Road train operation and maintenance decision-making system and method based on dynamic knowledge graph

Abstract

The invention relates to a road train operation and maintenance decision-making system and method based on a dynamic knowledge graph, belongs to the technical field of road train operation and maintenance, and solves the problem that fault mode recognition and performance degradation analysis in the prior art depend on empirical rules or static models and cannot dynamically adapt to complex and changeable operation environments. Comprising the following steps: S1, acquiring original data of a road train; s2, decomposing each equipment unit structure, and identifying key components and connection modes thereof; s3, establishing a data mining model; s4, establishing a comprehensive reliability evaluation model of the equipment, and calculating to obtain a reliability index of each key component of each equipment unit; s5, calculating equipment health indexes, and dividing equipment states; s6, constructing an equipment knowledge graph; s7, performing global reliability analysis based on the equipment knowledge graph, and determining a fault risk sequence and a weak link as a reliability evaluation result; and S8, generating a maintenance plan.

Description

Technical Field

[0001] This invention relates to the field of highway train operation and maintenance technology, specifically to a highway train operation and maintenance decision-making system and method based on dynamic knowledge graphs. Background Technology

[0002] With the continuous development of intelligent transportation systems, highway trains, as a highly efficient multi-car transport tool, have seen their operational stability and maintenance response speed become key factors in ensuring transport service quality and reducing operating costs. Highway trains typically consist of a lead car and multiple accompanying cars. Each car contains several key components, such as the power system, braking system, and suspension system. Each component corresponds to multiple state monitoring variables, including temperature, vibration, pressure, and current. Due to the large number of cars and the complexity of components, the total number of state variables for the entire highway train is enormous and intricately interconnected, making it difficult to accurately model and jointly predict the overall system operating status. Currently, highway train operation and maintenance management generally adopts time-based preventive maintenance or fault-driven corrective maintenance strategies, failing to fully utilize the massive amounts of data generated during equipment operation for health status assessment and fault prediction. Faced with multi-source heterogeneous data, existing systems lack effective data fusion and knowledge association mechanisms, making it difficult to achieve real-time monitoring and predictive maintenance of equipment health status. Traditional methods rely on empirical rules or static models for fault mode identification and performance degradation analysis, which cannot dynamically adapt to complex and changing operating environments, leading to delayed operation and maintenance decisions and insufficient system reliability.

[0003] Chinese invention patent application publication number CN119646473A, entitled "An Intelligent Fault Prediction Model Based on Machine Learning and Big Data Analysis," achieves fault prediction through machine learning and big data analysis. However, this model does not fully consider the deep fusion of multi-source heterogeneous data and the whole life cycle health management.

[0004] Chinese invention patent application publication number CN117994728A, invention title "Intelligent Operation and Maintenance Management System for High-speed Trains Based on Digital Twins", realizes intelligent management of operation and maintenance of high-speed trains through digital twin technology. However, this method has limitations in real-time data-driven dynamic decision support and multi-dimensional fault prediction analysis.

[0005] Therefore, there is a need in this field for improved decision-making methods for highway train operation and maintenance, in order to integrate multi-source heterogeneous data from different vehicles and components to achieve accurate prediction of equipment health status and maintenance decisions. Summary of the Invention

[0006] To address the aforementioned problems, this invention proposes a highway train operation and maintenance decision-making system and method based on dynamic knowledge graphs. This system integrates multi-source heterogeneous data from different vehicles and components to construct a full lifecycle database for equipment. It utilizes graph neural networks to achieve intelligent correlation analysis between failure modes and performance degradation, and then dynamically generates optimal maintenance strategies based on health status assessment and remaining life prediction. This method effectively solves the challenge of joint modeling and prediction of highway trains under multi-vehicle and multi-state conditions, significantly improving the intelligence level of operation and maintenance decision-making and the overall reliability of the system.

[0007] This invention addresses the challenges of multi-source heterogeneous data fusion, insufficient fault mode identification capabilities, and low levels of intelligent maintenance decision-making in existing highway train operation and maintenance management. It proposes a highway train operation and maintenance decision-making system and method based on a dynamic knowledge graph. By constructing a dynamic knowledge graph with spatiotemporal evolution capabilities, it enables real-time reasoning and adaptive maintenance decisions regarding equipment health status. First, multi-source data fusion technology is used to integrate heterogeneous data from different sensors such as vibration, temperature, and batteries to construct a database covering the entire equipment lifecycle. Second, a dynamic knowledge graph is constructed based on graph neural networks to achieve intelligent correlation and evolutionary modeling between fault modes and performance degradation mechanisms. Finally, by combining health status assessment and remaining life prediction results, the optimal maintenance strategy is automatically generated, thereby significantly improving the efficiency and reliability of highway train operation and maintenance.

