Equipment health diagnosis and training support method

Abstract

The application discloses a method for equipment health diagnosis and training support, and relates to the technical field of intelligent equipment support. The specific steps of the method are as follows: firstly, task priority calculation is performed, priority coefficients and grades are calculated according to collected parameters, and then the parameters are transmitted; secondly, fault atlas reasoning is performed, and a fault is located according to the atlas; thirdly, health state evaluation is performed, a health index is calculated by fusing data, and an early warning is generated; fourthly, resource scheduling decision is carried out, a scheduling scheme is generated by receiving the early warning; and finally, closed-loop parameter correction is performed, algorithm parameters are corrected by collecting execution data, and the scheme is reconstructed. In one aspect, the method fuses multiple features to complete priority calculation, builds a fault space-time knowledge atlas, and realizes accurate equipment health determination and trend prediction. In another aspect, the method models support resources in a digital manner, generates a support scheduling scheme by using a multi-element coupling algorithm, collects data for closed-loop correction, realizes efficient resource allocation, and promotes intelligent upgrading of the support mode.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment support technology, specifically to methods for equipment health diagnosis and training support. Background Technology

[0002] The equipment industry is currently developing towards integration, intelligence, and complexity. The application scenarios for equipment training, drills, and maintenance support are constantly expanding. Task execution exhibits multi-dimensional characteristics, multi-stage advancement, and parallel execution under multiple constraints. The health status of equipment directly determines the continuity, stability, and overall efficiency of training, drills, and maintenance operations. Equipment systems contain multiple functional subsystems, generating massive amounts of multi-source, heterogeneous monitoring data during operation. Fault manifestations are also shifting from single, independent faults to complex, interconnected faults. Traditional equipment management and support models are no longer adequate for the health control needs of complex scenarios. With the rapid development of cutting-edge technologies such as digital technology, artificial intelligence, and knowledge graphs, health diagnosis and intelligent support throughout the entire equipment lifecycle have become a core industry development trend. How to build an integrated support system combining task characteristics, equipment status, and resource allocation has become a key technology urgently needing breakthroughs in equipment maintenance and training support, and is also an important support for improving equipment operational reliability and ensuring the efficient implementation of various tasks.

[0003] Traditional equipment health diagnosis and training support technologies have many inherent flaws, making them ill-suited to the operation and maintenance requirements of modern complex equipment. Task prioritization relies solely on single-dimensional indicators, failing to integrate the full range of task characteristics and spatiotemporal attributes for comprehensive analysis. This lack of scientific rigor and rationality in task classification directly leads to a disconnect between support resource allocation and actual task requirements. Fault diagnosis excessively depends on manual experience and single-point data monitoring, lacking a systematic fault correlation knowledge system. This hinders accurate fault location and the ability to deduce fault propagation paths, resulting in low efficiency and a high misjudgment rate. Health status assessment fails to achieve deep integration of fault reasoning data and sensor monitoring data. Pre-warning thresholds are fixed and cannot be dynamically adjusted according to task requirements, severely compromising the relevance and timeliness of health warnings. The lack of full-attribute digital modeling of support resources, the failure of scheduling algorithms to achieve coupled calculation of multi-dimensional elements of tasks, equipment, and resources, the unreasonable allocation of resources, the lag in support response, the lack of a closed-loop correction mechanism for execution data feedback, the inability of algorithm parameters to iteratively optimize with actual scenarios, and the rigidity and lag of the overall support system have greatly restricted the improvement of equipment support efficiency and operation management level. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for equipment health diagnosis and training support. It constructs a full-process intelligent support system for training equipment through five key stages: task priority calculation, fault graph reasoning, health status assessment, resource scheduling decision-making, and closed-loop parameter correction. The method utilizes a spatiotemporal weighted algorithm to classify tasks by level, combines a fault spatiotemporal knowledge graph and graph attention network to accurately locate the root cause of faults and deduce propagation paths, achieves dynamic health early warning based on a graph fusion health algorithm, and completes efficient matching of tasks, equipment, and resources through a ternary coupled scheduling algorithm. Finally, it uses execution data to close-loop correct algorithm parameters. This invention realizes intelligent, collaborative, and closed-loop health diagnosis and support scheduling for training equipment, effectively improving the accuracy of fault diagnosis and resource utilization efficiency, adapting to the needs of complex training scenarios, and significantly enhancing the equipment's training support capabilities.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for equipment health diagnosis and training support, the specific steps of which are as follows: S1, Task Priority Calculation: Collect multi-dimensional feature parameters of training tasks, use a spatiotemporal weighted priority algorithm to calculate task priority coefficients and classify task levels, and transmit the priority coefficients and task levels to subsequent steps. S2, Fault Graph Reasoning: Receives the priority coefficient and task level transmitted by the task priority calculation step, collects equipment operation monitoring data and fault history data, constructs a spatiotemporal knowledge graph of equipment faults and performs incremental learning, locates the root cause node of the fault through the graph attention network and infers the fault propagation path. S3, Health Status Assessment: The fault propagation path output from the fault graph reasoning step is integrated with the equipment sensor monitoring data. The graph fusion health algorithm is used to calculate the equipment health index and predict the health change trend. The warning judgment threshold is adjusted according to the priority coefficient of the task priority solution step to generate a dynamic health warning signal. When adjusting the warning judgment threshold based on the priority coefficient transmitted in the task priority calculation step, the initial warning judgment threshold of each equipment component is first preset. When the priority coefficient is greater than 0.7, the initial warning judgment threshold of the corresponding equipment component is lowered by 10%-15%. When the priority coefficient is greater than or equal to 0.3 and less than or equal to 0.7, the initial warning judgment threshold of each equipment component remains unchanged. When the priority coefficient is less than 0.3, the initial warning judgment threshold of each equipment component remains unchanged. When generating a dynamic health warning signal, the signal includes the equipment abnormal part identifier, health index value, risk level identifier, warning judgment threshold adjustment range, and handling requirement identifier. The risk level identifier is divided according to the health index value, and the handling requirement identifier is determined according to the risk level identifier. After the signal is generated, it is synchronously transmitted to the resource scheduling decision step. S4, Resource Scheduling Decision: Receives dynamic health warning signals from the health status assessment step, performs full-attribute digital modeling of support resources and constructs a support resource efficiency matrix, uses a ternary coupled scheduling algorithm to calculate task-equipment-resource matching decision variables, generates a support resource scheduling plan and sends it to the execution terminal; The resource allocation plan includes a resource allocation list, an optimal scheduling route, a standardized work sequence, and execution time nodes. The resource allocation list specifies the quantity, specifications, and allocation direction of various resources. The optimal scheduling route is planned based on the geographical location of the resources and the location of equipment failures. The standardized work sequence specifies the order of failure detection, component disassembly, spare parts replacement, and equipment debugging. The execution time nodes specify the start and end times of each support link. After the plan is generated, it is sent to the execution terminal according to the preset process. S5, Closed-loop parameter correction: Collect data on the execution process of the guarantee scheme in the guarantee resource scheduling decision-making step, compare the execution data with the model prediction data, correct the parameters of the algorithm used in the previous steps, and reconstruct the guarantee resource scheduling scheme based on the corrected parameters.

