Power distribution control cabinet fault response method and device

Abstract

The invention provides a power distribution control cabinet fault response method and device, and relates to the technical field of electrical automation control, and the key points of the technical scheme are that a weighted directed graph model representing the relation of internal components of a power distribution control cabinet is constructed in advance; mapping component fault information predicted by the digital twin system to attributes of corresponding nodes in the weighted directed graph model; identifying a fault path or a fault cluster satisfying a predetermined condition; for each identified fault path or fault cluster, calculating a composite risk index according to the prediction probability of the nodes included in the fault path or fault cluster, the weight of the connection edge and the severity of the influence on the system; and generating early warning information according to the composite risk index. According to the power distribution control cabinet fault response method and device provided by the invention, a potential fault propagation path or a fault aggregation area can be identified based on physical, electrical and influence relationships among components, and a composite risk formed by a plurality of potential fault combinations is quantified, so that a preventive maintenance decision is guided.

Description

Technical Field

[0001] The present application relates to the field of electrical automation control technology, and in particular to a method and device for responding to a fault in a power distribution control cabinet. Background Art

[0002] In industrial environments like continuous chemical production, where power supply continuity is paramount, the stable and reliable operation of the power distribution network, particularly critical distribution control cabinets such as feeder control cabinets and motor control centers (MCCs), is crucial for ensuring the safe and efficient operation of core production equipment. To improve power supply reliability and optimize maintenance strategies, advanced predictive maintenance technologies such as digital twins have been gradually applied to distribution systems. Existing technologies typically build digital twins for key components within distribution control cabinets (such as molded case circuit breakers, vacuum contactors, thermal relays, and control transformers). These models encompass the electrical and thermodynamic characteristics of these components and their topological connectivity. Sensors deployed within the cabinets (such as temperature, current, voltage, vibration, and number of switching cycles) collect real-time operational data to drive state updates in the digital twin model. Based on degradation models for individual components (such as contactor vacuum loss, circuit breaker contact wear, and relay insulation aging), these models predict single failure modes and their likely occurrence times, providing maintenance recommendations to operations personnel.

[0003] However, in practical applications within continuous chemical production enterprises, particularly within distribution control cabinets (particularly MCC cabinets) with dense component density and complex internal environments, current digital twin systems, which primarily rely on single-component fault prediction, have significant shortcomings. When the system simultaneously predicts potential fault risks for multiple different components, even if the predicted probability of each individual risk is low, these potential faults are often not independent of each other. Due to the close physical proximity of components within the cabinet (such as the adjacent main circuit contactors and thermal relays within an MCC cabinet), significant thermal conduction occurs, and an abnormal temperature rise in one component can accelerate the aging of adjacent components. Furthermore, these components are electrically closely connected via shared busbars or feeders. Electrical disturbances in one circuit (such as harmonics, overloads, and voltage sags) can affect components in other circuits through electrical pathways, increasing operational stress or the risk of malfunction. Existing technologies generally lack an effective mechanism to systematically identify, model, and quantify the coupling relationships and impact propagation paths between these potential faults, which arise from physical proximity (such as thermal coupling) and electrical connections (such as stress transfer through shared pathways). Therefore, existing systems struggle to accurately assess system-level compound risks, which are composed of multiple seemingly independent, low-probability predicted events that may occur along specific paths or clustered within specific areas. This is particularly true of the potential threats these compound risks pose to the continuity of power supply to core loads such as critical chemical reactors. This neglect of the coupling effects between components and the fault propagation paths makes it difficult for operations and maintenance personnel, faced with a series of scattered single alerts, to identify high-priority fault chains or clusters caused by multi-factor coupling that truly threaten the continuous production process. Consequently, the formulated preventive maintenance plans may not effectively avoid unplanned downtime of critical equipment and the resulting significant production losses and potential safety hazards. Therefore, there is an urgent need for a distribution control cabinet fault response method that can comprehensively consider the multiple relationships between components, identify potential fault propagation paths, and assess compound risks.

[0004] In view of the above problems, the existing technology is in urgent need of improvement. Summary of the Invention

[0005] The purpose of this application is to provide a distribution control cabinet fault response method and device, which can identify potential fault propagation paths or fault aggregation areas based on the physical, electrical and influence relationships between components, and quantify the compound risks formed by the combination of multiple potential faults, thereby guiding preventive maintenance decisions.

[0006] In the first aspect, the present application provides a method for responding to a fault in a power distribution control cabinet, and the technical solution is as follows:

[0007] Pre-building a weighted directed graph model representing the relationships between components within the power distribution control cabinet, wherein the weighted directed graph model includes nodes representing physical components and their failure modes, and weighted edges representing physical proximity relationships, electrical connection relationships, and potential impact relationships;

[0008] Mapping component fault information predicted by the digital twin system to attributes of corresponding nodes in the weighted directed graph model;

[0009] Based on the weighted directed graph model and the attributes of the mapped corresponding nodes, executing a graph analysis algorithm to identify fault paths or fault clusters that meet predetermined conditions;

[0010] For each identified fault path or fault cluster, a composite risk index is calculated based on the predicted probability of the nodes it contains, the weight of the connecting edges, and the severity of the impact on the system;

[0011] Early warning information is generated based on the composite risk index to guide preventive maintenance decisions.

[0012] Furthermore, in the present application, the step of executing a graph analysis algorithm based on the weighted directed graph model and the attributes of the mapped corresponding nodes to identify a fault path or fault cluster that meets a predetermined condition includes:

[0013] Construct a fault path index structure to record the set of nodes and their associated edges contained in each identified fault path or fault cluster;

[0014] Monitoring changes in node attributes in the weighted directed graph model and determining a set of nodes that have undergone changes;

[0015] Based on the changed node set, query the fault path index structure to obtain a list of affected fault paths or fault clusters;

[0016] Performing local graph analysis only on the paths or clusters in the affected fault paths or fault cluster lists, and recalculating their risk states;

[0017] If it is detected that a new node attribute change may form a new fault path or fault cluster, incremental graph analysis is performed in the local range of the changed area to identify the newly added fault path or fault cluster.

[0018] Furthermore, in the present application, the step of calculating the composite risk index for each identified fault path or fault cluster based on the predicted probability of the included nodes, the weight of the connecting edges, and the severity of the impact on the system includes:

[0019] Detect load condition change events and determine the difference between the conditions before and after the change;

[0020] According to the difference in the working conditions, determine whether it is necessary to recalculate the composite risk index;

[0021] If recalculation is required, obtain the current load operating parameters and their corresponding influence coefficients;

[0022] Adjusting the predicted probability values ​​of nodes in the fault path or fault cluster according to the load operating parameters and the influence coefficient;

[0023] The composite risk index is recalculated based on the adjusted node prediction probability value, the weight of the connecting edge, and the severity of the system impact under the current working conditions.

[0024] Furthermore, in the present application, the step of detecting a load condition change event and determining the difference between the conditions before and after the change includes:

[0025] Obtaining time series data of electrical parameters of the load within a preset time window, wherein the electrical parameters include current, voltage, power, and power factor;

[0026] Calculating the duration and magnitude of the load change based on the electrical parameter time series data;

[0027] Obtain current process stage identification and process conversion plan information from the chemical process control system;

[0028] Determining whether the current load change is consistent with the expected process transition based on the process stage identifier and process transition plan information;

[0029] If the load change is consistent with the expected process transition and the duration exceeds a first threshold, identifying it as a continuous load change;

[0030] If the load change does not conform to the expected process transition or the duration is less than a second threshold, identifying it as a temporary load change;

[0031] The working condition difference is calculated according to the type and magnitude of the load change and the sensitivity coefficients of the corresponding components.

[0032] Furthermore, in the present application, the step of calculating the composite risk index for each identified fault path or fault cluster based on the predicted probability of the included nodes, the weight of the connecting edges, and the severity of the impact on the system includes:

[0033] Obtain the current process stage identifier and its corresponding key parameter threshold value from the chemical process control system;

[0034] Establish a mapping table between process stages and load importance, and determine the importance level of each load based on the current process stage identifier;

[0035] Obtain backup power supply path information and switching time parameters for each load;

[0036] Based on the load importance level, backup power supply path information and switching time parameters, a system impact severity assessment matrix for the current process stage is constructed;

[0037] For each fault path or fault cluster, identify the set of loads that it potentially affects;

[0038] Calculating the comprehensive impact coefficient of the load set at the current process stage according to the system impact severity assessment matrix;

[0039] The composite risk index is calculated by combining the predicted probabilities of the nodes in the fault path or fault cluster, the weights of the connecting edges, and the comprehensive impact coefficient.