[0008] According to one embodiment of the present invention, a method for highway train operation and maintenance decision-making based on dynamic knowledge graphs is provided, characterized in that it includes: S1: Acquire the raw data of the road train and preprocess it to obtain the model input data; the raw data includes historical fault data extracted from the equipment maintenance system of the road train, real-time operating data of the equipment collected from the sensor network as multi-source heterogeneous data, and maintenance logs of each equipment unit collected; the road train includes a lead car and at least one accompanying car following the lead car, the entire road train is represented by equipment, and each vehicle in the road train is represented by equipment unit; S2: Perform hardware structure characteristic analysis and operating environment analysis, decompose the structure of each equipment unit, identify key components and their connection methods; record the operating environment parameters of the equipment units; obtain the structural topology list of the equipment units and the initial values ​​of the performance degradation model parameters of the key components, and establish the performance degradation model of the key components; the key components include the power system, braking system and suspension system of each vehicle; S3: Establish a data mining model, receive multi-source heterogeneous data, use machine learning algorithms to extract the degradation curves or failure curves of equipment units under different working conditions and time periods, calculate the comprehensive degradation index of key components of equipment units at each time, identify the degradation mode of equipment units, and determine the degradation characteristics under different working conditions and time periods. S4: Establish a comprehensive reliability assessment model for the equipment, calculate the reliability indicators of each key component of each equipment unit based on historical failure data, and generate a comprehensive reliability score for each key component of each equipment unit; among which the reliability indicators include mean time between failures, failure rate, and coefficient of variation of failure interval; S5: Calculate the equipment health index using the dynamic Bayesian network method, and classify the equipment status into three levels: healthy, sub-healthy, and severely degraded; S6: Establish a set of entity nodes and a set of relationship edges, and construct a device knowledge graph in the form of entity-relation-entity; S7: Based on the device knowledge graph, a global reliability analysis is performed using a graph neural network algorithm to determine the failure risk ranking and weak points, which serve as the reliability assessment result; S8: Set dynamic thresholds for health index and prediction thresholds for remaining service life. Based on equipment status and reliability assessment results, determine and trigger the activation of the intelligent maintenance system, generate maintenance plans, and provide them to the maintenance team of the highway train.

[0009] Optionally, in the raw data of S1: Historical fault data includes the number of faults, fault time, fault type, and maintenance measures. Real-time operational data includes temperature, pressure, and vibration data, presented as multi-source heterogeneous data; The maintenance log is a detailed record of all maintenance procedures, including maintenance time, replaced parts, and maintenance personnel.

[0010] Optionally, the preprocessing of the raw data in S1 includes: Missing values ​​in the real-time running data of the original data are filled using the mean imputation method to obtain the sample mean of the numerical data in the original data, which is used as the original data after imputation. The DBSCAN clustering algorithm is used to identify and process outliers in the original data after filling, resulting in DBSCAN-processed data. The data processed using DBSCAN will be standardized using the Min-Max standardization algorithm to obtain the model input data.

[0011] Optionally, the comprehensive degradation indices of key components of the computing device unit in S3 at each time step include: Based on the structural topology list of equipment units, the preprocessed real-time operating data is grouped according to the vehicle-key component granularity; The grouped real-time running data is time-aligned according to a unified time base to form a synchronized multidimensional time series. Based on the statistical correlation strength between each variable in historical fault records and maintenance logs and braking system failure events, fixed weighting coefficients are pre-set; By combining the set fixed weight coefficients, the comprehensive degradation index of the key components of the equipment unit at each time moment is calculated, which is used to characterize the overall performance degradation degree of the key components of the equipment unit at each time moment.

[0012] Optionally, the degradation modes of the identified device unit in S3 include: For the comprehensive degradation index of each key component of the equipment unit at each time, the least squares method is used to fit the linear degradation curve. The degradation curves of all key components of all equipment units are used as input to perform K-means clustering analysis to obtain clustering results and identify degradation patterns. The types of degradation patterns include slow linear degradation pattern, accelerated degradation pattern, step degradation pattern and periodic modulation degradation pattern.

[0013] Optionally, in S4, a comprehensive reliability score is generated for each key component of each device unit, including: Based on historical failure data, the average failure interval time is calculated, which is the sample mean of the normal operating time of each equipment unit or key component between two consecutive independent failures. Calculate the failure rate based on the mean time between failures (MTBF). Calculate the coefficient of variation of the failure interval based on the mean time between failures (MTBF). Based on the obtained mean time between failures, failure rate, and coefficient of variation of the failure interval, normalization is performed, and combined with preset weights, to obtain the comprehensive reliability score of the equipment unit or key component.

[0014] Optionally, S6 includes: By treating key equipment components, degradation modes, and degradation factors as entity nodes, a set of entity nodes is obtained; Establish a set of relation edges to determine the connection relationships between components, the correlation between degradation modes and degradation factors, and treat each correlation relationship as a relation edge to obtain the set of relation edges; Based on the set of entity nodes and the set of relation edges, entities and relations are organized into a graph structure to obtain a device knowledge graph.

[0015] Optionally, S7 includes: S7.1: Establish a graph attention network, receive the device knowledge graph, process it to obtain the final embedded representation of each node, so as to capture its structural and attribute information in the device knowledge graph; S7.2: Based on the final embedded representation of the obtained nodes, calculate the failure risk of each node, sort the key components, and use a linear classifier to identify high-risk components; S7.3: Analyze the correlation between key components, calculate the attention weight between each node, and locate potential weak links.

[0016] Optionally, S8 includes: S8.1: Based on the equipment status and reliability assessment results, use the statistical distribution characteristics based on historical operating data to set dynamic thresholds for health index and prediction thresholds for remaining service life; S8.2: Trigger maintenance decision: Set a threshold for the predicted remaining useful life of the equipment unit or critical component. When the predicted remaining useful life is less than the predicted threshold, a maintenance decision is triggered. The predicted remaining useful life of the equipment unit or critical component is the time required from the current moment until the performance of the equipment unit or critical component degrades to the failure threshold. S8.3: Establish an optimization model with the goal of minimizing maintenance costs. Based on the equipment status and the failure risk of each node, generate a maintenance task list, allocate maintenance resources, and schedule maintenance time.

[0017] According to another embodiment of the present invention, a highway train operation and maintenance decision-making system based on dynamic knowledge graph is provided, which executes the highway train operation and maintenance decision-making method based on dynamic knowledge graph, including: a data acquisition module, a data mining and machine learning module, a reliability assessment module, an equipment health index calculation module, an equipment knowledge graph component module, a risk correlation analysis module, and a maintenance timing determination module.