[0006] Furthermore, in S1, the task priority calculation involves collecting multi-dimensional characteristic parameters of the training mission, specifically including mission type, execution stage, strategic value, time node, coverage area, participating forces, equipment dependence, environmental constraints, mission continuity, alternative feasibility, delay loss, and review weight. Tactical personnel, equipment support personnel, and training command personnel jointly assign weights to the above twelve types of characteristic parameters. Consistency verification is performed on the weight allocation results, and data with logical contradictions in weight allocation are eliminated. A spatiotemporal weighted priority algorithm is used, which combines the weights and standardized values ​​of various parameters, the remaining time of the mission from the key execution node, and the ratio of the strategic weight of the mission execution area to the total strategic weight of the entire training domain to calculate the task priority coefficient with a value range of 0 to 1. Based on the priority coefficient value, the mission is divided into three categories: key mission, routine mission, and auxiliary mission.

[0007] Furthermore, in S1, the spatiotemporal weighted priority algorithm used in task priority calculation employs the following mathematical expression:

[0008] in , , These are the weighting coefficients, and , This represents the task priority coefficient corresponding to the spatial location s at time t. This represents the weighting of the feature parameters for the i-th task class. This represents the standardized value of the feature parameters of the i-th task class. This indicates the remaining time until the critical execution node is reached. This indicates the total duration of the task. This represents the strategic weight value of the task execution area. This represents the overall strategic weight value across the entire training and exercise domain.

[0009] Furthermore, in S2, the fault graph reasoning process collects equipment operation monitoring data including vibration data, temperature data, pressure data, speed data, current data, voltage data, sealing data, and wear data of the equipment's power system, navigation system, hydraulic system, and electrical system. All equipment operation monitoring data are collected in real time through dedicated sensor terminals deployed on the equipment. The collected fault history data includes historical fault records of the equipment since its commissioning, maintenance history of each component, various fault handling cases, and maintenance cycle information of each component. Among them, historical fault records include fault occurrence time, fault location, fault manifestation, and fault level; component maintenance history includes maintenance time, maintenance components, maintenance methods, and maintenance consumables usage; fault handling cases include fault type, handling process, tools used, and handling time; and maintenance cycle information includes the routine maintenance cycle of each component and the basis for adjusting the maintenance cycle under special working conditions. After the data collection is completed, a spatiotemporal knowledge graph of equipment faults is constructed and incremental learning is performed. The root cause node of the fault is located and the fault propagation path is deduced through a graph attention network.

[0010] Furthermore, in S2, during fault graph reasoning, when constructing the equipment fault spatiotemporal knowledge graph, equipment physical components, sensing points, fault modes, and historical fault events are used as entity nodes. Among them, equipment physical component nodes include component model, installation location, and functional parameters; sensing point nodes include installation location, monitoring parameters, and measurement range; fault mode nodes include fault type and manifestation characteristics; and historical fault event nodes include fault occurrence time, occurrence location, and handling process. The internal physical connection relationship of the equipment, the causal transmission relationship of faults between components, and the temporal evolution dependency relationship of fault states are used as association edges to build an initial equipment fault spatiotemporal knowledge graph. During incremental learning, newly collected fault data, newly discovered component associations, and new fault handling cases are injected into the initial knowledge graph in real time, and the entity node features and association edge weights in the knowledge graph are updated synchronously. The update frequency is consistent with the equipment operation monitoring data collection frequency. After the update is completed, the root cause node of the fault is located and the fault propagation path is deduced through a graph attention network.

[0011] Furthermore, in S2, during fault graph reasoning, when the graph attention network is running, the set of entity nodes and the set of associated edges of the equipment fault spatiotemporal knowledge graph are first input. The feature data of all entity nodes are normalized, and then the attention weight between each entity node and its neighboring entity nodes is calculated. The weight calculation process combines the entity node's own features, the features of neighboring entity nodes, and the attribute information of the associated edges to normalize and calibrate the calculated attention weights. Based on the calibrated attention weights, the features of each entity node are weighted and aggregated to obtain the fused features of each entity node. Then, the fused features are processed by a multilayer perceptron to output the association score of each entity node. The root cause node of the fault is located based on the association score, and the propagation path of the fault between entity nodes is inferred based on the association score.

[0012] Furthermore, in S3, the mathematical expression used in the map fusion health algorithm for health status assessment is:

[0013] in This represents the health index of the equipment at position s at time t. The weight coefficients representing the reasoning results of a knowledge graph. Let N represent the reasoning score of the fault spatiotemporal knowledge graph, where N represents the set of entity nodes in the knowledge graph, and R represents the set of related edges in the knowledge graph. This represents the standardized value of the sensor data of the j-th type of equipment. This represents the fault propagation attenuation coefficient. This represents a standardized value indicating the propagation distance from the root cause node of the fault to the monitoring component.

[0014] Furthermore, in S4, during resource scheduling decision-making, when constructing the resource efficiency matrix, the resource type is used as the matrix row dimension, and the equipment failure mode is used as the matrix column dimension. Each element in the matrix corresponds to the efficiency value of a single resource for a single equipment failure mode. The efficiency value is obtained by weighting the resource's repair rate, allocation cost, response time, coverage radius, and available quantity. The weighting operation uses a preset fixed weight coefficient. The matrix row dimension covers all resource guarantees that have completed full-attribute digital modeling, and the matrix column dimension covers all marked failure modes of the equipment. After the matrix is ​​constructed, it is synchronously input into the ternary coupled scheduling algorithm for calculating the task-equipment-resource matching decision variables.

[0015] Furthermore, in S4, the mathematical expression used in the ternary coupled scheduling algorithm for resource scheduling decision-making is:

[0016] in This represents the guaranteed scheduling efficiency value corresponding to position s at time t. This represents the resource efficiency value of the k-th type of protection resource for the m-th type of failure mode. This represents the weight coefficients in multi-objective optimization. This represents the actual cost of protection. This indicates the preset maximum protection cost value. This indicates the actual response time. This indicates the preset maximum guaranteed response time value. This represents the actual utilization rate of guaranteed resources. This indicates the preset maximum guaranteed resource utilization rate.