[0040] Furthermore, in the present application, the step of calculating the composite risk index by combining the predicted probabilities of the nodes in the fault path or fault cluster, the weights of the connecting edges, and the comprehensive impact coefficient includes:

[0041] Establish a three-tier scoring matrix, corresponding to the technical risk layer, production impact layer, and security risk layer;

[0042] Mapping the predicted probability of nodes in the fault path or fault cluster to a technical risk layer and calculating a technical risk score;

[0043] Mapping the comprehensive impact coefficient of the load set at the current process stage to the production impact layer and calculating the production impact score;

[0044] Based on the potential failure modes of the fault paths or fault clusters, and in combination with chemical safety regulations, the types and severity of possible safety incidents are assessed, and a safety risk score is calculated;

[0045] According to the characteristics of the current process stage, determine the weight coefficients of the three levels of technical risk, production impact and safety risk;

[0046] The technical risk score, production impact score and safety risk score are weighted according to corresponding weight coefficients to obtain an initial composite risk value;

[0047] The initial composite risk value is converted into a standardized composite risk index through a nonlinear mapping function, wherein the nonlinear mapping function is pre-calibrated according to historical failure data and expert evaluation results.

[0048] Furthermore, in the present application, the step of adjusting the predicted probability value of the node in the fault path or fault cluster according to the load operating parameter and the influence coefficient includes:

[0049] Obtaining time series data of load operating parameters within a preset time window;

[0050] Calculating a rate of change of the load operating parameter and classifying the rate of change into multiple levels;

[0051] Establish a corresponding relationship table between the rate of change of load operating parameters and the stress increment of components;

[0052] According to the change rate level of the load operating parameter, query the corresponding component stress increment from the corresponding relationship table;

[0053] Multiplying the component stress increment by the influence coefficient to obtain an adjustment value of the node prediction probability;

[0054] The original node prediction probability is calculated with the adjustment value to obtain an adjusted node prediction probability value.

[0055] Furthermore, in the present application, the step of calculating the operating condition difference according to the type and magnitude of the load change and the sensitivity coefficient of the corresponding component includes:

[0056] Obtaining preset reference baseline data, wherein the preset reference baseline data includes standard operating parameters of various loads under normal operating conditions;

[0057] Calculating a deviation value of the current load change relative to the preset reference baseline data;

[0058] Setting a weight coefficient according to the type of load change, wherein the weight coefficient of a continuous load change is higher than the weight coefficient of a temporary load change;

[0059] Multiplying the deviation value by the weight coefficient to obtain a weighted deviation value;

[0060] Multiplying the weighted deviation value by the sensitivity coefficient of the corresponding component to obtain a component response value;

[0061] Accumulate the response values ​​of all affected components to obtain a cumulative response value;

[0062] According to a preset normalization function, the cumulative response value is converted into a standardized working condition difference within the range of 0-100;

[0063] According to the standardized working condition difference, the corresponding warning level is determined, wherein the warning level includes four levels: normal, attention, warning and emergency, and each warning level corresponds to a different difference threshold range.

[0064] Furthermore, in the present application, the step of mapping the component fault information predicted by the digital twin system to the attributes of the corresponding nodes in the weighted directed graph model includes:

[0065] Establishing standardized mapping rules between component fault information and graph model node attributes, wherein the standardized mapping rules include attribute field definitions, value range constraints, and outlier processing strategies;

[0066] Receiving component failure prediction information output by the digital twin system, wherein the component failure prediction information includes component identification, failure mode, prediction probability, and prediction time window;

[0067] Detect missing fields or abnormal values ​​in the component fault prediction information and determine an information completeness score based on preset data completeness requirements;

[0068] When the information completeness score is lower than a first threshold, retrieving historical records under similar working conditions from a historical database to extract supplementary information;

[0069] When the information completeness score is lower than a second threshold and valid supplementary information cannot be obtained from the historical database, generating an estimated value based on the component type and its location in the power distribution system;

[0070] Converting the complete or supplemented component fault prediction information into attribute values ​​of graph model nodes according to the standardized mapping rules;

[0071] Record the information source and uncertainty indicators of each mapping operation for credibility analysis in subsequent risk assessment.

[0072] Secondly, this application also proposes a distribution control cabinet fault response device for digital twin predictive maintenance of manufacturing enterprises. The system includes:

[0073] A modeling module is used to pre-build a weighted directed graph model representing the relationships between components within the power distribution control cabinet, wherein the weighted directed graph model includes nodes representing physical components and their failure modes, and weighted edges representing physical proximity relationships, electrical connection relationships, and potential impact relationships;

[0074] A mapping module, configured to map component fault information predicted by the digital twin system to attributes of corresponding nodes in the weighted directed graph model;

[0075] An analysis module, configured to execute a graph analysis algorithm based on the weighted directed graph model and the attributes of the mapped corresponding nodes to identify a fault path or a fault cluster that meets predetermined conditions;

[0076] A calculation module is used to calculate the composite risk index for each identified fault path or fault cluster based on the predicted probability of the nodes it contains, the weight of the connecting edges, and the severity of the impact on the system;

[0077] The early warning module is used to generate early warning information according to the composite risk index to guide preventive maintenance decisions.

[0078] From the above, it can be seen that the present application provides a distribution control cabinet fault response method and device, which maps and predicts fault information by constructing a weighted directed graph model that includes physical proximity, electrical connection and potential impact relationships, and uses a graph analysis algorithm to identify fault paths or clusters, and calculates a composite risk index that takes into account node probabilities, edge weights and system impacts. It solves the problem that the existing technology is difficult to evaluate composite risks under the interaction of multiple components, and has the ability to identify potential fault propagation paths or fault aggregation areas based on the physical, electrical and impact relationships between components, and quantify the composite risks formed by multiple potential fault combinations, thereby guiding preventive maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 A flow chart of a power distribution control cabinet fault response method provided in this application.

[0080] Figure 2 This is a schematic diagram of the structure of a power distribution control cabinet fault response system provided in this application.

[0081] Figure 3 This is a system block diagram of a power distribution control cabinet fault response system provided in this application.

[0082] In the figure: 210, modeling module; 220, mapping module; 230, analysis module; 240, calculation module; 250, early warning module. DETAILED DESCRIPTION

[0083] The technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments. The components of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for which protection is claimed, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0084] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and should not be understood as indicating or implying relative importance.

[0085] Please refer to Figure 1 This application proposes a distribution control cabinet fault response method, which is applied to the digital twin predictive maintenance system of the manufacturing enterprise, including:

[0086] S110. Pre-constructing a weighted directed graph model representing relationships between components within the power distribution control cabinet, wherein the weighted directed graph model includes nodes representing physical components and their failure modes, and weighted edges representing physical proximity relationships, electrical connection relationships, and potential impact relationships;

[0087] S120, mapping component fault information predicted by the digital twin system to attributes of corresponding nodes in a weighted directed graph model;

[0088] S130 , executing a graph analysis algorithm based on the weighted directed graph model and the attributes of the mapped corresponding nodes to identify a fault path or fault cluster that meets a predetermined condition;

[0089] S140. For each identified fault path or fault cluster, calculate a composite risk index based on the predicted probability of the included nodes, the weight of the connecting edges, and the severity of the impact on the system;

[0090] S150. Generate early warning information based on the composite risk index to guide preventive maintenance decisions.

[0091] When constructing a weighted directed graph model, nodes are designed to represent specific physical components (e.g., circuit breakers, contactors) and their possible failure modes (e.g., refusal to operate, malfunction, insulation degradation). Edges are assigned weights to quantify the strength of the relationship. For example, the weight of a physical proximity edge can be set based on thermal conductivity or distance, the weight of an electrical connection edge can be set based on impedance or shared load ratio, and the weight of a potential impact edge (e.g., control loop dependency) can be set based on logical determinism.

[0092] Mapping prediction information to node attributes means assigning information such as the single component failure prediction probability and prediction time window output by the digital twin system to the data field of the corresponding node in the graph.

[0093] Execute graph analysis algorithms, such as node screening based on probability thresholds combined with path search algorithms (such as finding paths between high-probability nodes connected by strongly correlated edges) or community discovery algorithms (such as identifying groups of high-probability nodes that have close mutual influence) to identify fault paths or fault clusters.

[0094] The composite risk index is calculated by combining the predicted probability of nodes within a path or cluster, the weight of the connecting edges, and the severity of the consequences of the failure of the path or cluster on the downstream load or system operation (which can be pre-assessed or dynamically obtained), and then calculated using a preset mathematical model or scoring rules.

[0095] The early warning information is generated by comparing the calculated composite risk index with the preset risk level threshold (e.g., low, medium, high, urgent) to trigger the corresponding level of alarm or maintenance recommendation.

[0096] Specifically, the working principle of this method aims to overcome the limitation of focusing only on single component failure prediction while ignoring the coupling effects between components.

[0097] First, by constructing a weighted directed graph model, the potential interaction relationships between components in the distribution control cabinet, such as physical layout (thermal impact), electrical topology (electrical stress conduction) and functional dependency (control logic impact), are structured and quantified, providing a basis for analyzing coupled faults.

[0098] Subsequently, real-time or recent single component failure prediction data from the digital twin system is dynamically injected into the model, updating the risk status of the nodes in the graph.

[0099] Next, using graph analysis techniques, we search this network for potential high-risk scenarios, known as failure paths or failure clusters, that may be formed by the concatenation or aggregation of multiple seemingly low-risk single prediction events through coupling relationships (strongly weighted edges). This step can reveal hidden, complex failure modes caused by the coupling of multiple factors.