[0018] The purpose of this invention is to provide a highway train operation and maintenance decision-making system and method based on dynamic knowledge graphs, which aims to solve problems such as difficulty in fusion of multi-source heterogeneous data, insufficient ability to identify new faults, and low level of intelligence in collaborative maintenance decision-making, thereby improving the efficiency and reliability of highway train operation and maintenance.

[0019] Compared with the prior art, the highway train operation and maintenance decision-making system and method based on dynamic knowledge graph provided by the present invention has at least the following beneficial effects.

[0020] 1. To address the issue of poor consistency in the fusion of multi-source heterogeneous data during highway train operation and maintenance, a data standardization and ontology processing method based on dynamic knowledge graphs is proposed. This method, through data preprocessing in step S1, unifies data format and semantic structure, resolving analysis errors caused by inconsistencies in format and scale between different data sources, and significantly improving the accuracy and reliability of data processing.

[0021] 2. To address the problem of complex data coupling relationships and difficulty in effective modeling under multi-vehicle collaborative operation conditions, a knowledge graph is used to explicitly model data entities and their relationships, and a graph neural network algorithm is combined for in-depth analysis, which improves the system's ability to understand complex coupling relationships and significantly improves the accuracy of fault diagnosis and the efficiency of operation and maintenance decisions.

[0022] 3. To address the lack of real-time monitoring and predictive maintenance capabilities in traditional operation and maintenance methods, a knowledge graph with dynamic update capabilities is constructed and graph neural network algorithms are integrated to achieve real-time assessment of equipment health status and prediction of remaining lifespan. This improves the timeliness and foresight of maintenance strategies, effectively reduces unexpected downtime, extends equipment lifespan, and enhances the safety and operational efficiency of the transportation system. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly introduced below. The features and advantages of the present invention can be more clearly understood by referring to the accompanying drawings. The accompanying drawings are schematic and should not be construed as limiting the present invention in any way. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating a highway train operation and maintenance decision-making method based on a dynamic knowledge graph, provided according to an embodiment of the present invention.

[0025] Figure 2 The flowchart illustrates the construction and calculation of reliability assessment indicators in a road train operation and maintenance decision-making method based on dynamic knowledge graphs, provided according to an embodiment of the present invention.

[0026] Figure 3 The flowchart illustrates the risk association analysis based on knowledge graphs in the highway train operation and maintenance decision-making method based on dynamic knowledge graphs provided according to an embodiment of the present invention. Detailed Implementation

[0027] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0028] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein. Therefore, the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0029] The following detailed description, with reference to the accompanying drawings, illustrates a road train operation and maintenance decision-making system and method based on a dynamic knowledge graph, according to embodiments of the present invention. The road train operation and maintenance decision-making method based on a dynamic knowledge graph, according to embodiments of the present invention, constructs a dynamic knowledge graph for road train systems with "multi-car formations and multi-component coupling," and uses graph neural networks to achieve joint reasoning and maintenance decisions based on health status across vehicles and components. A road train may include a lead car and at least one accompanying car. Each vehicle in the road train may be an autonomous heavy-duty truck.

[0030] like Figure 1 As shown, a highway train operation and maintenance decision-making method based on dynamic knowledge graphs according to an embodiment of the present invention includes the following steps.

[0031] S1: Acquire the raw data of the highway train and preprocess it to obtain the model input data. Among them, the set equipment includes all vehicles in the highway train (lead car and each accompanying car), that is, the system regards each car as an independent but related equipment unit; the key components of each equipment unit are the power system, braking system, suspension system, etc., which are the basic objects of condition monitoring and fault analysis; the multi-source heterogeneous data are: various sensor data (e.g., temperature, vibration, pressure, etc.), fault records, maintenance logs, etc. on each car (i.e., each equipment unit), organized according to the three-dimensional structure of "car-component-time".

[0032] S1.1: Obtain raw data from the road train, including: extracting historical fault data from the road train's equipment maintenance system, including the number of faults, fault time, fault type, maintenance measures, etc.; collecting real-time operating data of the equipment from the sensor network installed on the road train, including temperature, pressure, vibration, etc.; and collecting maintenance logs (i.e., detailed records of each maintenance), including maintenance time, replaced parts, maintenance personnel records, etc.

[0033] S1.2: The raw data obtained above is preprocessed to obtain the model input data. The specific process is as follows.

[0034] S1.2.1: For the real-time running data in the original data, use the mean imputation method to fill in the missing values ​​to obtain the sample mean of the numerical data in the original data.

[0035] Mean imputation involves using the calculated mean to fill in the missing values ​​in the original data sample, resulting in imputed original data. Mean imputation is performed within the time series of the same physical quantity, component, or vehicle. A "sample" is an observation of a specific monitoring variable over time (i.e., a data point in the time series).

[0036] The mean-filling method for missing values ​​can be applied only to numerical variables in real-time operational data, excluding historical fault data and maintenance logs. In this implementation, only numerical variables with the same physical meaning have their means calculated within their respective sequences. Optionally, mean filling is performed within the time series of sensors of the same vehicle, component, and type, without crossing types, components, or vehicles.

[0037] S1.2.2: For the padded original data, use the DBSCAN clustering algorithm to identify and process outliers to obtain the DBSCAN-processed data.

[0038] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm. Its core idea is that high-density regions form clusters, while points in low-density regions are labeled as "noise," or outliers. The algorithm depends on two parameters: (eps): Neighborhood radius; MinPts: Minimum number of neighborhood points required to form a cluster. If a point is within its... Points with fewer than MinPts in their neighborhood are considered outliers. As an example, the DBSCAN clustering algorithm can be performed as follows.

[0039] 1) Time series of a certain variable (For example, the temperature data of a certain component of a vehicle) is divided into fixed time windows, with a window length of 1 hour and a sliding step of 30 minutes.