[0017] Furthermore, in S5, during the closed-loop parameter correction, the collected execution data includes support resource location data, on-site operation progress data, equipment repair status data, support response time data, and resource consumption data. The support resource location data is updated at a fixed frequency, the equipment repair status data includes a post-repair health index, and the support response time data records the total time from the issuance of the scheduling instruction to the completion of maintenance. The corrected algorithm parameters include the weight coefficients of the spatiotemporal weighted priority algorithm, the graph fusion health algorithm, and the ternary coupled scheduling algorithm; the spatiotemporal feature factor of the spatiotemporal weighted priority algorithm; the fault propagation coefficient of the graph fusion health algorithm; and the resource efficiency coefficient of the ternary coupled scheduling algorithm. When the training mission is adjusted or the equipment health status changes abruptly, the support resource scheduling scheme is reconstructed based on the corrected parameters.

[0018] Compared with existing technologies, this equipment health diagnosis and training support method has the following beneficial effects: I. This invention achieves intelligent priority calculation by integrating the full-dimensional features and spatiotemporal attributes of the task. It constructs a spatiotemporal knowledge graph of faults by combining equipment operation monitoring data and fault history, and continuously performs incremental learning. Relying on graph attention networks, it accurately locates the root cause of faults and infers fault propagation paths. Then, it integrates fault reasoning results with sensor monitoring data to conduct health status assessments, dynamically adjusting the early warning threshold based on task requirements. This method completely changes the traditional situation of one-sided task classification, fault diagnosis relying on human experience, and health assessments being disconnected from tasks. It significantly improves the accuracy of fault location and path inference, making health warnings more aligned with actual task needs. It achieves accurate judgment and trend prediction of equipment health status, strengthens the dynamic control capability of equipment operation status, provides a scientific basis for equipment health management, ensures the stable and reliable status of equipment during task execution, and lays a solid foundation for subsequent support operations.

[0019] II. This invention utilizes a comprehensive digital model of support resources, constructing a resource efficiency matrix. A multi-factor coupling algorithm is employed to achieve collaborative matching decisions among tasks, equipment, and resources, generating support scheduling schemes adapted to specific scenarios. Simultaneously, data from the entire support execution process is collected, and the algorithm parameters are iteratively corrected by comparing actual execution results with model predictions. Based on these corrected parameters, the scheduling scheme is reconstructed. This model overcomes the limitations of traditional resource scheduling, such as rigidity, delayed response, and lack of closed-loop optimization. It achieves efficient allocation and rational utilization of support resources, reduces resource idleness and loss, shortens support response and operation cycles, and improves the accuracy and flexibility of resource scheduling. The closed-loop correction mechanism enables the support system to possess autonomous optimization capabilities, quickly adapting to task adjustments and equipment status changes. This comprehensively improves the overall efficiency of equipment health diagnosis and training support, driving the support model towards intelligence, refinement, and dynamism.

[0020] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0022] Figure 1 A flowchart for equipment health diagnosis and training support methods; Figure 2 A schematic diagram illustrating data transmission between various steps in the equipment health diagnosis and training support methods; Figure 3 A detailed flowchart for fault map reasoning. Detailed Implementation

[0023] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0024] Reference Figure 1One embodiment of the present invention proposes an equipment health diagnosis and training support method. It adopts an integrated processing mechanism that includes task priority spatiotemporal weighted solution, equipment fault spatiotemporal knowledge graph construction and incremental learning, graph attention network fault root cause localization and path inference, graph fusion health index calculation and dynamic early warning, three-element coupling support resource scheduling, and execution data feedback closed-loop parameter correction. It can adapt to actual scenarios such as dynamic changes in training tasks, real-time fluctuations in equipment operating status, and changes in support resource distribution. It can achieve accurate classification of training task levels, accurate location of equipment faults, quantitative assessment of health status, and optimal scheduling of support resources. At the same time, it can achieve automated, precise, and closed-loop operation of the entire process of equipment health diagnosis and training support by iteratively optimizing algorithm parameters through actual execution data.

[0025] The method described in this embodiment specifically includes: S1, Task Priority Calculation: Collect multi-dimensional feature parameters of training tasks, use a spatiotemporal weighted priority algorithm to calculate task priority coefficients and classify task levels, and transmit the priority coefficients and task levels to subsequent steps. S2, Fault Graph Reasoning: Receives the priority coefficient and task level transmitted by the task priority calculation step, collects equipment operation monitoring data and fault history data, constructs a spatiotemporal knowledge graph of equipment faults and performs incremental learning, locates the root cause node of the fault through the graph attention network and infers the fault propagation path. S3, Health Status Assessment: The fault propagation path output from the fault graph reasoning step is integrated with the equipment sensor monitoring data. The graph fusion health algorithm is used to calculate the equipment health index and predict the health change trend. The warning judgment threshold is adjusted according to the priority coefficient of the task priority solution step to generate a dynamic health warning signal. S4, Resource Scheduling Decision: Receives dynamic health early warning signals from the health status assessment step, performs full-attribute digital modeling of the support resources and constructs a support resource efficiency matrix, uses a ternary coupled scheduling algorithm to calculate the task equipment resource matching decision variables, generates a support resource scheduling scheme and sends it to the execution terminal; S5, Closed-loop parameter correction: Collect data on the execution process of the guarantee scheme in the guarantee resource scheduling decision-making step, compare the execution data with the model prediction data, correct the parameters of the algorithm used in the previous steps, and reconstruct the guarantee resource scheduling scheme based on the corrected parameters.

[0026] S1, Task priority calculation: Specifically, when performing task priority calculation in this step, the first step is to collect multi-dimensional characteristic parameters of the training mission. A total of twelve categories of multi-dimensional characteristic parameters are collected, namely, mission type, execution stage, strategic value, time node, coverage, participating forces, equipment dependence, environmental constraints, mission continuity, alternative feasibility, delay loss, and review weight. The above twelve categories of parameters can comprehensively cover all dimensions of information such as the attribute characteristics, execution conditions, resource dependence, strategic significance, and loss risk of the training mission, providing complete data support for task priority calculation.

[0027] After the feature parameters were collected, three types of professionals—tactical personnel, equipment support personnel, and training command personnel—collaborated to carry out the weight allocation work. From the professional perspectives of tactical execution, equipment support, and training command, respectively, and in combination with the actual scenario and mission requirements of the current training exercise, the three types of personnel assigned corresponding weights to the twelve feature parameters. After the weight allocation was completed, all weight results were checked for consistency. By using preset check rules, logical contradictions, numerical anomalies, duplicate assignments, and other problems in the weight allocation were checked. Weight data that did not meet the check requirements were removed, and compliant and reasonable weight results were retained.