[0100] Then, for each identified composite failure scenario, a composite risk index is derived by comprehensively calculating the failure probability of each component within it (node ​​probability), the strength of their mutual influence (edge ​​weight), and the impact of the scenario on the production system (impact severity). This index provides a quantitative system-level risk assessment, going beyond the simple addition of single failure probabilities.

[0101] Finally, based on the composite risk index, early warning information with clear direction is generated to inform operation and maintenance personnel which composite risk scenarios require priority attention and intervention, thereby guiding the formulation of more effective preventive maintenance plans. Its effect is that it can identify and quantify system-level composite risks caused by coupling effects between components, improve the accuracy and pertinence of predictive maintenance, and help avoid unexpected downtime of critical loads.

[0102] In some specific implementations, a motor control center cabinet that supplies power to a key reactor in a chemical production line is taken as an example.

[0103] First, a weighted directed graph model is constructed: nodes are created for the main incoming circuit breaker Q0, feeder contactor KM1, thermal overload relay FR1, and contactor KM2 in the adjacent circuit. Node attributes include "predicted failure probability." Edges are created: (Q0->KM1, type=Electrical, weight=0.8), (KM1->FR1, type=Electrical, weight=0.9), (KM1->KM2, type=Thermal_Proximity, weight=0.6), and (FR1->KM1, type=Control_Influence, weight=1.0). Weights are estimated based on design parameters and physical distances.

[0104] Next, the digital twin maps the predicted information: the digital twin predicts a 0.15 probability of a sticking failure occurring at KM1 within the next 72 hours, and a 0.10 probability of increased pick-up uncertainty at KM2 due to coil aging. These probability values ​​are then updated to the "Predicted Failure Probability" attribute of the KM1 and KM2 nodes.

[0105] Next, we perform graph analysis, setting a probability threshold of 0.08 and an edge weight threshold of 0.5. The algorithm identifies that both nodes KM1 and KM2 exceed the probability threshold and that a thermal proximity edge with a weight of 0.6 exists between them. This identifies a potential fault cluster {KM1, KM2}, indicating that thermal coupling could exacerbate the risk of simultaneous failure of both nodes. The fault path Q0->KM1->FR1 is also identified.

[0106] Next, calculate the composite risk index: Assume that the system impact severity score of a reactor shutdown caused by a KM1 or KM2 failure is 90 (out of 100). Calculate the composite risk index for the fault cluster {KM1, KM2}, taking into account P(KM1) = 0.15, P(KM2) = 0.10, Edge(KM1, KM2) = 0.6, and Impact = 90, resulting in a risk value of, for example, 78. Calculate the risk index for the path Q0->KM1->FR1, taking into account P(Q0), P(KM1), P(FR1), and the associated edge weights and impact severity.

[0107] Finally, an early warning is generated: a high-risk threshold is set at 70. Because the risk index of 78 for the fault cluster {KM1, KM2} exceeds the threshold, the system generates an early warning: "High-risk alert: A high risk of combined failure (thermal coupling) has been detected for contactors KM1 and KM2, which may cause an unexpected shutdown of a critical reactor. It is recommended to immediately check the status of KM1 and KM2 and consider preventive replacement." Compared to reporting the low-probability risks of KM1 and KM2 separately, this early warning can better reveal potential serious consequences, enabling more accurate maintenance decisions and effectively avoiding production interruptions caused by coupled failures.

[0108] This application further proposes that based on the weighted directed graph model and the attributes of the mapped corresponding nodes, a graph analysis algorithm is executed to identify the fault path or fault cluster that meets the predetermined conditions, including the following steps:

[0109] Construct a fault path index structure to record the set of nodes and their associated edges contained in each identified fault path or fault cluster;

[0110] Monitor changes in node attributes in a weighted directed graph model and determine the set of nodes that have changed;

[0111] Based on the set of changed nodes, query the fault path index structure to obtain the affected fault paths or fault cluster lists;

[0112] Perform local graph analysis only on the paths or clusters in the affected fault paths or fault cluster lists and recalculate their risk status;

[0113] If it is detected that a new node attribute change may form a new fault path or fault cluster, incremental graph analysis is performed in the local range of the changed area to identify the newly added fault path or fault cluster.

[0114] Among them, the fault path index structure is constructed to establish a fast mapping relationship between a node and the identified fault path or fault cluster containing the node. For example, it can be implemented using a hash table or database index, with the key being the node identifier and the value being a list of fault path or fault cluster identifiers containing the node.

[0115] Monitor changes in node attributes in the weighted directed graph model through event monitoring mechanisms or periodic checks. Once the digital twin system updates component fault prediction information and maps it to graph node attributes, subsequent processing is triggered.

[0116] Determine the set of nodes that have changed, that is, collect all nodes whose attribute values ​​have changed during the monitoring period. Based on the set of changed nodes, query the fault path index structure. Using the aforementioned index, find the associated fault path or fault cluster for each changed node, and summarize it to obtain the affected list. This step avoids traversing all identified paths or clusters.

[0117] Performing local graph analysis only on the paths or clusters in the list of affected fault paths or clusters means calling the risk calculation function for each path or cluster in the list and re-evaluating its risk using the updated node properties, rather than executing the analysis algorithm on the entire graph.

[0118] If it is detected that a new node attribute change may form a new fault path or fault cluster, incremental graph analysis is performed locally within the changed area. This means starting from the changed node, a limited depth search (for example, breadth-first search or depth-first search) is performed in the graph model, and combined with predetermined conditions to determine whether a new fault path or fault cluster that meets the conditions is formed. This incremental method avoids the overhead of global search. Therefore, through the combination of the above steps, efficient fault pattern identification and risk assessment of the dynamically updated graph model are achieved.

[0119] Specifically, to address the inefficiency caused by repeated global analysis due to dynamic updates of node attributes when executing graph analysis algorithms to identify fault paths or fault clusters, this technical solution provides an optimized analysis execution strategy. This strategy first constructs a fault path index structure, pre-storing information about identified fault paths or fault clusters and the nodes they contain, and establishing a fast search path from a node to its corresponding risk pattern.

[0120] The system then continuously monitors changes in node attributes (such as predicted failure probability) in the weighted directed graph model. Once a change is detected, the system identifies the specific set of nodes that have changed.

[0121] By using the pre-built fault path index structure, a fast query based on the changed node set can be performed to directly locate the identified fault paths or fault clusters containing these changed nodes to form an affected list.

[0122] The core optimization lies in that the system only performs local graph analysis on the paths or clusters in the affected list and recalculates their risk status, avoiding global analysis of the entire graph model and significantly reducing computational complexity.

[0123] At the same time, in order to cope with the possibility that changes in node attributes may give rise to new risk patterns, an incremental graph analysis mechanism has also been designed. When it is detected that changes in node attributes may form new fault paths or fault clusters, the system only executes an incremental graph search algorithm in the local graph area around the changed node to discover and identify the newly added fault paths or fault clusters.

[0124] This strategy, which combines index query, local update, and incremental discovery, can efficiently maintain the identification results of fault paths or fault clusters and their risk assessments in an environment where node attributes change dynamically. It solves the computational burden and response delay problems brought by global analysis, improves the real-time nature and efficiency of fault response methods, and thus can provide more timely decision support for preventive maintenance.

[0125] In some specific embodiments, for example, a weighted directed graph model of a distribution control cabinet includes nodes N1 (circuit breaker), N2 (contactor), N3 (thermal relay), edges E1 (N1->N2, electrical connection, weight w1), E2 (N2->N3, electrical connection, weight w2), and E3 (N2<->N3, physical proximity thermal influence, weight w3).

[0126] Initially, a global analysis identifies a fault path P1 = (N1->N2->N3) with a risk index of R1. The fault path index structure is constructed as: Index = {N1:[P1], N2:[P1], N3:[P1]}. At this point, the digital twin system updates the predicted fault probability attribute for node N2 (contactor).

[0127] A change in the N2 attribute is detected, and the set of changed nodes is determined to be {N2}. The system queries Index[N2] and obtains the list of affected fault paths as [P1]. Next, the system performs a local graph analysis on only P1. Using the updated probability value from N2, combined with the original attributes of N1 and N3 and the edge weights w1, w2, and w3, the system recalculates the risk index of P1, obtaining R1_new.

[0128] At the same time, the system performs incremental graph analysis, centering around N2 and examining its neighborhood (for example, considering nodes with first- or second-degree connections) to determine whether there are new fault paths or clusters that meet the criteria due to changes in N2's attributes. Assuming the probability of N2 increases, combined with the existing probability of N3 and the weight w3 of edge E3 (thermal impact), a new fault cluster C1 = {N2, N3} (indicating that N2's heat generation has accelerated N3's aging) may meet the preset risk threshold.

[0129] The system identifies the newly added fault cluster C1 and adds it to the index structure: Index = {N1:[P1], N2:[P1, C1], N3:[P1, C1]}. Through this process, the system efficiently updates the status of the existing risk pattern P1 and discovers the new risk pattern C1 without reanalyzing the entire graph model. This significantly improves analysis efficiency and ensures timely risk assessment under dynamic changes.