[0040] 2) Within each window, the observation sequence is treated as a one-dimensional set of data points, using only numerical values. As input features.

[0041] 3) Run the DBSCAN algorithm independently on the data within each window, with MinPts set to 5. Set it to 1.8 times the standard deviation of the data in this window.

[0042] 4) Mark the data points with a label of -1 in the DBSCAN output as outliers.

[0043] 5) Replace all outliers with the median of the non-outliers in the corresponding window.

[0044] 6) Concatenate all the processed data in chronological order, prioritize the retention of non-outlier values ​​in overlapping regions, and finally generate preprocessed data of the same length as the original sequence.

[0045] It should be understood that in other implementations, the specific variables, parameters, execution methods, etc. of the DBSCAN clustering algorithm can be adjusted and changed as needed.

[0046] S1.2.3: The data processed using DBSCAN will be further standardized using the Min-Max standardization algorithm:

[0047] in, It is standardized data. This is the data processed by DBSCAN. and These are the minimum and maximum values ​​of the data, respectively. To find the minimum value function, This is the function for finding the maximum value.

[0048] Based on the above Min-Max standardized data As input data for the model, it can be directly used in subsequent machine learning modeling and knowledge graph construction. Min-Max standardization is performed independently within each component variable of each vehicle (i.e., the lead vehicle and accompanying vehicles).

[0049] S2: Perform hardware structure characteristic analysis and operating environment analysis, decompose the equipment unit structure, identify key components and their connection methods; record equipment unit operating environment parameters, including temperature, humidity, vibration, etc.; obtain the structural topology list of equipment units and the initial values ​​of performance degradation model parameters for key components, and establish performance degradation models for key components for mechanistic analysis. The aforementioned key components may include the power system, braking system, suspension system, etc., of each vehicle in a highway train.

[0050] The following explanation uses one of the trains accompanying a highway train as an example to illustrate the execution method of S2.

[0051] 1) Decomposition equipment unit structure: The vehicles are disassembled layer by layer according to their functional modules: Vehicle → Subsystem → Key Components → Sub-assemblies

[0052] 2) Identify key components (as explicitly listed in the instruction manual): Powertrain: Diesel engine, gearbox, drive motor (if electric drive); Braking system: master cylinder, wheel-side brakes, ABS control unit, air / hydraulic lines; Suspension system: air springs, shock absorbers, stabilizer bars.

[0053] 3) Identify the connection method (i.e., physical or functional association): Physical connection: Engine → (via driveshaft) → Gearbox "Master brake cylinder → (through hydraulic lines) → left front wheel brake"; Functional dependencies: "ABS control unit ← (receive signal) — wheel speed sensor" "Air spring ↔ (working together) - shock absorber".

[0054] These structures and relationships will serve as the initial skeleton for entities (nodes) and relationships (edges) in the knowledge graph.

[0055] 4) Establish the performance degradation model for key components as follows:

[0056] in, Indicates the time of critical components Performance, It is the initial performance of key components. It is the rate of degradation. For time.

[0057] The above theoretical model is not used directly for calculation. Its functional form (exponential decay) can be used as one of the candidate models for fitting the degradation curve. It can be used to provide a mechanism reference for S3 and a priori properties of components for S6.

[0058] The performance degradation models of the above key components are theoretical models that can be calibrated in subsequent steps.

[0059] Initial performance of key components These are the design performance indicators of components in their brand-new factory condition, which can be obtained from: equipment technical manuals (such as "new motor rated efficiency 95%)", type test reports, industry standards or manufacturer nominal values, etc.

[0060] Equipment unit operating environment parameters (e.g., temperature, humidity, vibration, etc.) are external operating condition variables that can be used in subsequent steps, such as as covariates to correct the degradation rate in S3. (e.g., high temperature accelerates degradation); added to the knowledge graph as an environmental entity in S6 (e.g., "Braking system, affected by ambient temperature").

[0061] The data acquired by S1 is the current or historical operating status, which can be used for updates or calibration. and However, it is not directly defined. For example, the actual residual coefficient of friction of the brake pads of a vehicle that has been used for two years may have dropped to 0.38, but the coefficient of friction in the model may not be defined. The value remains at 0.45 (nominal new part value), while the current performance P(t) = 0.38 is derived from the S1 data. Degradation rate. Then it can be corrected by fitting historical data of S1 (e.g., actual λ=0.0005 / hour).

[0062] The results obtained in S2 may include: a list of equipment structure topologies (used for data grouping in S3 and map skeleton construction in S6) and initial values ​​of performance degradation model parameters for key components. , This serves as the initial attribute for component entities in the knowledge graph and provides theoretical initial values ​​or a basis for model form selection during S3 degradation curve fitting. The equipment structure topology list can include a list of key components for all vehicles (lead vehicle and accompanying vehicles), a table of connections / dependencies between components (e.g., a triple: <brake master cylinder, connected to, left front wheel brake>); and initial values ​​for the performance degradation model parameters of key components, which can include theoretical or nominal values ​​(non-real-time measurements) for the initial performance and degradation rate of each key component.

[0063] It should be understood that the above is an exemplary description. In other embodiments, the specific execution method, process, parameter settings, etc. of S2 can be adjusted and changed as needed.

[0064] S3: Establish a data mining model, receive multi-source heterogeneous data, use machine learning algorithms to extract degradation curves or failure curves of equipment units under different operating conditions and time periods, and identify the degradation modes of equipment units. Among them, the multi-source heterogeneous data can mainly come from S1, partially rely on the semantic tags (component attribution, etc.) provided by S2, but do not directly use the structural model of S2.

[0065] The degradation curve of a device unit can be an empirical curve fitted from the actual operating data of S1. The following example illustrates how the degradation curve of S3 is generated.