[0028] After completing the weight allocation and consistency verification, the spatiotemporal weighted priority algorithm is used to calculate the task priority coefficient. This algorithm combines the weights and standardized values ​​of twelve types of feature parameters, the remaining time of the task from the key execution node, and the ratio of the strategic weight of the task execution area to the total strategic weight of the entire training domain for comprehensive calculation. During the calculation process, all input data are standardized to eliminate the differences in the dimensions and numerical ranges of different parameters, and finally obtain the task priority coefficient with a value range of 0 to 1.

[0029] After obtaining the priority coefficient, the tasks are divided into three categories: critical tasks, routine tasks, and auxiliary tasks according to the preset value range. After the level division is completed, the task priority coefficient and task level are synchronously transmitted to the subsequent fault map reasoning steps.

[0030] The mathematical expression used in the spatiotemporal weighted priority algorithm is:

[0031] in , , These are the weighting coefficients, and , This represents the task priority coefficient corresponding to the spatial location s at time t. This represents the weighting of the feature parameters for the i-th task class. This represents the standardized value of the feature parameters of the i-th task class. This indicates the remaining time until the critical execution node is reached. This indicates the total duration of the task. This represents the strategic weight value of the task execution area. This represents the overall strategic weight value across the entire training and exercise domain.

[0032] For example, when performing task priority calculation, the full collection of twelve types of training task characteristic parameters is first completed. The parameter weights are allocated by three types of professionals: tactical, equipment support, and training command, and consistency verification is performed. After eliminating contradictory weight allocation data, the standardized values ​​of the twelve types of characteristic parameters, the allocated weights, the remaining task duration, the total task duration, the strategic weight of the execution area, and the total strategic weight of the entire training domain are substituted into the spatiotemporal weighted priority algorithm formula. The weighted summation of the twelve types of characteristic parameters is first completed, followed by the time dimension ratio calculation and the spatial strategic dimension ratio calculation. Finally, the comprehensive calculation is completed by combining the α, β, and γ weight coefficients to obtain the task priority coefficient in the range of 0 to 1. The corresponding task level is matched according to the priority coefficient value, and the judgment result of critical task / routine task / auxiliary task is transmitted to the fault map reasoning step along with the priority coefficient.

[0033] S2, Fault Diagram Reasoning: Specifically, when performing fault map reasoning in this step, the task priority coefficient and task level data transmitted in step S1 are received first. Based on this, equipment data collection is carried out. The collected data is divided into two categories: equipment operation monitoring data and fault history data.

[0034] Equipment operation monitoring data is collected in real time through dedicated sensor terminals deployed on the equipment. The collection scope covers the equipment's power system, navigation system, hydraulic system, and electrical system. Specific parameters collected include vibration data, temperature data, pressure data, speed data, current data, voltage data, sealing data, and wear data. The above data can reflect the actual operating status of each system of the equipment in real time and provide real-time status basis for fault reasoning.

[0035] The fault history data is the full-cycle historical data of the equipment since its commissioning, specifically including historical fault records, maintenance history of each component, various fault handling cases, and maintenance cycle information of each component. Among them, historical fault records include the fault occurrence time, fault location, fault manifestation, and fault level; component maintenance history includes maintenance time, repaired components, maintenance methods, and maintenance consumables usage; fault handling cases include fault type, handling process, tools used, and handling time; and maintenance cycle information includes the routine maintenance cycle of each component and the basis for adjusting the maintenance cycle under special operating conditions. The above historical data provides experience support for the construction of the fault knowledge graph.

[0036] After data collection is completed, a spatiotemporal knowledge graph of equipment failure is constructed, with equipment physical components, sensor points, failure modes, and historical failure events as entity nodes. Among them, equipment physical component nodes include component model, installation location, and functional parameters; sensor point nodes include installation location, monitoring parameters, and measurement range; failure mode nodes include failure type and manifestation characteristics; and historical failure event nodes include failure occurrence time, occurrence location, and handling process. The initial spatiotemporal knowledge graph of equipment failure is constructed by using the internal physical connection relationship of the equipment, the causal transmission relationship of failure between components, and the temporal evolution dependency relationship of failure state as association edges.

[0037] After the initial knowledge graph is built, incremental learning is performed. Newly collected fault data, newly discovered component relationships, and new fault handling cases are injected into the initial knowledge graph in real time. The entity node features and associated edge weights are updated synchronously. The update frequency is consistent with the equipment operation monitoring data collection frequency to ensure that the knowledge graph always matches the latest operating status of the equipment.

[0038] After completing the knowledge graph construction and incremental learning, the graph attention network is run for fault reasoning: First, the set of entity nodes and the set of associated edges of the knowledge graph are input, and the feature data of all entity nodes are normalized; then, the attention weights between nodes are calculated and normalized by combining the features of the entity nodes themselves, the features of adjacent nodes, and the attributes of associated edges; based on the calibrated attention weights, the node features are weighted and aggregated to obtain the node fusion features; the fusion features are processed by a multilayer perceptron to output the association score of each entity node; the root cause node of the fault is located based on the association score, and the propagation path of the fault between entity nodes is deduced. After the reasoning is completed, the fault propagation path is transmitted to the health status assessment step.

[0039] For example, when performing fault graph reasoning, the system first receives task priority coefficients and task level data, completes the collection of operational monitoring data for the four major systems of equipment power, navigation, hydraulics, and electrical, and retrieves the equipment's full-cycle fault history data. Using equipment physical components, sensor points, fault modes, and historical fault events as entity nodes, and physical connections, fault causal propagation, and temporal evolution dependencies as associated edges, an initial fault spatiotemporal knowledge graph is constructed. New fault data and component relationships are injected into the graph to complete incremental learning, and the updated graph is input into a graph attention network. After node feature normalization, attention weight calculation, feature weighted aggregation, and multilayer perceptron processing, the association score of each entity node is obtained. Based on the score, the root cause node of the fault is determined, and the complete propagation path of the fault from the root cause node to each associated component is deduced. The fault propagation path data is then transmitted to the health status assessment step.

[0040] S3, Health Status Assessment: Specifically, in this step of health status assessment, the fault propagation path data output from step S2 is first fused with equipment sensor monitoring data. Based on the fused data, a graph fusion health algorithm is used to calculate the equipment health index. Simultaneously, based on historical and real-time data of the health index, the future health trend of the equipment is predicted. The graph fusion health algorithm integrates the results of fault spatiotemporal knowledge graph reasoning, 23 types of equipment sensor monitoring data, and the propagation distance from the fault root cause node to the monitored component to complete the quantitative calculation of the equipment health level. The calculated health index can intuitively reflect the current health status of the equipment.