[0130] The present application further proposes that for each identified fault path or fault cluster, the steps of calculating a composite risk index based on the predicted probability of the nodes it contains, the weight of the connecting edges, and the severity of the impact on the system include: detecting a load condition change event and determining the difference in the conditions before and after the change; judging whether it is necessary to recalculate the composite risk index based on the condition difference; if recalculation is required, obtaining the current load operating parameters and their corresponding influence coefficients; adjusting the predicted probability values ​​of the nodes in the fault path or fault cluster based on the load operating parameters and the influence coefficient; and recalculating the composite risk index based on the adjusted node predicted probability values, the weights of the connecting edges, and the severity of the system impact under the current conditions.

[0131] Among them, the detection of load condition change events is achieved by continuously monitoring the operating parameters of the load connected to the distribution control cabinet, such as current, voltage, power and other timing data.

[0132] To determine the difference between the operating conditions before and after the change, the current operating parameters can be compared with the preset baseline or the parameters of the previous stable state, and a quantitative difference index can be generated by calculating factors such as the size and duration of the deviation.

[0133] Furthermore, whether the composite risk index needs to be recalculated is determined based on the difference in operating conditions. A difference threshold is set. Only when the calculated operating condition difference exceeds this threshold is the subsequent risk recalculation process triggered, thereby avoiding unnecessary calculations for small or short-term operating condition fluctuations.

[0134] If it is determined that recalculation is required, obtaining the current load operating parameters means reading the real-time electrical measurement values, and obtaining the influence coefficient means looking up a pre-established mathematical relationship or lookup table that describes the degree of influence of specific operating parameters (such as current size and temperature) on the failure probability of a specific type of component (node).

[0135] Adjusting the node prediction probability value according to the load operating parameters and influence coefficient is to substitute the acquired real-time operating parameters into the influence coefficient model, calculate the adjustment amount for the original prediction probability of the node, and apply this adjustment amount to the original probability value to obtain a new prediction probability that reflects the current working pressure.

[0136] Finally, based on the adjusted node prediction probability value, the weight of the connecting edge (representing the fault propagation or correlation strength), and the severity of the system impact reassessed or confirmed under the current working conditions, a composite risk calculation model (such as weighted sum, product or other risk synthesis functions) is applied to obtain an updated composite risk index.

[0137] Specifically, this technical solution aims to solve the problem that the composite risk index calculation fails to dynamically adapt to changes in load conditions. The working principle is as follows:

[0138] The system continuously monitors load conditions. Once a change in load conditions is detected, the system first quantifies the significance of the change, calculating the degree of variance. Based on the magnitude of the variance, the system then determines whether to update the risk assessment. If the change is significant enough (the variance exceeds a threshold), the system collects the current actual load operating parameters and determines how these parameters affect the likelihood of failure of related components.

[0139] Using this information, the system adjusts the predicted failure probabilities of component nodes involved in the fault path or fault cluster so that the probability values ​​reflect the additional stress or relief effects brought about by the current operating conditions.

[0140] Finally, these adjusted probabilities are used, along with known fault associations (edge ​​weights) and an assessment of the system impact under the current operating conditions, to recalculate the composite risk index. This closed-loop process dynamically updates the composite risk index based on load conditions, ensuring that risk assessment results closely align with real-time operating conditions. This provides a more accurate basis for preventive maintenance decisions and overcomes the limitation of static risk assessments, which can be inaccurate under dynamic conditions.

[0141] In some specific implementations, for example, a fault path within a motor control center cabinet powering a chemical reactor includes a main contactor node and a thermal overload relay node. Initially, based on digital twin predictions, the contactor node has a predicted failure probability of 0.02, the thermal relay node has a predicted failure probability of 0.01, the edge weight connecting the two is 0.6, and the system impact severity score under the current operating conditions is 80 (out of 100). This results in an initial composite risk index.

[0142] Subsequently, adjustments to the production process required the reactor stirring motor to increase its speed, causing the motor load current to increase from 80% to 110% of the rated value, and this continued. The system detected this continued increase in load current and calculated a working condition difference of 70, exceeding the preset trigger threshold of 50. The system determined that the risk needed to be recalculated. The system obtained the current motor current as 150A (assuming a rated current of 135A) and found that for this model of contactor, the failure probability adjustment factor is 0.00005 for every 1A increase in current; for the thermal relay, the failure probability adjustment factor is 0.00008 for every 1A increase in current.

[0143] The calculated contactor probability adjustment amount = (150-135)A*0.00005 / A = 0.00075.

[0144] Adjusted contactor predicted probability = 0.02 + 0.00075 = 0.02075.

[0145] The calculated thermal relay probability adjustment amount = (150-135)A*0.00008 / A = 0.0012.

[0146] The adjusted predicted probability of the thermal relay = 0.01 + 0.0012 = 0.0112. At the same time, the assessment determined that under this high-load condition, the impact of the path failure on production was more severe, and the system impact severity score was adjusted to 90. Finally, based on the adjusted node probabilities (0.02075 and 0.0112), the original edge weight (0.6), and the new system impact severity score (90), the composite risk index was recalculated. The resulting new risk index will be higher than the initial value, accurately reflecting the actual risk level increased due to the increased load, which may trigger a higher level of warning or adjust the maintenance plan.

[0147] This implementation achieves more accurate risk assessment by dynamically adjusting node probabilities and considering system impacts related to working conditions.

[0148] The present application further proposes obtaining electrical parameter timing data of the load within a preset time window, the electrical parameters including current, voltage, power and power factor; calculating the duration and magnitude of the load change based on the electrical parameter timing data; obtaining the current process stage identification and process conversion plan information from the chemical process control system; judging whether the current load change is consistent with the expected process conversion based on the process stage identification and process conversion plan information; if the load change is consistent with the expected process conversion and the duration exceeds a first threshold, it is identified as a continuous load change; if the load change is inconsistent with the expected process conversion or the duration is less than a second threshold, it is identified as a temporary load change; and calculating the operating condition difference based on the type of load change, the magnitude of the change and the sensitivity coefficient of the corresponding components.

[0149] Among them, the step of obtaining electrical parameter time series data can be achieved by configuring a data acquisition unit to read current, voltage, power and power factor data from a measuring device (such as a smart meter) installed on the load feeder at a fixed time interval (for example, once per second).

[0150] To calculate the duration and magnitude of load changes, time series analysis techniques can be applied, such as setting a baseline value, detecting the magnitude of the deviation of the data point from the baseline value, and recording the continuous length of time the deviation state lasts.

[0151] To obtain information from the chemical process control system, it is necessary to establish a data interface with the system (such as DCS or MES) and regularly query or subscribe to relevant process status tags and production plan data.

[0152] Determining whether the load change is consistent with the expected process transition involves logically comparing the time and magnitude of the detected electrical change with the planned transition time and target process parameters obtained from the process system.

[0153] Identifying the type of load change depends on the comparison results and duration judgment: when the electrical change matches the planned process conversion in terms of time and impact load, and the duration of the changed state exceeds the first threshold (for example, set to 5 minutes), it is determined to be a continuous load change; when the electrical change does not match the plan, or although it matches but the duration does not reach the second threshold (for example, set to 30 seconds, usually the second threshold is less than the first threshold), it is determined to be a temporary load change.

[0154] To calculate the operating condition difference, the determined change type (weight coefficients are assigned to different types, and the weight of continuous change is usually higher than that of temporary change), the calculated change amplitude, and the sensitivity coefficient of the distribution components related to the load change to this type of change stored in the database (a parameter indicating the component's tolerance or the degree of impact of aging) are comprehensively utilized to obtain a quantitative operating condition difference value through a predefined calculation formula (such as weighted product summation).

[0155] Specifically, this technical solution aims to solve the problem of how to specifically detect load condition change events and determine the degree of condition difference.

[0156] By acquiring time-series data on electrical parameters during load operation, we established a data foundation for analysis. Furthermore, we calculated the duration and magnitude of these changes to quantify electrical variations. By incorporating process stage identification and transition plan information from the chemical process control system, we established a link between electrical variations and the production process.

[0157] By comparing the detected load change with the expected process transition and combining the duration of the change (compared with the first and second thresholds), it is possible to distinguish between planned, long-lasting, continuous load changes and unplanned, short-term temporary load changes.

[0158] This distinction is useful for subsequent judgment on whether the risk index needs to be adjusted, because different types of changes have different impacts on components.

[0159] Finally, a quantitative working condition difference is calculated by combining the type of load change, the quantitative change amplitude and the sensitivity coefficient of the affected components to the change.

[0160] This operating condition difference reflects the specific impact of the current load condition change on related components, providing a clear, quantitative input for determining whether to recalculate the composite risk index based on this difference. This resolves the technical issue of only generally mentioning detection events and determining differences without specific implementation details. This enables more accurate assessment of operating condition changes, triggering subsequent risk analysis and decision-making.