[0066] 1) For each key component (such as "Vehicle Braking System No. 2"), extract its long-term monitoring sequence, such as oil temperature, pressure, and wear indicators over the past 6 months. In S1, real-time operating data of the equipment has been collected from the sensor network and processed through missing value imputation, outlier handling, and standardization to form a continuous and clean time series. These processed real-time operating data are stored cumulatively over time to form a historical monitoring database. The long-term monitoring sequence can be the S1 output extracted from this historical database.

[0067] 2) Integrate these multidimensional sensor data into a comprehensive degradation index, such as through principal component analysis (PCA) or a weighted health index.

[0068] 3) Using running time (or mileage) as the horizontal axis and comprehensive degradation index as the vertical axis, an empirical degradation trajectory is formed.

[0069] 4) Fit a function (such as linear, exponential, power law, etc.) to the trajectory to obtain the degradation curve.

[0070] In step S3, to extract the degradation characteristics of the equipment unit, the preprocessed real-time operating data output in step S1 is first grouped according to the "vehicle-key component" granularity based on the equipment unit structure information determined in step S2. Taking any braking system on a highway train as an example, its corresponding sensor signals include four categories: brake fluid temperature, hydraulic line pressure, effective value of wheel end vibration acceleration (RMS), and control current.

[0071] Next, the four types of signals after being grouped are time-aligned according to a unified time base (sampling period of 5 minutes) to form a synchronous multidimensional time series. Each signal has undergone missing value imputation, outlier correction and Min-Max standardization in step S1. The standardized values ​​range from [0,1], where the closer the value is to 1, the greater the degree to which the variable deviates from the normal operating condition.

[0072] Subsequently, based on the statistical correlation strength between each variable and braking system failure events in historical fault records and maintenance logs, fixed weight coefficients are pre-set. For example, taking the above example, the weight of brake fluid temperature can be set to 0.3, the weight of hydraulic line pressure to 0.4, the weight of wheel end vibration RMS to 0.2, and the weight of control current to 0.1, with the sum of all weights being 1.

[0073] Based on this, calculate each time step Comprehensive degradation index as follows:

[0074] in, The brake oil temperature, hydraulic pressure, vibration RMS, and control current of the equipment unit at time t are respectively... The standardized value.

[0075] The resulting comprehensive degradation index , is a scalar time series with values ​​in the range [0,1], used to characterize the braking system of the device unit in this example at time . The degree of overall performance degradation.

[0076] The resulting sequence can be used as a degradation proxy variable and input into subsequent degradation curve fitting and cluster analysis modules to identify potential degradation patterns.

[0077] The K-means clustering algorithm was used to identify potential degradation patterns and determine degradation characteristics under different working conditions and time periods.

[0078] Degradation patterns and degradation characteristics may specifically include the following items and contents.

[0079] Degradation mode refers to the typical pattern of equipment unit performance evolution over time. Examples include: slow linear degradation mode: the degradation rate is constant and small, and the performance declines steadily; accelerated degradation mode: the degradation rate increases over time, and the performance deteriorates rapidly with an upward convex curve; step degradation mode: the performance recovers abruptly after maintenance or component replacement, and then begins to degrade again; periodic modulated degradation mode: fluctuations synchronized with the operating cycle (such as shifts or routes) are superimposed on the overall degradation trend.

[0080] Degradation features are used to quantify specific parameters describing degradation patterns, and may include: degradation onset time. : The time point at which the comprehensive degradation index D(t) first exceeds the health threshold; Average degradation rate a: the slope of the degradation curve (e.g., a in the linear model D(t)=at+b); Maximum curvature or acceleration: the extreme value of the second derivative used to identify the accelerated degradation stage; Warning window length: the time interval from the sub-healthy state (e.g., D(t)=0.6) to the failure threshold (e.g., D(t)=0.9); Environmental sensitivity index: the percentage increase in degradation rate under operating conditions such as high temperature (>40℃) and high humidity (>80%RH); Periodic fluctuation frequency and amplitude: for periodic modulation modes, extract its main frequency (e.g., 24 hours) and fluctuation amplitude.

[0081] The aforementioned degradation patterns and characteristics are stored in a structured manner and, as "failure mechanism" class entities and their attributes, are input into the S6 knowledge graph to support the risk propagation analysis in S7 and the remaining lifetime prediction in S8.

[0082] In step S3, for the comprehensive degradation index sequence D(t) of each key component under specific operating conditions, a linear degradation curve is fitted using the least squares method:

[0083] in, Cumulative operating time of key components (unit: hours). This refers to the standardized degradation index constructed above. Parameters and It can be a regression coefficient calculated using historical operating data of the key component.

[0084] Specifically, this refers to all valid ( ) components of the critical component since it was put into use or last major overhauled up to the present moment. , Given a sample set of data points, find the solution that minimizes the sum of squared residuals. and :

[0085] in, Indicates the sequence number of the sample point. This indicates the total number of valid data points. Indicates the first The cumulative running time (in hours) for each data point.

[0086] Parameters obtained This indicates the average degradation rate of the component under current operating conditions. This represents the initial degradation offset. Due to differences in the location and usage intensity of different vehicles in a road train, the degradation rates of various components vary significantly. For example, the braking system of the lead car, due to frequent starts and stops, can have a degradation rate of... The braking system at the rear of the formation is used less frequently, and its degradation rate can be [missing information]. This difference was accurately captured and used for subsequent differential maintenance decisions.

[0087] Furthermore, the degradation curves of all key components of the vehicle (from their respective...) and Using the definition as input, perform K-means clustering analysis, with the clustering dimension including degradation rate. Trend of curve slope change, time of degradation onset (For example, it can be defined as) The data includes the moment when the health threshold of 0.3 is first exceeded, and the second derivative of the degradation trajectory (used to identify acceleration phases). The clustering results form several degradation pattern categories, each corresponding to a typical performance evolution pattern.