[0041] After the health index is calculated, the warning threshold is dynamically adjusted based on the task priority coefficient obtained in step S1. The warning threshold is adjusted accordingly for different task priority coefficients. The higher the priority of the task, the stricter the warning threshold is set. After the threshold adjustment is completed, the health index is compared with the adjusted threshold, and a dynamic health warning signal is generated in combination with the health change trend. The warning signal is then transmitted to the resource scheduling decision step.

[0042] The mathematical expression used in the graph fusion health algorithm is:

[0043] in This represents the health index of the equipment at position s at time t. The weight coefficients representing the reasoning results of a knowledge graph. Let N represent the reasoning score of the fault spatiotemporal knowledge graph, where N represents the set of entity nodes in the knowledge graph, and R represents the set of related edges in the knowledge graph. This represents the standardized value of the sensor data of the j-th type of equipment. This represents the fault propagation attenuation coefficient. This represents a standardized value indicating the propagation distance from the root cause node of the fault to the monitoring component.

[0044] For example, when performing a health status assessment, the fault propagation path data and equipment sensor monitoring data are first integrated. The fault spatiotemporal knowledge graph reasoning score, knowledge graph entity node set, knowledge graph associated edge set, standardized values ​​of 23 types of equipment sensor data, fault propagation attenuation coefficient, and standardized values ​​of the propagation distance from the fault root cause node to the monitored component are substituted into the graph fusion health algorithm formula. The weighted calculation of the knowledge graph reasoning results is first completed, then the mean calculation of the 23 types of sensor data is completed, and finally, the comprehensive calculation is completed by combining the fault propagation distance to obtain the equipment health index. The early warning judgment threshold is adjusted according to the task priority coefficient, the health index is compared with the threshold, and a dynamic health early warning signal is generated by combining the trend of health index changes. The early warning signal is then transmitted to the resource scheduling decision-making step.

[0045] S4, Resource Scheduling Decision: Specifically, when making resource scheduling decisions in this step, the dynamic health warning signal transmitted in step S3 is received first, and the resource protection processing work is carried out as a trigger condition.

[0046] First, we conduct full-attribute digital modeling of the resources to transform all attributes of the resources into standardized digital data, forming a unified format resource digital model, which provides a data foundation for the construction of the resource efficiency matrix.

[0047] After completing the digital modeling, a support resource efficiency matrix is ​​constructed. The support resource type is the row dimension of the matrix, and the equipment failure mode is the column dimension. Each element in the matrix corresponds to the efficiency value of a single support resource for a single failure mode. The efficiency value is obtained by weighted calculation of repair rate, allocation cost, response time, coverage radius, and available quantity. The weighting calculation uses a preset fixed weight coefficient. The matrix row dimension covers all support resources that have been modeled, and the column dimension covers all marked failure modes of the equipment, fully matching the adaptation relationship between resources and failures.

[0048] After the matrix is ​​constructed, a ternary coupled scheduling algorithm is used to calculate the decision variables for matching task equipment resources. This algorithm combines task priority coefficient, equipment health index, and support resource efficiency value, while taking into account multiple objectives such as support cost, response time, and resource utilization to complete the calculation of decision variables. Based on the decision variables, a support resource scheduling plan is generated, which includes resource configuration, allocation path, and operation requirements. After generation, the plan is sent to the execution terminal.

[0049] The mathematical expression used in the ternary coupled scheduling algorithm is:

[0050] in This represents the guaranteed scheduling efficiency value corresponding to position s at time t. This represents the resource efficiency value of the k-th type of protection resource for the m-th type of failure mode. This represents the weight coefficients in multi-objective optimization. This represents the actual cost of protection. This indicates the preset maximum protection cost value. This indicates the actual response time. This indicates the preset maximum guaranteed response time value. This represents the actual utilization rate of guaranteed resources. This indicates the preset maximum guaranteed resource utilization rate.

[0051] For example, when executing resource scheduling decisions, the system first receives dynamic health warning signals, completes digital modeling of all resource attributes, constructs a resource efficiency matrix with resource type as rows and equipment failure mode as columns, and calculates the efficiency values ​​of each resource for each failure mode. Then, it substitutes the task priority coefficient, equipment health index, resource efficiency values, multi-objective optimization weight coefficient, ratio of actual support cost to maximum cost, ratio of actual response time to maximum response time, and ratio of actual resource utilization rate to maximum utilization rate into the ternary coupled scheduling algorithm formula. The system first completes the coupled calculation of tasks, equipment, and resources, then completes the calculation of multi-objective optimization terms, and finally obtains the resource scheduling efficiency values ​​and decision variables. Based on the decision variables, a complete resource scheduling plan is generated and distributed to the field execution terminal.

[0052] S5, Closed-loop parameter correction: Specifically, when performing closed-loop parameter correction in this step, the first step is to collect full-dimensional data during the execution of the support plan. The collected execution data includes support resource location data, on-site operation progress data, equipment repair status data, support response time data, and resource consumption data. Among them, the support resource location data is updated in real time at a fixed frequency, the equipment repair status data includes the health index after repair, and the support response time data records the total time from the issuance of the dispatch instruction to the completion of maintenance. The above data fully reflects the actual execution effect of the support plan.

[0053] After completing the data collection, the actual execution data is compared with the model prediction data item by item. The value, direction and cause of the data deviation are analyzed to form the deviation analysis results. Based on the deviation results, the parameters of the preceding algorithms are modified in a targeted manner. The modification scope includes the weight coefficients of the spatiotemporal weighted priority algorithm, the graph fusion health algorithm and the ternary coupling scheduling algorithm, the spatiotemporal characteristic factor of the spatiotemporal weighted priority algorithm, the fault propagation coefficient of the graph fusion health algorithm and the resource efficiency coefficient of the ternary coupling scheduling algorithm.

[0054] After the parameters are corrected, if the training mission is adjusted or the equipment health status changes abruptly, the entire process of mission priority calculation, fault map reasoning, health status assessment, and resource scheduling decision-making will be re-executed based on the corrected parameters. This will reconstruct a support resource scheduling scheme that is adapted to the current scenario and achieve closed-loop optimization of the entire process.

[0055] For example, when performing closed-loop parameter correction, all execution data, including the location of support resources, on-site operation progress, equipment repair status, support response time, and resource consumption, are first collected. The collected execution data is then compared item by item with the corresponding data predicted by the model, and the deviation values ​​and patterns of each data point are statistically analyzed. Based on the deviation results, the weight coefficients and spatiotemporal feature factors of the spatiotemporal weighted priority algorithm, the weight coefficients and fault propagation coefficients of the graph fusion health algorithm, and the weight coefficients and resource efficiency coefficients of the ternary coupled scheduling algorithm are corrected respectively. When the training mission is adjusted or the equipment health status changes abruptly, all corrected algorithm parameters are invoked to re-complete the mission priority calculation, fault graph inference, health status assessment, and resource scheduling decision-making operations, generating and issuing a support resource scheduling scheme adapted to the latest scenario.