[0161] In some embodiments, the system collects data on the motor feeder's current, voltage, active power, and power factor. At one point, the monitored current value rose from the rated value of 50A to 75A and remained stable at this level for eight minutes. The system also queried the DCS system through an interface and obtained information: the current process phase was "feeding and heating," and the planned "constant temperature reaction" phase was five minutes later. This phase transition was expected to cause the stirring motor current to rise to approximately 75A.

[0162] Since the detected current change (amplitude of 25A, i.e., a 50% increase) is consistent with the expected process transition provided by the DCS (the time is basically consistent and the target current value is consistent), and the change duration (8 minutes) exceeds the set first threshold (e.g., 5 minutes), the load change is identified as a "continuous load change."

[0163] Subsequently, the system searches the database for the sensitivity coefficients of the contactor and thermal relay of this model to the continuous current overload (eg, 0.7 and 0.9, respectively).

[0164] According to a preset formula, for example, operating condition difference = change type weight * (current change amplitude / rated current) * average sensitivity coefficient. Assuming the continuous change weight is 1.0, the operating condition difference = 1.0 * (25A / 50A) * (0.7 + 0.9) / 2 = 1.0 * 0.5 * 0.8 = 0.4. This calculated operating condition difference value (0.4, or calibrated to 40 in the range of 0-100 as needed) will be used to determine whether it is necessary to recalculate the composite risk index of the relevant fault path or fault cluster due to this operating condition change. This method distinguishes between planned operations and abnormal fluctuations by combining process information, and quantifies the impact of changes on specific components, providing a more accurate basis for risk assessment.

[0165] The present application further proposes that for each identified fault path or fault cluster, the steps of calculating a composite risk index based on the predicted probability of the included nodes, the weight of the connecting edges and the severity of the impact on the system include: obtaining the current process stage identifier and the corresponding key parameter threshold from the chemical process control system; establishing a process stage and load importance mapping table, and determining the importance level of each load based on the current process stage identifier; obtaining the backup power supply path information and switching time parameters of each load; constructing a system impact severity assessment matrix for the current process stage based on the load importance level, backup power supply path information and switching time parameters; for each fault path or fault cluster, identifying the set of potentially affected loads; calculating the comprehensive impact coefficient of the load set in the current process stage based on the system impact severity assessment matrix; and calculating the composite risk index by combining the predicted probability of the nodes in the fault path or fault cluster, the weight of the connecting edges and the comprehensive impact coefficient.

[0166] Among them, obtaining the current process stage identifier is achieved through an interface with the chemical company's distributed control system or production execution system to ensure that risk assessment is synchronized with production status. The established process stage and load importance mapping table is a data structure, such as a database table or configuration file, which associates process stage codes (such as "reaction", "separation", "distillation") with the load identifiers powered by the distribution control cabinet (such as pumps, mixers, compressors), and assigns an importance level to each association (for example, 1-5 levels, 5 being the highest). Obtaining backup power supply path information involves querying the electrical configuration database or digitized single-line diagram to determine whether the load has a backup power supply, the configuration of the automatic transfer switch (ATS), and the typical switching time.

[0167] The system impact severity assessment matrix is ​​dynamically generated based on the current process stage. The rows or columns of the matrix represent loads, and the values ​​in the cells are calculated based on the importance level of the load at the current stage, the presence or absence of backup power, and the switching speed. High-importance loads with no backup or long switching time receive high severity scores.

[0168] Identifying potentially affected load sets requires utilizing the topology information of the distribution network to trace the electrical connections from the component nodes (such as circuit breakers and contactors) in the fault path or fault cluster down to the final power-consuming equipment.

[0169] When calculating the comprehensive impact coefficient, the system searches for the scores corresponding to the affected load set in the evaluation matrix and summarizes them using a predetermined algorithm (such as summation, weighted average, or maximum value) to obtain a quantitative value representing the overall impact of the fault path or fault cluster at the current process stage.

[0170] Finally, the calculation of the composite risk index integrates the node failure probability predicted by the digital twin, the edge weight representing the association strength in the graph model, and the dynamically calculated comprehensive impact coefficient. The final risk score is obtained through a mathematical formula (for example, the product of each node probability and the edge weight multiplied by the comprehensive impact coefficient).

[0171] Specifically, this method aims to address the problem that fixed or general system impact assessments cannot accurately reflect the differences in the risks of the same fault at different chemical process stages.

[0172] By acquiring the current process stage identifier from the chemical process control system in real time, a risk assessment process closely linked to the current production status is initiated. Using a pre-established mapping table of process stage and load importance, this method can determine the importance of each load to production continuity based on the current process stage. For example, the feed pump of a reactor may be a critical load during the reaction phase, but its importance is reduced during the cleaning phase.

[0173] Furthermore, the method takes into account the redundancy design of the system and evaluates the possibility and speed of power restoration after a fault occurs by obtaining the backup power supply path information and switching time parameters of the load.

[0174] If critical loads have fast-switching backup power sources, their actual impact severity is reduced. Based on this dynamically acquired and configured information (importance, backup power source, and switchover time), a system impact severity assessment matrix is ​​constructed for the current process stage. This matrix quantifies the consequences of each load failure under the current operating conditions.

[0175] When a potential fault path or fault cluster is identified (for example, both the upstream circuit breaker and the downstream contactor of a feeder have a high probability of failure), the method first determines which end loads will be affected by the failure of this path or cluster.

[0176] Then, using the constructed evaluation matrix, the comprehensive impact coefficient of these affected loads at the current process stage is calculated. This coefficient dynamically quantifies the severity of the specific fault in the current production context.

[0177] Finally, this comprehensive impact coefficient is combined with the predicted failure probability of each node in the fault path and the weight of the connecting edges between nodes (representing the strength of the impact transmission) to calculate a composite risk index. This composite risk index more accurately reflects the true risk level to the system caused by the coupling of multiple potential faults at a specific chemical process stage. It guides operations and maintenance personnel to prioritize potential faults with the greatest impact on current production, achieving more accurate and effective preventive maintenance.

[0178] In some embodiments, the current process stage identifier is obtained from the DCS system as "polymerization reaction stage." The process stage and load importance mapping table defines that in this stage, the agitator motor M-101 of reactor R-101 has an importance level of 5 (the highest), and the catalyst injection pump P-201 has an importance level of 4. The electrical configuration shows that M-101 is powered by the main feeder L1 and is equipped with an ATS that can switch to the backup feeder L2 within 2 seconds; P-201 is powered only by the main feeder L1 and has no backup.

[0179] Based on this information, a system impact severity assessment matrix for the "aggregation phase" is constructed. M-101 might receive a severity score of 0.7 (high importance but with a fast backup), while P-201 might receive a severity score of 0.9 (high importance with no backup). At this point, the digital twin system predicts a 0.05 probability of failure for circuit breaker CB-1 on feeder L1 and a 0.08 probability of failure for contactor K-1 (controlling M-101 and P-201). These two components form a fault cluster. The set of loads potentially affected by this fault cluster is {M-101, P-201}.

[0180] Based on the evaluation matrix, the comprehensive impact coefficient of this load set is calculated. For example, using the maximum method, the comprehensive impact coefficient is max(0.7, 0.9) = 0.9. Assuming the edge weight from CB-1 to K-1 is 1.0, the composite risk index of this fault cluster can be calculated as (probability factor) * (comprehensive impact coefficient). The probability factor can simply be the maximum value or product of the node probabilities, for example, taking the maximum value of 0.08. The composite risk index is approximately 0.08 * 0.9 = 0.072.

[0181] If the system switches to the "Equipment Cleaning Phase," the importance of P-201 drops to 2, and its severity score drops to 0.3. The combined impact coefficient for the same fault cluster becomes max(0.7, 0.3) = 0.7, and the composite risk index decreases accordingly to 0.08 * 0.7 = 0.056. This dynamic calculation enables early warning information to accurately reflect the true severity of faults at different production stages, optimizing the allocation of maintenance resources.

[0182] The present application further proposes the steps of calculating the composite risk index by combining the predicted probability of nodes in the fault path or fault cluster, the weight of the connecting edge and the comprehensive influence coefficient, including: establishing a three-layer scoring matrix, corresponding to the technical risk layer, the production impact layer and the safety risk layer respectively; mapping the predicted probability of nodes in the fault path or fault cluster to the technical risk layer, and calculating the technical risk score; mapping the comprehensive influence coefficient of the load set under the current process stage to the production impact layer, and calculating the production impact score; based on the potential failure mode of the fault path or fault cluster, combined with chemical safety regulations, evaluating the type and severity of possible safety incidents, and calculating the safety risk score; determining the weight coefficients of the three levels of technical risk, production impact and safety risk according to the characteristics of the current process stage; weighting the technical risk score, production impact score and safety risk score according to the corresponding weight coefficients to obtain an initial composite risk value; converting the initial composite risk value into a standardized composite risk index through a nonlinear mapping function, wherein the nonlinear mapping function is pre-calibrated based on historical fault data and expert evaluation results.

[0183] Among them, three scoring matrices are established for the technical risk layer, production impact layer and safety risk layer, which set independent dimensions for risk assessment. For example, each matrix can be a lookup table or function that maps input values ​​(probability, impact coefficient, safety assessment results) to 0-100 scores.