[0088] The degradation patterns identified in this implementation can specifically include the following four categories: Slow linear degradation mode: degradation rate Constant and small (e.g.) D(t) rises steadily over time, reflecting the normal wear process, which is typical of air springs in suspension systems that operate stably for a long time. Accelerated degradation mode: The degradation rate is low in the early stages and high in the later stages. Significantly increased (confirmed by local slope detection via sliding window). An upward convex growth pattern indicates a risk of sudden failure and is commonly seen in the aging of motor insulation in power systems. Step degradation mode: A sudden drop occurs at a certain moment (e.g., from 0.7 to 0.2), and then accumulation resumes. This corresponds to a maintenance or component replacement event, such as performance recovery after brake pad replacement. Periodic modulation degradation mode: The overall upward trend is superimposed with periodic fluctuations, and the fluctuation cycle is consistent with the operating schedule (such as a cycle of 24 hours). This reflects the cumulative effect of thermal-mechanical fatigue in mountain transportation, which is often seen in the transmission system of the lead car.

[0089] The quantitative characteristics corresponding to each degradation mode include: degradation onset time. Average degradation rate Maximum curvature value (used to determine the degree of acceleration), from sub-healthy state ( ) to failure threshold ( The time interval (i.e., the length of the warning window) and the frequency and rate of occurrence of this pattern under environmental conditions such as high temperature (>40℃) and high humidity (>80%RH).

[0090] Furthermore, the degradation patterns and their characteristic parameters obtained above can be stored in a structured manner and used as "failure mechanism" type entities and their attributes. In step S6, a dynamic knowledge graph can be injected to support the risk propagation analysis in step S7 and the remaining life prediction and maintenance timing determination in step S8.

[0091] S3 focuses on the performance degradation trend in the "not failed" state, while S4 focuses on the statistical patterns of "occurred" failure events. S4 is a pure failure event-driven reliability analysis, independent of the degradation modeling of S3.

[0092] S4: Establish a comprehensive reliability assessment model for the equipment. Based on reliability indicators, generate comprehensive reliability scores for each key component of each equipment unit. The obtained comprehensive reliability scores can be used to calculate the equipment health index in S5, where R is composed of MTBF, Composed of CV and other factors, it reflects The confidence and stability of R can be used in engineering implementations for weighted correction or uncertainty calibration. It can also be used for knowledge graph construction in S6. In S7, in GNN node embedding initialization or attention mechanisms, R can be used as a prior weight to suppress false risk signals from high-reliability components and improve inference accuracy. It can also be used in S8 for prior repair of low-scoring components. Figure 2 As shown, a comprehensive reliability assessment model for equipment can be established by combining reliability indicators such as MTBF (Mean Time Between Failures), failure rate, and coefficient of variation of failure intervals. The specific process is as follows. The aforementioned reliability indicators can be calculated based on historical failure data obtained in S1; this data can come from the historical failure data output by S1, including the historical failure occurrence time series of each vehicle and each key component. S4 may specifically include the following steps.

[0093] Based on historical fault data obtained from S1, the mean fault interval time is calculated: the sample mean of the normal operating time of the same equipment unit or critical component between two consecutive independent faults.

[0094] in, It is the first Inter-failure time. It is the total number of effective fault intervals (number of faults - 1). Indicates the number of times the fault occurred. This indicates the mean time between failures.

[0095] Calculate the failure rate based on the mean time between failures (MTBF):

[0096] in, This refers to the failure rate.

[0097] Calculate the coefficient of variation of the failure interval based on the mean time between failures:

[0098] in, The fault interval variation coefficient is . It is the standard deviation of the interval between failures. It is the average of the intervals between failures. = .

[0099] By integrating the above indicators, a comprehensive reliability assessment model is constructed, resulting in a comprehensive reliability score. A normalized weighted scoring method is employed, as explained below: Set the MTBF of a key component of a device unit. CV and CV are respectively:

[0100]

[0101] Among them, subscript Index for key components.

[0102] Normalize each indicator: The higher the MTBF, the better:

[0103] can be It was rewritten.

[0104] The smaller the better:

[0105] A smaller CV is better (more stable):

[0106] in, , , Key components MTBF, CV normalization intermediate variables.

[0107] Set weights (based on expert experience or historical data analysis):

[0108]

[0109]

[0110] Ultimately, the overall reliability score was... for:

[0111] in, Overall reliability score A higher value indicates a higher reliability of the critical component.

[0112] The result is a comprehensive reliability score generated for each critical component (such as the lead vehicle braking system and the No. 3 vehicle power system) of each vehicle (i.e., each equipment unit). At the same time, the original indicators (MTBF, CV) serves as the basis for fine-grained analysis.

[0113] S5: The equipment health index is calculated using a dynamic Bayesian network method, classifying equipment status into three levels: healthy, sub-healthy, and severely degraded. This includes the following steps. In S5, "equipment" refers to the entire highway train convoy as a single operating unit, i.e., a complete transportation system consisting of a lead car and multiple accompanying cars.

[0114] S5.1: Based on the failure rates of each key component of the equipment obtained from the comprehensive reliability assessment model, calculate the health index, and use a dynamic Bayesian network to calculate the overall health index of the equipment. The calculation formula is:

[0115] in, It is the first The failure rate of each key component (which can be derived from the failure rate of each key component calculated in step S4). ), It is runtime (time variable) This represents the current cumulative running time, in hours. It can be a unified time base for the entire system, ensuring that the timing of each step is aligned. Indexes representing the key components of the equipment. This represents the total number of all critical components of the equipment.

