain the local abnormal events of the production line, and use the local abnormal events as the precise trigger points of the production line. The reverse causal backtracking module 103 is used to perform reverse causal backtracking on the temporal feature entities of the precise trigger point based on the temporal production knowledge graph, and obtain the directed backtracking subgraph of the production line. The dual-dimensional quantization judgment module 104 is used to perform dual-dimensional root cause quantization judgment on the causal path of the directed backtracking subgraph to obtain the suspected root cause sequence of the production line. The multi-path hybrid localization module 105 is used to perform multi-path hybrid weighted root cause quantization localization on the temporal feature entities of the suspected root cause sequence to obtain the root cause localization result. The traceability analysis report module 106 is used to output an intelligent traceability analysis report of the production line based on the full-link related data of the root cause localization results.

[0069] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0070] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0071] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0072] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0073] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for tracing the root causes of production process anomalies based on time-series knowledge graphs, characterized in that, The method includes: A1: Acquire time-series data from multiple sources of sensors in the production line, and perform hierarchical dynamic update graph modeling on the time-series data to obtain the time-series production knowledge graph of the production line; A2: Based on preset anomaly identification rules, the abnormal events of the production line are captured and classified to obtain the local abnormal events of the production line, and the local abnormal events are used as the precise trigger points of the production line. A3: Based on the time-series production knowledge graph, reverse causal backtracking is performed on the time-series feature entities of the precise trigger point to obtain the directed backtracking subgraph of the production line; A4: Perform two-dimensional root cause quantification on the causal path of the directed backtracking subgraph to obtain the suspected root cause sequence of the production line; A5: Perform multi-path hybrid weighted root cause quantization localization on the temporal feature entities of the suspected root cause sequence to obtain the root cause localization result of the production line; A6: Based on the full-chain related data of the root cause localization results, output the intelligent traceability analysis report of the production line.

2. The method for tracing the root causes of production process anomalies based on time-series knowledge graphs as described in claim 1, characterized in that, The process of acquiring time-series data from multiple sensors in the production line and performing hierarchical dynamic update graph modeling on the time-series data to obtain a time-series production knowledge graph of the production line includes: Based on the process flow design document of the production line, a topological model of the material and energy flow relationship of the workstations in the production line is performed to obtain the static ecological production topology map of the production line. Based on the multi-source sensors in the production line, the time-series data of the multi-source sensors are labeled with workstation spatiotemporal tags to obtain the time-series data stream of the production line; Based on the order of the timestamps in the workstation spatiotemporal labels, the data in the time-series data stream is matched and entered into the workstation nodes of the static ecological production topology map to obtain the time-series data sequence of the workstation nodes. The time-series data sequence is dynamically updated using a sliding window to obtain the time-series production knowledge graph of the production line.

3. The method for tracing the root causes of production process anomalies based on time-series knowledge graphs as described in claim 1, characterized in that, The method of capturing and classifying abnormal events on the production line based on preset anomaly identification rules to obtain local anomaly events on the production line, and using these local anomaly events as precise trigger points for the production line, includes: Based on preset anomaly identification rules, anomaly detection and judgment are performed on the workstation parameters in the time-series production knowledge graph to obtain the abnormal events of the production line. The deviation of parameters from the abnormal events is assessed to determine the level of abnormal events on the production line. The hierarchical events are filtered and located to obtain local abnormal events of the production line, and these local abnormal events are used as precise trigger points for the production line.

4. The method for tracing the root causes of production process anomalies based on time-series knowledge graphs as described in claim 1, characterized in that, The step of performing reverse causal backtracking on the temporal feature entities of the precise trigger point based on the temporal production knowledge graph to obtain the directed backtracking subgraph of the production line includes: Queries the upstream associated entities of the workstation entity in the precise trigger point; Perform anomaly detection on the historical time series data of the upstream associated entity to obtain the anomaly detection result of the workstation entity itself; Based on the results of the self-abnormal event determination, the upstream associated entities are subjected to abnormal effective screening to obtain the set of effective upstream nodes of the workstation entity, and the associated edges of the effective upstream nodes in the set of effective upstream nodes are retained. The effective upstream nodes are iteratively backtracked until no effective upstream nodes are available, thus obtaining the abnormal link tracing result of the production line; Based on the abnormal link tracing results, the effective upstream nodes and the associated edges are constructed to obtain the directed backtracking subgraph of the production line.

5. The method for tracing the root causes of production process anomalies based on time-series knowledge graphs as described in claim 4, characterized in that, The step of performing anomaly detection on the historical time-series data of the upstream associated entity to obtain the anomaly event detection result of the workstation entity itself includes: Based on the time-series production knowledge graph, obtain the time-series data of the upstream related entities; The time-series data is subjected to time-series anomaly detection to obtain the self-abnormal event determination result of the workstation entity; If the time series data has a period of parameter change, then the result of the self-abnormal event determination is that there is a self-abnormal event; If the time series data does not have a period of parameter change, then the result of the self-abnormal event determination is that the abnormality condition is not met.