[0184] The predicted probability of a node in a fault path or fault cluster is converted into a score at the technical risk layer through predetermined rules (eg, a mapping table between probability intervals and scores). This score reflects the technical possibility of a fault occurring.

[0185] The comprehensive impact coefficient of the load set under the process stage (a value calculated in a previous step) is converted into a score of the production impact layer through mapping (e.g., linear or piecewise linear mapping), which quantifies the interference of the fault on the production process.

[0186] Based on the failure modes assessed by fault paths or fault clusters (e.g., short circuit, open circuit, overheating), and in comparison with chemical safety regulations (e.g., factory safety procedures, industry standards), the types of safety events (e.g., fire, equipment damage, personal injury) and their severity levels are evaluated, and a safety risk score is calculated. For example, a risk matrix approach is used to determine the score by combining the likelihood of an event occurring and the severity of the consequences.

[0187] According to the characteristics of the process stages (for example, the startup stage, the stable operation stage, and the material switching stage), the weight coefficients of the three levels of technical risk, production impact, and safety risk are determined; for example, in the stable operation stage, the production impact weight coefficient can be set to 0.5, and the technical and safety risk weight coefficients are 0.25 each; in the specific hazardous materials handling stage, the safety risk weight coefficient can be set to 0.6.

[0188] The technical risk score, production impact score, and safety risk score are weighted using their corresponding weight coefficients (the calculation formula is: technical risk score × technical weight coefficient + production impact score × production weight coefficient + safety risk score × safety weight coefficient) to obtain an initial composite risk value. This initial composite risk value is input into a nonlinear mapping function (for example, a sigmoid function or a piecewise function). This function is calibrated based on historical failure data and expert assessment results, and outputs a standardized composite risk index (for example, ranging from 0 to 100). This step uniformly scales the risk values, facilitating comparison and threshold setting, while also enhancing the ability to distinguish high-risk areas.

[0189] Specifically, this technical solution addresses the question of how to integrate technical risk, production impact, and safety risk factors, and determine their weightings based on process stages to generate a standardized composite risk index. This method first establishes a three-level scoring hierarchy for technical risk, production impact, and safety risk, providing a framework for risk quantification. The technical risk score is determined by the node prediction probability and reflects the failure rate at the technical level.

[0190] The production impact score is determined by the comprehensive impact coefficient, reflecting the impact on production continuity. The safety risk score is determined by evaluating the consequences of failure and referring to safety regulations, incorporating safety considerations.

[0191] Introducing a weighting factor that varies based on the characteristics of the process stage allows for dynamic adjustment of the proportion of different risk dimensions in the final assessment, tailoring the risk assessment to the needs of the working conditions and improving its accuracy.

[0192] By performing weighted calculation on the scores of the three dimensions, the initial composite risk value is obtained, and the method of integrating multi-dimensional information is clarified.

[0193] Finally, a nonlinear mapping function calibrated based on historical data and expert experience is applied to convert the initial composite risk value into a standardized composite risk index. This standardization step improves the comparability and interpretability of the index, facilitating subsequent early warning and maintenance decision-making. These steps establish a systematic, contextualized, and standardized composite risk assessment process, addressing the lack of a specific fusion calculation method and enabling a more comprehensive and quantitative assessment of system-level risks caused by coupled multi-component faults within distribution control cabinets.

[0194] The present application further proposes that the steps of adjusting the predicted probability value of the node in the fault path or fault cluster according to the load operating parameters and the influence coefficient include: obtaining the time series data of the load operating parameters within a preset time window; calculating the change rate of the load operating parameters, and dividing the change rate into multiple levels; establishing a correspondence table between the load operating parameter change rate and the component stress increment; according to the change rate level of the load operating parameter, querying the corresponding component stress increment from the correspondence table; multiplying the component stress increment by the influence coefficient to obtain the adjusted value of the node predicted probability; calculating the original node predicted probability and the adjusted value to obtain the adjusted node predicted probability value.

[0195] The purpose of obtaining the time series data of the load operating parameters within the preset time window is to provide data input for the subsequent change rate calculation.

[0196] Calculating the rate of change of load operating parameters is to quantify how quickly the parameters change over time, and then classify the calculated rate values ​​into different level intervals based on preset thresholds, such as "stable", "slowly changing", "rapidly changing", and other levels.

[0197] Establishing a corresponding relationship table between the rate of change of load operating parameters and the stress increment of components is to predefine the additional stress generated by different change rate levels on related components. The table can be obtained through experiments, simulations or analysis based on physical models.

[0198] According to the level of the calculated change rate, the corresponding stress increment value is found in the corresponding relationship table.

[0199] The stress increment of the component found is multiplied by the influence coefficient specific to the component to obtain a specific value used to adjust the prediction probability. The influence coefficient reflects the sensitivity of the component to load changes.

[0200] The original predicted failure probability of the component is combined with the calculated adjustment value through a predetermined operation method such as addition or multiplication to generate the final adjusted node prediction probability value. This probability value reflects the immediate impact of the current dynamic change of the load on the component failure risk.

[0201] Specifically, this method aims to quantify the impact of changes in load operating parameters on the predicted probability of component failure and generate probability adjustment values ​​based on the impact coefficients. By acquiring time-series data of load parameters, calculating their rate of change, and grading them, a quantitative connection between load dynamics and component stress is established.

[0202] Using a pre-set correspondence table, the rate of change level is converted into a specific stress increment. This stress increment is then combined with the influence coefficient that characterizes the sensitivity of the component to calculate the adjustment to the predicted probability.

[0203] Finally, this adjustment is applied to the original predicted probability to obtain an updated probability value. This process provides a clear and actionable step for incorporating dynamic load changes under current conditions into the assessment of composite risk. This provides a quantitative basis for adjusting node predicted probabilities, thereby improving the accuracy of risk assessment.

[0204] The present application further proposes that the steps of calculating the working condition difference according to the type of load change, the amplitude of the change and the sensitivity coefficient of the corresponding components include: obtaining preset reference baseline data, the preset reference baseline data including the standard operating parameters of various loads under normal working conditions; calculating the deviation value of the current load change relative to the preset reference baseline data; setting a weight coefficient according to the type of load change, wherein the weight coefficient of the continuous load change is higher than the weight coefficient of the temporary load change; multiplying the deviation value by the weight coefficient to obtain a weighted deviation value; multiplying the weighted deviation value by the sensitivity coefficient of the corresponding component to obtain a component response value; accumulating the response values ​​of all affected components to obtain a cumulative response value; converting the cumulative response value into a standardized working condition difference in the range of 0-100 according to a preset normalization function; determining the corresponding warning level according to the standardized working condition difference, wherein the warning level includes four levels: normal, attention, warning and emergency, and each warning level corresponds to a different difference threshold range.

[0205] Among them, the step of obtaining preset reference baseline data involves retrieving pre-set standard values ​​of electrical parameters representing various types of loads under design or historical stable operating conditions, such as the average value or range of current, voltage, and power, from a storage medium.

[0206] The step of calculating the deviation value is achieved by performing mathematical subtraction or ratio operation on the load parameter value monitored in real time and the corresponding baseline data.

[0207] The step of setting the weight coefficient is based on the previously determined load change type (persistent or temporary), and selecting a corresponding coefficient from a preset mapping relationship (for example, persistent corresponds to a weight of 1.0, and temporary corresponds to a weight of 0.5).

[0208] The step of calculating the weighted deviation value is to multiply the deviation value obtained by the above calculation by the selected weight coefficient.

[0209] The step of calculating the component response value requires the establishment of a sensitivity coefficient library for each component or component type in the distribution control cabinet in advance. This coefficient represents the response degree or vulnerability of the component to a specific load parameter change, and then multiplies the weighted deviation value by the sensitivity coefficient of the affected component.

[0210] The step of accumulating response values ​​is to sum up the response values ​​of all relevant components affected by the same load change event.

[0211] The standardization processing step applies a predefined mathematical function (such as linear mapping, Sigmoid function, etc.) to map the aforementioned cumulative response values ​​to a uniform numerical range of 0 to 100. The parameters of the function are calibrated based on historical data or expert experience.

[0212] The step of determining the warning level is to compare the calculated standardized working condition difference with the preset threshold intervals of each level (for example, normal [0, 20], concern (20, 50], warning (50, 80], emergency (80, 100]) to determine the level it belongs to.

[0213] Specifically, this technical approach aims to address the problem of how to calculate operating condition variability in a specific and standardized manner and associate it with warning levels. By introducing a reference baseline, the magnitude of load variations can be quantified. By distinguishing the type of load variation (persistent or temporary) and assigning different weights, the assessment better reflects the potential impact of the change, with persistent variations receiving greater attention.

[0214] The introduction of component sensitivity coefficients allows the evaluation to take into account the differential responses of different components to the same change.

[0215] The cumulative response of each component forms a comprehensive measure of the degree of impact on the system level. Through standardization functions, the differences calculated under different operating conditions are comparable, facilitating unified management.