[0116] S5.2: Equipment Status Classification: Based on Health Index Classify device status and set health thresholds. and sub-health threshold The equipment status is divided into three levels: healthy, sub-healthy, and severely degraded.

[0117] in, This represents a preset health status threshold; This indicates the preset threshold for sub-health status.

[0118] In this step, states with a health index equal to or higher than a preset health status threshold are classified as healthy levels; states with a health index lower than a preset health status threshold but higher than or equal to a preset sub-health status threshold are classified as sub-health levels; and states with a health index lower than a preset sub-health status threshold are classified as severely degraded levels.

[0119] S6: Establish an equipment knowledge graph, constructing the knowledge graph in the form of entity-relationship-entity for information such as key components, degradation modes, degradation factors, and environmental variables of the equipment unit. The knowledge graph construction in S6 can rely on the output results of steps S2, S3, and S4. Specifically, S2 provides entity node types (e.g., braking system) and provides topological relationships between components (e.g., <brake master cylinder, connected to, wheel-end brake>); S3 provides degradation mode and degradation factor entities (e.g., "thermal fatigue degradation") and provides analysis results of the degradation mode-degradation factor association relationship (e.g., <brake overheating, leading to, accelerated degradation>); S4 provides evidence of environmental sensitivity and failure tendency, and supports the quantitative weights of the <environmental variable, influence, degradation mode> relationship.

[0120] A set of entity nodes is established, which includes key equipment components, degradation modes, degradation factors, and environmental variables. Key equipment components can be derived from the structural topology list of equipment units obtained from the equipment unit structure decomposition in S2; degradation modes can be derived from the K-means clustering results in S3; degradation factors can be derived from the correlation analysis between multi-source heterogeneous data and degradation in S3 (reflected in the causal explanation and feature extraction of degradation modes in S3; the associated operating conditions or triggers identified during the causal analysis of various degradation modes, these statistical or mechanistic correlations between multi-source data in S1 and the degradation trajectory D(t) constitute the source of degradation factors); environmental variables can be derived from operating environment parameters collected in S1 and recorded in S2. Entity identification refers to the structured technical elements output from the preceding steps.

[0121] Establish a set of relational edges to determine the connections between components, the correlations between degradation modes and degradation factors, and treat each correlation as a relational edge. The connections between components can use the physical / functional connection methods identified in S2. Relationship edges can be determined based on explicit or implicit causal, dependency, and influence logic from previous steps. The correlation between degradation modes and degradation factors can be obtained based on the statistical correlation results between degradation modes and operating conditions / sensors in S3. The influence of environmental variables on components can be obtained by combining the reliability indicators in S4 with the degradation characteristics in S3. The attribution of degradation factors to key components can be determined by the component labels in S1 data and the fusion process in S3.

[0122] Based on the set of entity nodes and the set of relation edges, a knowledge graph is constructed. The knowledge graph is stored using the Neo4j graph database, and the entities and relations are organized into a graph structure to obtain the device knowledge graph.

[0123] in, It is a set of entity nodes. It is a set of relation edges. This represents a device knowledge graph.

[0124] S7: Based on a device knowledge graph, and using a graph neural network algorithm, it ranks fault risks and component correlations to quickly locate potential weaknesses, enabling global reliability analysis of complex systems. For example... Figure 3 As shown, this step may specifically include the following steps.

[0125] S7.1: Input the device knowledge graph obtained in S6 into the Graph Attention Network (GAT). The GAT learns the embedding representations of nodes and edges through a message passing mechanism. The GAT can be a graph neural network model trained on the knowledge graph built on S6, and it is a deep learning model with the S6 graph as its topology. The two have a "data-model" relationship.

[0126] The layers of a Graph Attention Network (GAT) represent the embedding of each node by aggregating its neighboring nodes. The information has been updated:

[0127] in, It is a node The set of neighboring nodes, Represents a node Embedded representation, Represents a node The Embedded representation of each neighbor node Represents a node The neighbor node index, the superscript in the symbol This represents the layer index of a multi-layer graph neural network. It is an aggregation function, which can be, for example, summation, averaging, or max pooling.

[0128] In this implementation, GAT can be used as the specific implementation of GNN.

[0129] By stacking multiple layers of a graph neural network, the output of the final layer is the final embedding representation of each node. It can capture the structural and attribute information of the device in the device knowledge graph.

[0130] S7.2: Based on the final embedded representation of the obtained nodes, calculate the failure risk of each node, sort the key components, and use a linear classifier to identify high-risk components.

[0131] Calculate the failure risk of each node:

[0132] in, It is a weight matrix. It is a bias term. It is an activation function. Represents a node The risk of failure.

[0133] The failure risks of all critical components (i.e., the failure risks of each node) are sorted to identify high-risk components. Components with failure risk values ​​exceeding a set threshold can be designated as high-risk components, or the top few components with the highest failure risk values ​​(this number can be set as needed, for example, the top 3 components with the highest failure risk values) can be designated as high-risk components.

[0134] S7.3: Analyze the correlation between key components, calculate the attention weight between each node, quickly locate potential weak links, and realize global reliability analysis of complex systems.

[0135] By analyzing the attention weights or edge embeddings in graph neural networks, the degree of correlation between components can be quantified.

[0136] In Graph Attention Networks (GAT), nodes Its neighboring nodes attention weights It can be calculated in the following ways.

[0137] First, the embedding representation of nodes x and y is... , Perform concatenation and use a learnable weight matrix Transformation; Then utilize the shared attention vector Calculate the unnormalized score:

[0138] Finally, softmax normalization is performed on all neighbor nodes to obtain the attention weights:

[0139] in, Represents the input features (or previous layer embeddings) of nodes x and y; It is a learnable linear transformation matrix; This indicates vector concatenation. Represents a learnable attention vector; This represents the unnormalized attention score; , Represents the vector dimension.