[0056] The preferred features in the above embodiments can be used individually in any embodiment, or in any combination thereof, provided they do not conflict with each other. Furthermore, parts not described in detail in the embodiments can be implemented using existing technologies.

[0057] The following examples further illustrate this application in order to better understand the above-mentioned technical solutions. It should be understood that the following are merely some examples and are not intended to limit this application.

[0058] Reference Figure 2 This application employs a method for equipment health diagnosis and training support to conduct health diagnosis and training support work on a general support equipment cluster for routine base-based training exercises. In some specific implementations of this application, a full-process health diagnosis and training support work is carried out on a general support equipment cluster consisting of communication support equipment, detection support equipment, transportation support equipment, and power supply support equipment in a routine base-based training scenario. This equipment cluster encompasses real-time sensor monitoring data, full-cycle fault history data, multi-type training task data, and support resource distribution data for its power system, navigation system, hydraulic system, and electrical system. The method for equipment health diagnosis and training support provided in this application is used to complete the full-process support work. The specific implementation steps are as follows: S1, Task priority calculation: Multi-dimensional characteristic parameters of training tasks of the general support equipment cluster for routine training exercises in a base are acquired in real time. After weight allocation and consistency verification, the task priority is calculated using a spatiotemporal weighted priority algorithm to generate task priority coefficients and task level data.

[0059] Specifically, the multi-dimensional characteristic parameters of the training mission were collected and preprocessed in a comprehensive manner. The collected parameters included twelve categories of characteristics: mission type, execution stage, strategic value, time node, coverage, participating forces, equipment dependence, environmental constraints, mission continuity, alternative feasibility, delay loss, and review weight. The parameters were then verified for validity, missing values ​​were filled in, and outliers were removed. After testing, the parameters of this training mission were found to be free of missing, outliers, and non-standard formats, forming standardized mission characteristic data, which defined the effective range for subsequent weight allocation and priority calculation.

[0060] Subsequently, personnel from the tactical domain, equipment support domain, and exercise command domain collaborated to complete the weight allocation of twelve types of characteristic parameters. Consistency verification was performed on the weight allocation results, eliminating weight data with logical contradictions and numerical conflicts, and retaining compliant weight results. A spatiotemporal weighted priority algorithm was used to calculate the standardized task characteristic data. Combining the characteristic parameter weights with standardized values, the remaining time of the task from the critical execution node, and the ratio of the strategic weight of the task execution area to the total strategic weight of the entire exercise domain, a task priority coefficient with a value of 0 to 1 was generated. Based on the coefficient, three task levels were divided into critical tasks, routine tasks, and auxiliary tasks, forming task priority calculation result data, which provides a task dimension basis for subsequent fault map reasoning.

[0061] S2, Fault Diagram Reasoning: Based on task priority coefficients and task level data, real-time operation monitoring data and fault history data of general support equipment clusters are collected to construct a spatiotemporal knowledge graph of equipment faults and complete incremental learning. The root cause nodes of faults are located and the fault propagation path is deduced through graph attention network.

[0062] Specifically, equipment operation monitoring data is collected in real time through dedicated equipment sensing terminals, covering vibration, temperature, pressure, speed, current, voltage, sealing performance, and wear data of the power system, navigation system, hydraulic system, and electrical system; fault history data is collected from historical fault records, component maintenance history, fault handling cases, and maintenance cycle information since the equipment was commissioned, and the two types of data are normalized and formatted.

[0063] Using physical components, sensing points, fault modes, and historical fault events as entity nodes, and physical connections, fault causal transmission relationships, and fault temporal evolution dependencies as associated edges, an initial equipment fault spatiotemporal knowledge graph is constructed. New fault data, new component relationships, and new handling cases are injected into the graph in real time to complete incremental learning, with the update frequency consistent with the monitoring data collection frequency.

[0064] The updated knowledge graph is input into the graph attention network. After node feature normalization, attention weight calculation, feature weighted aggregation, and multilayer perceptron processing, the entity node association score is output. Based on the score, the root cause node of the fault is located, the propagation path of the fault among equipment components is deduced, and the fault graph reasoning result data is formed.

[0065] S3, Health Status Assessment: By integrating fault propagation path data with equipment sensor monitoring data, a graph fusion health algorithm is used to calculate the equipment health index and predict health change trends. The warning threshold is adjusted in conjunction with the task priority coefficient to generate dynamic health warning signals.

[0066] Specifically, the results of fault spatiotemporal knowledge graph reasoning, standardized data of 23 types of equipment sensing and monitoring, and data on the propagation distance from the root cause node of the fault to the monitoring component are fused and processed, and then substituted into the graph fusion health algorithm to complete the calculation of the equipment health index. Based on the historical time series data of the health index, the future health change trend of the equipment is predicted.

[0067] The health warning threshold is dynamically adjusted based on the task priority coefficient. High-priority tasks correspond to strict warning thresholds, while low-priority tasks are adapted to regular warning thresholds. The health index is compared with the adjusted thresholds, and a hierarchical dynamic health warning signal is generated by combining the health change trend, providing a basis for equipment health status for subsequent resource scheduling decisions.

[0068] S4, Resource Scheduling Decision: Receive dynamic health early warning signals, conduct full-attribute digital modeling of support resources, construct a support resource efficiency matrix, use a ternary coupled scheduling algorithm to calculate task equipment resource matching decision variables, generate support resource scheduling schemes and send them to execution terminals.

[0069] Specifically, a full-attribute digital model of support resources such as maintenance personnel, spare parts and consumables, testing equipment, and transport vehicles at the training site is carried out, integrating attribute data such as resource type, available quantity, deployment location, response time, repair rate, allocation cost, and coverage radius. A support resource efficiency matrix is ​​constructed with support resource type as the row dimension and equipment failure mode as the column dimension, and the efficiency value within the matrix is ​​calculated by weighting repair rate, allocation cost, response time, coverage radius, and available quantity.

[0070] The task priority coefficient, equipment health index, and support resource efficiency value are substituted into the three-element coupled scheduling algorithm. Taking into account multiple objective optimization factors such as support cost, response time, and resource utilization, the algorithm calculates the support scheduling efficiency value and decision variables. Based on the decision variables, it generates a support resource scheduling plan that includes resource allocation, allocation path, and operation specifications, and sends it to the field execution terminal in real time.

[0071] S5, Closed-loop parameter correction: Collect full-process execution data of the resource scheduling plan, compare the execution data with the model prediction data, correct the parameters of each algorithm, and reconstruct the resource scheduling plan based on the corrected parameters to form a closed-loop optimization mechanism.