[0216] Finally, the standardized variability is mapped to clear warning levels, providing direct, graded guidance for operations and maintenance personnel. This series of steps constitutes a complete calculation process, from raw data input to final warning level output. This solves the problem of lack of specific implementation details and standardization in operating condition variability calculation, making the operating condition variability assessment process clear, quantitative, and comparable, and can directly guide warning responses.

[0217] The present application further proposes the steps of mapping the component failure information predicted by the digital twin system to the attributes of the corresponding nodes in the weighted directed graph model, including: establishing standardized mapping rules for component failure information and graph model node attributes, the standardized mapping rules including attribute field definitions, numerical range constraints and outlier processing strategies; receiving component failure prediction information output by the digital twin system, the component failure prediction information including component identification, failure mode, prediction probability and prediction time window; detecting missing fields or outliers in the component failure prediction information, and determining the information completeness score based on preset data integrity requirements; when the information completeness score is lower than a first threshold, retrieving historical records under similar working conditions from the historical database to extract supplementary information; when the information completeness score is lower than a second threshold and valid supplementary information cannot be obtained from the historical database, generating an estimated value based on the component type and its position in the distribution system; according to the standardized mapping rules, converting the complete or supplemented component failure prediction information into the attribute value of the graph model node; recording the information source and uncertainty index of each mapping operation for credibility analysis in subsequent risk assessment.

[0218] The step of establishing standardized mapping rules is intended to provide a basis for subsequent data processing and conversion. This rule defines the conversion logic from prediction information to graph model node attributes, including specifying which prediction information fields correspond to which node attributes, stipulating the data type and allowed value range of each attribute (for example, the prediction probability is between 0 and 1), and clarifying how to identify and initially process out-of-range or non-formatted data as outliers.

[0219] The step of receiving prediction information is to obtain basic data input from the digital twin system, which includes the basic elements required for fault analysis: which component (identification), what kind of problem may occur (failure mode), the possibility of occurrence (prediction probability) and the time range (prediction time window).

[0220] The detection and scoring step assesses the quality of the received information. It checks for missing fields or values ​​that do not conform to the preset range, and calculates a quantitative completeness score based on preset criteria (for example, the weight of each field).

[0221] The data supplementation and estimation steps are remedial measures taken when the score indicates incomplete data. If the score is below the first threshold but above the second threshold, the system attempts to fill in the missing information by querying the historical database to find historical failure prediction records for the same component or component type under similar operating conditions (such as load and ambient temperature).

[0222] If the score is lower than the second threshold and the historical data search is unsuccessful, an alternative solution is activated to obtain a default value or a statistically estimated value from a preset knowledge base or configuration table based on the type of component itself (such as circuit breaker, contactor) and its installation location information in the distribution cabinet.

[0223] The conversion step according to the rules is to assign the prediction information after the above-mentioned inspection, supplementation or estimation processing to the attribute field of the corresponding node in the weighted directed graph model according to the established standardized mapping rules.

[0224] The step of recording the source of information and uncertainty indicators is to attach metadata to the data generated by each mapping, indicating whether its source is the original forecast, historical supplement or system estimate, and may be accompanied by an indicator reflecting the credibility of the data (for example, the uncertainty of the supplementary data is higher than that of the original data, and the uncertainty of the estimated value is the highest).

[0225] Specifically, this method aims to solve the problem that when the prediction information of the digital twin system is introduced into the graphical model, the accuracy of subsequent analysis is affected by missing or abnormal original data.

[0226] Here’s how it works:

[0227] First, standardized mapping rules are established to unify data processing. Upon receiving prediction information from the digital twin system, a data quality check is immediately performed to identify missing fields or values ​​that do not meet specifications, and these are quantified into an information completeness score.

[0228] Based on this score, the system determines the processing strategy: if the score is high enough, it will directly enter the mapping; if the score is lower than the first threshold, the historical data retrieval mechanism will be activated to use past experience to supplement the information; if the score further drops below the second threshold and the historical data supplementation fails, an estimation method based on component type and location will be used to generate alternative values.

[0229] After data preparation (raw, supplemented, or estimated) is complete, standardized mapping rules are applied to transform the information into graph model node attributes. Crucially, each mapping operation records the data's provenance and uncertainty, for example, marking an attribute value as "estimated" and assigning it a lower confidence score.

[0230] In this way, even when processing imperfect data, the graphical model can be ensured to obtain attribute inputs with quality descriptions. Subsequent graph analysis and risk calculations can be weighted or adjusted based on these information sources and uncertainty indicators, thereby improving the robustness of the entire fault response process and the strength of the basis for the final decision.

[0231] Reference Figure 3 In some preferred embodiments, the system block diagram is as follows Figure 3As shown in the figure, the system adopts a three-layer architecture design, including input layer, core processing layer and output layer.

[0232] Specifically, the system receives three types of key data inputs:

[0233] The digital twin system provides component failure prediction information, including component identification, failure mode, predicted probability, and prediction time window. This data is generated based on the degradation model and real-time operating parameters of a single component.

[0234] The chemical process control system transmits process status information, including the current process stage identifier, process transition plan, and key parameter thresholds. This information is used to determine the load importance level and the severity of the system impact.

[0235] The real-time load parameters collected by the load condition monitoring equipment, including time series data of electrical parameters such as current, voltage, power and power factor, are used to detect load change events and calculate the degree of condition difference.

[0236] After data enters the system, it flows and is processed along the following paths:

[0237] The weighted directed graph model, as the core data structure of the system, pre-constructs a graph structure that represents the relationships between components within the power distribution control cabinet. This model contains three key elements:

[0238] Node: represents physical components and their failure modes;

[0239] Edge: represents physical proximity, electrical connection, and potential impact.

[0240] Weight: quantifies the strength of the relationship and assigns a numerical value to each edge;

[0241] The fault information mapping module receives the predicted data from the digital twin system and converts it into attributes of the corresponding nodes in a weighted directed graph. During processing, this module also checks the data integrity and, when necessary, supplements information from the historical database or generates estimates.

[0242] The graph analysis algorithm module uses algorithms such as path search and community discovery based on the updated weighted directed graph to identify fault paths or fault clusters that meet predetermined conditions. This module uses an index structure and incremental analysis strategy to achieve efficient local update processing.

[0243] The composite risk index calculation module integrates multiple aspects of information:

[0244] The fault path / cluster structure obtained from the graph analysis algorithm module;

[0245] Predicted probabilities of nodes in a path / cluster;

[0246] The weight of the connecting edge;

[0247] Information related to the process stages obtained from chemical process control systems;

[0248] Load parameter change data obtained from load condition monitoring;

[0249] By establishing a three-tiered scoring matrix (technical risk, production impact, and safety risk), a comprehensive composite risk index is calculated. As load conditions change, the module dynamically adjusts node prediction probabilities to ensure real-time adaptability of risk assessment.

[0250] The early warning information generation module generates graded early warning information based on the calculated composite risk index and preset risk thresholds. This information not only indicates the risk level but also pinpoints high-risk component combinations and their potential impact, providing maintenance personnel with clear guidance for preventive maintenance decisions.

[0251] This system framework shifts from single-point prediction to system-level assessment. By effectively capturing the coupling relationships between components through a graphical model, it makes previously overlooked compound risks visible and quantifiable. The system's dynamic adaptability ensures real-time updates to risk assessments as process states and load conditions change, providing continuous and effective decision support.

[0252] Especially in industrial environments such as chemical production that have extremely high requirements for power supply continuity, the system can identify potential high-risk fault combinations caused by factors such as thermal coupling and electrical connections, thereby helping operation and maintenance personnel to formulate more targeted maintenance plans and effectively avoid unplanned downtime and production losses caused by coupling failures.

[0253] Reference Figure 2 , this application further proposes a distribution control cabinet fault response device for digital twin predictive maintenance of production enterprises, the device comprising:

[0254] A modeling module 210 is configured to pre-build a weighted directed graph model representing the relationships between components within the power distribution control cabinet. The weighted directed graph model includes nodes representing physical components and their failure modes, and weighted edges representing physical proximity relationships, electrical connection relationships, and potential impact relationships.

[0255] A mapping module 220 is used to map the component fault information predicted by the digital twin system to the attributes of the corresponding nodes in the weighted directed graph model;

[0256] An analysis module 230 is configured to execute a graph analysis algorithm based on the weighted directed graph model and the attributes of the mapped corresponding nodes to identify fault paths or fault clusters that meet predetermined conditions;

[0257] The calculation module 240 is used to calculate a composite risk index for each identified fault path or fault cluster based on the predicted probability of the nodes it contains, the weight of the connecting edges, and the severity of the impact on the system;

[0258] The early warning module 250 is used to generate early warning information according to the composite risk index to guide preventive maintenance decisions.

[0259] By constructing a weighted directed graph model that includes physical proximity, electrical connections, and potential impact relationships, mapping and predicting fault information, and using graph analysis algorithms to identify fault paths or clusters, the composite risk index that takes into account node probabilities, edge weights, and system impacts is calculated. This solves the problem that existing technologies are difficult to assess composite risks under the interaction of multiple components. It has the ability to identify potential fault propagation paths or fault clustering areas based on the physical, electrical, and impact relationships between components, and quantify the composite risks formed by the combination of multiple potential faults, thereby guiding preventive maintenance decisions.