6. The method for tracing the root causes of production process anomalies based on time-series knowledge graphs as described in claim 4, characterized in that, The step of performing two-dimensional root cause quantification on the causal path of the directed backtracking subgraph to obtain the suspected root cause sequence of the production line includes: Obtain the maximum value of parameter deviation and the duration of anomalies of the abnormal event nodes in the directed backtracking subgraph, and use them as the first dimension quantization value; Obtain the propagation energy value pointing to the downstream node in the abnormal event node, and use it as the second dimension quantization value; The first dimension quantization value and the second dimension quantization value are coupled and quantized to obtain the comprehensive impact score of the production line; The formula for calculating the comprehensive impact score is as follows: ； in, The overall impact score is as follows. For nodes The maximum deviation of the parameters corresponding to the abnormal event. The maximum parameter deviation of all nodes in the directed backtracking subgraph. For nodes The duration corresponding to the abnormal event, The preset reference duration, It is the natural logarithm. For nodes The set of all downstream nodes pointed to in the directed backtracking subgraph, For the node To the node The energy value propagated on the directed edge, The maximum propagation energy value on all directed edges in the directed backtracking subgraph; The comprehensive impact scores are sorted and filtered to obtain the suspected root cause sequence of the production line.

7. The method for tracing the root causes of production process anomalies based on time-series knowledge graphs as described in claim 6, characterized in that, The step of obtaining the propagation energy value pointing to the downstream node from the abnormal event node includes: The workstation physical association and timing characteristics of the abnormal event nodes are quantitatively analyzed to obtain the transmission attenuation coefficient of the abnormal event nodes. The magnitude and duration of parameter deviation of the abnormal event node are quantitatively evaluated to obtain the abnormal excitation intensity of the abnormal event node; The abnormal excitation intensity and the transmission attenuation coefficient are weighted and fused to obtain the propagation energy value of the abnormal event node pointing to the downstream node.

8. The method for tracing the root causes of production process anomalies based on time-series knowledge graphs as described in claim 6, characterized in that, The step of performing multi-path hybrid weighted root cause quantization localization on the temporal feature entities of the suspected root cause sequence to obtain the root cause localization result of the production line includes: The candidate nodes in the suspected root cause sequence are traced and aggregated along the entire path to obtain the path set of the suspected root cause sequence. The path weights of the path set are obtained by accumulating the comprehensive influence scores of the nodes on the paths in the path set. The extreme value of the path weight is used as the multi-path mixing weight of the candidate node; The nodes in the suspected root cause sequence are reordered according to the multi-path mixed weight to obtain the root cause localization result of the production line.

9. The method for tracing the root causes of production process anomalies based on time-series knowledge graphs as described in claim 1, characterized in that, The step of outputting an intelligent traceability analysis report for the production line based on the end-to-end related data of the root cause localization results includes: Based on the time-series production knowledge graph, multi-dimensional information parsing is performed on the workstation nodes of the root cause localization results to obtain the full-link related data of the workstation nodes; The propagation link of the directed backtracking subgraph is reconstructed to obtain the complete propagation path of the directed backtracking subgraph; Based on the relevant data across the entire supply chain, the root cause localization results and the relevant parameters of the complete propagation path are structurally integrated to obtain a traceability analysis report for the production line, and an intelligent traceability analysis report for the production line is output.

10. A production process anomaly root cause tracing system based on time-series knowledge graphs, characterized in that, The system for implementing the production process anomaly root cause tracing method based on time-series knowledge graph as described in claim 1, the system comprising: The time-series graph construction module is used to acquire time-series data from multiple sources of sensors in the production line, and to perform hierarchical dynamic updating graph modeling on the time-series data to obtain the time-series production knowledge graph of the production line. An anomaly capture and classification module is used to capture and classify abnormal events of the production line based on preset anomaly identification rules, obtain local anomaly events of the production line, and use the local anomaly events as precise trigger points of the production line. The reverse causal backtracking module is used to perform reverse causal backtracking on the temporal feature entities of the precise trigger point based on the temporal production knowledge graph, and obtain the directed backtracking subgraph of the production line; The dual-dimensional quantization judgment module is used to perform dual-dimensional root cause quantization judgment on the causal path of the directed backtracking subgraph to obtain the suspected root cause sequence of the production line. A multi-path hybrid localization module is used to perform multi-path hybrid weighted root cause quantization localization on the temporal feature entities of the suspected root cause sequence to obtain the root cause localization result. The traceability analysis report module is used to output an intelligent traceability analysis report for the production line based on the full-link related data of the root cause localization results.

Images