[0072] Specifically, data on the location of support resources, on-site operation progress, equipment repair status, support response time, and resource consumption are collected. The location of support resources is updated at a fixed frequency. The equipment repair status includes the health index after repair. The support response time fully records the entire cycle from the issuance of the instruction to the completion of the maintenance.

[0073] The executed data and model prediction data are compared item by item to analyze the statistical deviation values, directions, and causes. The weight coefficients of the spatiotemporal weighted priority algorithm, the graph fusion health algorithm, and the ternary coupled scheduling algorithm are then adjusted accordingly. The spatiotemporal feature factors of the spatiotemporal weighted priority algorithm, the fault propagation coefficient of the graph fusion health algorithm, and the resource efficiency coefficient of the ternary coupled scheduling algorithm are also corrected. When training missions are adjusted or equipment health status changes abruptly, the entire process of priority calculation, fault reasoning, health assessment, and resource scheduling is re-executed based on the corrected algorithm parameters to reconstruct a resource scheduling scheme adapted to the current scenario.

[0074] For example, data from the entire process of task priority calculation, fault graph reasoning, health status assessment, and resource scheduling decision-making is collected, along with execution feedback data from the support site. The results of the entire support process are integrated with the on-site execution information. Based on the above information, the effectiveness of each algorithm parameter is quantitatively evaluated to comprehensively test the actual application effect of each parameter. The parameter calculation deviation, early warning accuracy, scheduling matching rate, and support response efficiency are statistically analyzed. The evaluation found that the calculation deviations of some fault propagation coefficients and resource efficiency coefficients exceeded the preset range, and these were identified as parameters to be corrected, thus clarifying the direction of algorithm optimization.

[0075] The parameters to be corrected are imported into a preset simulation environment isolated from the training and support system for verification testing. Batch simulation calculations are carried out using historical equipment data from base-based training, boundary condition cases, and multi-equipment mixed operation cases. After multi-dimensional verification, the verification results are judged according to preset effectiveness quantification indicators, including calculation accuracy, early warning accuracy, scheduling matching rate, and support response efficiency. Parameter compliance thresholds are set: calculation accuracy ≥ preset threshold, early warning accuracy ≥ preset threshold, scheduling matching rate ≥ preset threshold, and support response efficiency ≥ preset threshold. In addition, calculation deviation, early warning false alarm rate, and scheduling deviation rate are all lower than preset upper limits. Parameters that meet all verification requirements are screened and organized to form a candidate parameter set for algorithm optimization.

[0076] The candidate parameter set is quantitatively calibrated and fused to optimize using a preset parameter correction logic. The parameters of each algorithm are updated incrementally. This parameter update took 6 minutes, which meets the update time requirement and efficiently completes the algorithm parameter optimization.

[0077] After the parameters are updated, the latest parameter configuration is pushed to the health diagnosis and support scheduling system to achieve hot loading of parameters. This allows the system to adapt to the latest parameters without restarting, improving the efficiency of diagnosis and scheduling response. At the same time, the parameter update results are used to optimize the weighting coefficients of each algorithm, and the weighting coefficients of fault map inference, health assessment and resource scheduling are adjusted to make the algorithm calculation effect more in line with the actual needs of base-based training support.

[0078] Every three days thereafter, the core indicators of equipment health diagnosis and support scheduling are summarized to dynamically grasp the actual execution effect of algorithms and plans, dynamically adjust the task priority judgment criteria, health warning sensitivity, and support resource scheduling priority, continuously optimize support strategies, generate a closed-loop iteration report every month, comprehensively summarize the optimization of algorithm parameters and changes in support efficiency, form a full-process closed-loop self-optimization mechanism, realize the continuous improvement of equipment health diagnosis and training support capabilities with actual application, and provide technical support and algorithm optimization foundation for equipment support work in different types of training activities such as base-based routine training and special skills training.

[0079] This application provides a method for equipment health diagnosis and training support, which effectively improves the accuracy of training mission priority calculation, the comprehensiveness of equipment fault reasoning, the scientific nature of health status assessment, the rationality of support resource scheduling, and the closed-loop nature of algorithm parameter correction. It reduces the loss of support efficiency caused by manual judgment and parameter deviation. Furthermore, through the systematic calibration and dynamic optimization function of unified algorithm parameters, it provides technical support and algorithm optimization foundation for subsequent health diagnosis and support work for different types of training scenarios and different equipment clusters. This enables the intelligent, standardized, and efficient operation of the entire process of training equipment from mission priority calculation, fault diagnosis, health assessment to support scheduling and parameter correction.

[0080] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for equipment health diagnosis and training support, characterized in that, The specific steps of this method are as follows: S1, Task Priority Calculation: Collect multi-dimensional feature parameters of training tasks, use a spatiotemporal weighted priority algorithm to calculate task priority coefficients and classify task levels, and transmit the priority coefficients and task levels to subsequent steps. S2, Fault Graph Reasoning: Receives the priority coefficient and task level transmitted by the task priority calculation step, collects equipment operation monitoring data and fault history data, constructs a spatiotemporal knowledge graph of equipment faults and performs incremental learning, locates the root cause node of the fault through the graph attention network and infers the fault propagation path. S3, Health Status Assessment: The fault propagation path output from the fault graph reasoning step is integrated with the equipment sensor monitoring data. The graph fusion health algorithm is used to calculate the equipment health index and predict the health change trend. The warning judgment threshold is adjusted according to the priority coefficient of the task priority solution step to generate a dynamic health warning signal. S4, Resource Scheduling Decision: Receives dynamic health warning signals from the health status assessment step, performs full-attribute digital modeling of support resources and constructs a support resource efficiency matrix, uses a ternary coupled scheduling algorithm to calculate task-equipment-resource matching decision variables, generates a support resource scheduling plan and sends it to the execution terminal; S5, Closed-loop parameter correction: Collect data on the execution process of the guarantee scheme in the guarantee resource scheduling decision-making step, compare the execution data with the model prediction data, correct the parameters of the algorithm used in the previous steps, and reconstruct the guarantee resource scheduling scheme based on the corrected parameters.

2. The equipment health diagnosis and training support method according to claim 1, characterized in that, In S1, the task priority calculation involves collecting multi-dimensional characteristic parameters of the training mission, specifically including mission type, execution stage, strategic value, time node, coverage area, participating forces, equipment dependence, environmental constraints, mission continuity, alternative feasibility, delay loss, and review weight. Tactical personnel, equipment support personnel, and training command personnel jointly assign weights to the above twelve types of characteristic parameters. Consistency verification is performed on the weight allocation results, and data with logical contradictions in weight allocation are eliminated. A spatiotemporal weighted priority algorithm is used, which combines the weights and standardized values ​​of various parameters, the remaining time of the mission from the key execution node, and the ratio of the strategic weight of the mission execution area to the total strategic weight of the entire training domain to calculate the task priority coefficient with a value range of 0 to 1. Based on the priority coefficient value, the mission is divided into three categories: key mission, routine mission, and auxiliary mission.