[0260] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A distribution control cabinet fault response method, applied to the digital twin predictive maintenance system of a manufacturing enterprise, characterized in that: include: Pre-building a weighted directed graph model representing the relationships between components within the power distribution control cabinet, wherein the weighted directed graph model includes nodes representing physical components and their failure modes, and weighted edges representing physical proximity relationships, electrical connection relationships, and potential impact relationships; Mapping component fault information predicted by the digital twin system to attributes of corresponding nodes in the weighted directed graph model; Based on the weighted directed graph model and the attributes of the mapped corresponding nodes, executing a graph analysis algorithm to identify fault paths or fault clusters that meet predetermined conditions; For each identified fault path or fault cluster, a composite risk index is calculated based on the predicted probability of the nodes it contains, the weight of the connecting edges, and the severity of the impact on the system; Early warning information is generated based on the composite risk index to guide preventive maintenance decisions.

2. A power distribution control cabinet fault response method according to claim 1, characterized in that: The step of executing a graph analysis algorithm based on the weighted directed graph model and the attributes of the mapped corresponding nodes to identify a fault path or a fault cluster that meets a predetermined condition includes: Construct a fault path index structure to record the set of nodes and their associated edges contained in each identified fault path or fault cluster; Monitoring changes in node attributes in the weighted directed graph model and determining a set of nodes that have undergone changes; Based on the changed node set, query the fault path index structure to obtain a list of affected fault paths or fault clusters; Performing local graph analysis only on the paths or clusters in the affected fault paths or fault cluster lists, and recalculating their risk states; If it is detected that a new node attribute change may form a new fault path or fault cluster, incremental graph analysis is performed in the local range of the changed area to identify the newly added fault path or fault cluster.

3. A power distribution control cabinet fault response method according to claim 1, characterized in that: The step of calculating the composite risk index for each identified fault path or fault cluster based on the predicted probability of the included nodes, the weight of the connecting edges, and the severity of the impact on the system includes: Detect load condition change events and determine the difference between the conditions before and after the change; According to the difference in the working conditions, determine whether it is necessary to recalculate the composite risk index; If recalculation is required, obtain the current load operating parameters and their corresponding influence coefficients; Adjusting the predicted probability values ​​of nodes in the fault path or fault cluster according to the load operating parameters and the influence coefficient; The composite risk index is recalculated based on the adjusted node prediction probability value, the weight of the connecting edge, and the severity of the system impact under the current working conditions.

4. A power distribution control cabinet fault response method according to claim 3, characterized in that: The step of detecting a load condition change event and determining the difference between the conditions before and after the change includes: Obtaining time series data of electrical parameters of the load within a preset time window, wherein the electrical parameters include current, voltage, power, and power factor; Calculating the duration and magnitude of the load change based on the electrical parameter time series data; Obtain current process stage identification and process conversion plan information from the chemical process control system; Determining whether the current load change is consistent with the expected process transition based on the process stage identifier and process transition plan information; If the load change is consistent with the expected process transition and the duration exceeds a first threshold, identifying it as a continuous load change; If the load change does not conform to the expected process transition or the duration is less than a second threshold, identifying it as a temporary load change; The working condition difference is calculated according to the type and magnitude of the load change and the sensitivity coefficients of the corresponding components.

5. A power distribution control cabinet fault response method according to claim 1, characterized in that: The step of calculating the composite risk index for each identified fault path or fault cluster based on the predicted probability of the included nodes, the weight of the connecting edges, and the severity of the impact on the system includes: Obtain the current process stage identifier and its corresponding key parameter threshold value from the chemical process control system; Establish a mapping table between process stages and load importance, and determine the importance level of each load based on the current process stage identifier; Obtain backup power supply path information and switching time parameters for each load; Based on the load importance level, backup power supply path information and switching time parameters, a system impact severity assessment matrix for the current process stage is constructed; For each fault path or fault cluster, identify the set of loads that it potentially affects; Calculating the comprehensive impact coefficient of the load set at the current process stage according to the system impact severity assessment matrix; The composite risk index is calculated by combining the predicted probabilities of the nodes in the fault path or fault cluster, the weights of the connecting edges, and the comprehensive impact coefficient.

6. A power distribution control cabinet fault response method according to claim 5, characterized in that: The step of calculating the composite risk index by combining the predicted probability of nodes in the fault path or fault cluster, the weight of the connecting edge, and the comprehensive impact coefficient includes: Establish a three-tier scoring matrix, corresponding to the technical risk layer, production impact layer, and security risk layer; Mapping the predicted probability of nodes in the fault path or fault cluster to a technical risk layer and calculating a technical risk score; Mapping the comprehensive impact coefficient of the load set at the current process stage to the production impact layer and calculating the production impact score; Based on the potential failure modes of the fault paths or fault clusters, and in combination with chemical safety regulations, the types and severity of possible safety incidents are assessed, and a safety risk score is calculated; According to the characteristics of the current process stage, determine the weight coefficients of the three levels of technical risk, production impact and safety risk; The technical risk score, production impact score and safety risk score are weighted according to corresponding weight coefficients to obtain an initial composite risk value; The initial composite risk value is converted into a standardized composite risk index through a nonlinear mapping function, wherein the nonlinear mapping function is pre-calibrated according to historical failure data and expert evaluation results.

7. A power distribution control cabinet fault response method according to claim 3, characterized in that: The step of adjusting the predicted probability value of the node in the fault path or fault cluster according to the load operating parameter and the influence coefficient includes: Obtaining time series data of load operating parameters within a preset time window; Calculating a rate of change of the load operating parameter and classifying the rate of change into multiple levels; Establish a corresponding relationship table between the rate of change of load operating parameters and the stress increment of components; According to the change rate level of the load operating parameter, query the corresponding component stress increment from the corresponding relationship table; Multiplying the component stress increment by the influence coefficient to obtain an adjustment value of the node prediction probability; The original node prediction probability is calculated with the adjustment value to obtain an adjusted node prediction probability value.

8. A power distribution control cabinet fault response method according to claim 4, characterized in that: The step of calculating the operating condition difference according to the type and magnitude of the load change and the sensitivity coefficient of the corresponding components includes: Obtaining preset reference baseline data, wherein the preset reference baseline data includes standard operating parameters of various loads under normal operating conditions; Calculating a deviation value of the current load change relative to the preset reference baseline data; Setting a weight coefficient according to the type of load change, wherein the weight coefficient of a continuous load change is higher than the weight coefficient of a temporary load change; Multiplying the deviation value by the weight coefficient to obtain a weighted deviation value; Multiplying the weighted deviation value by the sensitivity coefficient of the corresponding component to obtain a component response value; Accumulate the response values ​​of all affected components to obtain a cumulative response value; According to a preset normalization function, the cumulative response value is converted into a standardized working condition difference within the range of 0-100; According to the standardized working condition difference, the corresponding warning level is determined, wherein the warning level includes four levels: normal, attention, warning and emergency, and each warning level corresponds to a different difference threshold range.

9. A power distribution control cabinet fault response method according to claim 1, characterized in that: The step of mapping the component fault information predicted by the digital twin system to the attributes of the corresponding nodes in the weighted directed graph model includes: Establishing standardized mapping rules between component fault information and graph model node attributes, wherein the standardized mapping rules include attribute field definitions, value range constraints, and outlier processing strategies; Receiving component failure prediction information output by the digital twin system, wherein the component failure prediction information includes component identification, failure mode, prediction probability, and prediction time window; Detect missing fields or abnormal values ​​in the component fault prediction information and determine an information completeness score based on preset data completeness requirements; When the information completeness score is lower than a first threshold, retrieving historical records under similar working conditions from a historical database to extract supplementary information; When the information completeness score is lower than a second threshold and valid supplementary information cannot be obtained from the historical database, generating an estimated value based on the component type and its location in the power distribution system; Converting the complete or supplemented component fault prediction information into attribute values ​​of graph model nodes according to the standardized mapping rules; Record the information source and uncertainty indicators of each mapping operation for credibility analysis in subsequent risk assessment.

10. A power distribution control cabinet fault response device, used for digital twin predictive maintenance of manufacturing enterprises, characterized in that: The device includes: A modeling module is used to pre-build a weighted directed graph model representing the relationships between components within the power distribution control cabinet, wherein the weighted directed graph model includes nodes representing physical components and their failure modes, and weighted edges representing physical proximity relationships, electrical connection relationships, and potential impact relationships; A mapping module, configured to map component fault information predicted by the digital twin system to attributes of corresponding nodes in the weighted directed graph model; An analysis module, configured to execute a graph analysis algorithm based on the weighted directed graph model and the attributes of the mapped corresponding nodes to identify a fault path or a fault cluster that meets predetermined conditions; A calculation module is used to calculate the composite risk index for each identified fault path or fault cluster based on the predicted probability of the nodes it contains, the weight of the connecting edges, and the severity of the impact on the system; The early warning module is used to generate early warning information according to the composite risk index to guide preventive maintenance decisions.

Images