3. The equipment health diagnosis and training support method according to claim 1, characterized in that, In S1, the spatiotemporal weighted priority algorithm used in task priority calculation employs the following mathematical expression: in , , These are the weighting coefficients, and , This represents the task priority coefficient corresponding to the spatial location s at time t. This represents the weighting of the feature parameters for the i-th task class. This represents the standardized value of the feature parameter of the i-th task. This indicates the remaining time until the critical execution node is reached. This indicates the total duration of the task. This represents the strategic weight value of the task execution area. This represents the overall strategic weight value across the entire training and exercise domain.

4. The equipment health diagnosis and training support method according to claim 1, characterized in that, In S2, the fault graph reasoning process collects equipment operation monitoring data including vibration data, temperature data, pressure data, speed data, current data, voltage data, sealing data, and wear data of the equipment's power system, navigation system, hydraulic system, and electrical system. All equipment operation monitoring data are collected in real time through dedicated sensor terminals deployed on the equipment. The collected fault history data includes historical fault records since the equipment was commissioned, maintenance history of each component, various fault handling cases, and maintenance cycle information of each component. Among them, historical fault records include fault occurrence time, fault location, fault manifestation, and fault level; component maintenance history includes maintenance time, maintenance component, maintenance method, and maintenance consumables usage; fault handling cases include fault type, handling process, tools used, and handling time; and maintenance cycle information includes the routine maintenance cycle of each component and the basis for adjusting the maintenance cycle under special working conditions. After the data collection is completed, a spatiotemporal knowledge graph of equipment faults is constructed and incremental learning is performed. The root cause node of the fault is located and the fault propagation path is deduced through a graph attention network.

5. The equipment health diagnosis and training support method according to claim 1, characterized in that, In S2, during fault graph reasoning, when constructing the equipment fault spatiotemporal knowledge graph, equipment physical components, sensor points, fault modes, and historical fault events are used as entity nodes. Among them, equipment physical component nodes include component model, installation location, and functional parameters; sensor point nodes include installation location, monitoring parameters, and measurement range; fault mode nodes include fault type and manifestation characteristics; and historical fault event nodes include fault occurrence time, occurrence location, and handling process. The internal physical connection relationship of the equipment, the causal transmission relationship of faults between components, and the temporal evolution dependency relationship of fault states are used as association edges to build an initial equipment fault spatiotemporal knowledge graph. During incremental learning, newly collected fault data, newly discovered component associations, and new fault handling cases are injected into the initial knowledge graph in real time, and the entity node features and association edge weights in the knowledge graph are updated synchronously. The update frequency is consistent with the equipment operation monitoring data collection frequency. After the update is completed, the root cause node of the fault is located and the fault propagation path is deduced through a graph attention network.

6. The equipment health diagnosis and training support method according to claim 1, characterized in that, In S2, during fault graph reasoning, when the graph attention network is running, the set of entity nodes and the set of associated edges of the equipment fault spatiotemporal knowledge graph are first input. The feature data of all entity nodes are normalized, and then the attention weight between each entity node and its neighboring entity nodes is calculated. The weight calculation process combines the entity node's own features, the features of neighboring entity nodes, and the attribute information of the associated edges to normalize and calibrate the calculated attention weights. Based on the calibrated attention weights, the features of each entity node are weighted and aggregated to obtain the fused features of each entity node. Then, the fused features are processed by a multilayer perceptron to output the association score of each entity node. The root cause node of the fault is located based on the association score, and the propagation path of the fault between entity nodes is inferred based on the association score.

7. The equipment health diagnosis and training support method according to claim 1, characterized in that, In S3, the mathematical expression used in the map fusion health algorithm for health status assessment is: in This represents the health index of the equipment at position s at time t. The weight coefficients representing the reasoning results of a knowledge graph. Let N represent the reasoning score of the spatiotemporal knowledge graph, where N represents the set of entity nodes in the knowledge graph, and R represents the set of related edges in the knowledge graph. This represents the standardized value of the sensor data of the j-th type of equipment. This represents the fault propagation attenuation coefficient. This represents a standardized value indicating the propagation distance from the root cause node of the fault to the monitoring component.

8. The equipment health diagnosis and training support method according to claim 1, characterized in that, In S4, during resource scheduling decision-making, when constructing the resource efficiency matrix, the resource type is used as the matrix row dimension, and the equipment failure mode is used as the matrix column dimension. Each element in the matrix corresponds to the efficiency value of a single resource for a single equipment failure mode. The efficiency value is obtained by weighting the resource's repair rate, allocation cost, response time, coverage radius, and available quantity. The weighting operation uses a preset fixed weight coefficient. The matrix row dimension covers all resource guarantees that have completed full-attribute digital modeling, and the matrix column dimension covers all marked failure modes of the equipment. After the matrix is ​​constructed, it is synchronously input into the ternary coupled scheduling algorithm for calculating the task-equipment-resource matching decision variables.

9. The equipment health diagnosis and training support method according to claim 1, characterized in that, In S4, the mathematical expression used in the resource scheduling decision-making process for the ternary coupled scheduling algorithm is: in This represents the guaranteed scheduling efficiency value corresponding to position s at time t. This represents the resource efficiency value of the k-th type of protection resource for the m-th type of failure mode. This represents the weight coefficients in multi-objective optimization. This represents the actual cost of protection. This indicates the preset maximum protection cost value. This indicates the actual response time. This indicates the preset maximum guaranteed response time value. This represents the actual utilization rate of guaranteed resources. This indicates the preset maximum guaranteed resource utilization rate.

10. The equipment health diagnosis and training support method according to claim 1, characterized in that, In S5, during closed-loop parameter correction, the collected execution data includes support resource location data, on-site operation progress data, equipment repair status data, support response time data, and resource consumption data. Support resource location data is updated at a fixed frequency, equipment repair status data includes a post-repair health index, and support response time data records the total time from the issuance of the dispatch command to the completion of maintenance. The corrected algorithm parameters include the weight coefficients of the spatiotemporal weighted priority algorithm, the graph fusion health algorithm, and the ternary coupled scheduling algorithm; the spatiotemporal feature factor of the spatiotemporal weighted priority algorithm; the fault propagation coefficient of the graph fusion health algorithm; and the resource efficiency coefficient of the ternary coupled scheduling algorithm. When the training mission is adjusted or the equipment health status changes abruptly, the support resource scheduling scheme is reconstructed based on the corrected parameters.

Images