[0140] The obtained nodes Its neighboring nodes attention weights The larger the value, the stronger the association between nodes x and y.

[0141] By analyzing these correlations, potential weaknesses can be quickly identified, enabling global reliability analysis of complex systems and yielding reliability assessment results. This is determined through attention weighting. Size quantifies the correlation strength between nodes: A larger value indicates a higher importance of information transmission between nodes x and y. For example, the value of an edge (x, y) is... If the attention weight of a node p is significantly higher than that of other edges, it indicates that the connection is crucial to the overall system stability. If all outgoing edges of a node p have high attention weights, it indicates that it is a key hub node. If the incoming edges of a node p generally have low attention weights, it may be an isolated or weakly coupled node.

[0142] Vulnerable links include high-risk components and high-risk interconnections: High-risk components: have a high probability of failure and may trigger system failure; High-risk interconnections: although the component itself is normal, if its connection relationship is critical (such as main control unit → actuator), once interrupted, it will lead to loss of function.

[0143] The fault risk ranking and weak points output by S7 are direct results of reliability assessment and can be used to guide maintenance priorities.

[0144] S8: Sets dynamic thresholds for the health index and prediction thresholds for remaining service life. Based on equipment status and reliability assessment results, it determines and triggers the activation of the intelligent maintenance system, generates a maintenance plan, and provides it to the maintenance team of the highway train. The equipment's operating status can be the vehicle health index H (continuous value) output by S5; the reliability assessment results can be the component-level failure risk ranking and structural weak points (discrete results) output by S7; the two complement each other.

[0145] S8.1: Based on equipment status and reliability assessment results, use the statistical distribution characteristics based on historical operating data to set dynamic thresholds for health indices and prediction thresholds for remaining useful life. Different health status thresholds are set based on historical data and equipment reliability indicators:

[0146]

[0147] in, and These are the mean and standard deviation of the health index, respectively. To calculate the coefficients, The dynamic health threshold of the health index This refers to the dynamic sub-health threshold of the health index.

[0148] S8.2: Trigger maintenance decision: Set the remaining useful life prediction threshold as follows When the predicted remaining useful life (RUL) is less than This triggers a maintenance decision.

[0149] Remaining useful life (RUL) refers to the remaining useful life from the current moment. From the point where the performance of a device unit or critical component degrades to the failure threshold. The required time. In S3, degradation curves have been fitted for each critical component, for example, failure thresholds have been set. Current level of degradation It can be calculated in real time through sensors.

[0150] make The failure time is obtained by solving: ( )

[0151] The failure time is obtained: .

[0152] S8.3: Generate Maintenance Plan: The intelligent maintenance system generates a maintenance task list, allocates maintenance resources, and schedules maintenance time based on the equipment status and fault risk information of each node.

[0153] For example, maintenance tasks can be one or more of the following: preventive maintenance, corrective maintenance, predictive maintenance, brake pad replacement, cooling system cleaning, sensor calibration, control software updates, and bolt fastening.

[0154] To establish an optimization model with the goal of minimizing maintenance costs:

[0155] in, Key components The maintenance cost Key components Downtime costs, It is a key component for maintenance Time, This represents a binary decision variable, indicating whether or not to repair the component. If the value is positive, it is included in the maintenance cost; otherwise, it is zero.

[0156] After solving the above optimization model, the solution obtained may include: each key component Maintenance decision variables Each key component Repair time The final maintenance task list may include: task details, execution time, responsible person, required resources, etc.

[0157] By solving this optimization model, the obtained solution can be used as the optimal maintenance plan to ensure the reliable operation of the equipment, thereby realizing global reliability analysis and intelligent maintenance management of complex systems.

[0158] Example 1 To better understand the present invention, the following describes an embodiment 1 of a road train operation and maintenance decision-making method based on a dynamic knowledge graph provided by an embodiment of the present invention.

[0159] Background setup: 1 lead car + 2 follower cars (3 cars in total); Research objective: The braking system of vehicle No. 2; Historical data: 6 months; Sensor data collected: brake fluid temperature, hydraulic pressure, wheel end vibration RMS, control current (sampling period 5 minutes).

[0160] The process of the highway train operation and maintenance decision-making method based on dynamic knowledge graph in Example 1 is as follows.

[0161] S1: Data Preprocessing Calculate the sample mean and fill in the missing values:

[0162] Sliding window DBSCAN An outlier (92℃) was identified and replaced with the window median of 86℃.

[0163] Min-Max Standardization

[0164]

[0165] Output the standardized oil temperature sequence.

[0166] S2: The braking system includes: master cylinder, wheel-end brakes, ABS unit, and initial performance (coefficient of friction). =0.95, theoretical degradation rate λ=0.0001 / h (manual value).

[0167] S3: Fusion of multi-dimensional signals (fixed weights)

[0168] Comprehensive degradation index: D(4000)=0.3×0.40+0.4×0.35+0.2×0.20+0.1×0.10=0.12+0.14+0.04+0.01=0.31 Fitting degeneracy path: 5256 within 4380 hours Point fitting: D(t) = 0.00018⋅t + 0.05 Clustering identification degradation patterns: The degradation rate a = 0.00018 (relatively small) is determined to be a slow linear degradation mode.

[0169] S4: Reliability Index Calculation The braking system has experienced three independent failures in the past two years; Failure interval: t1 = 1800h (from first use to first failure) t2 = 1500h Effective interval number J=2

[0170]

[0171]

[0172] Overall reliability score: R=0.633, indicating moderate to low reliability.

[0173] S5: Vehicle Health Index Calculation The formation consists of m=12 key components, and the current running time is t=4000h. For vehicle braking system No. 2:

[0174] It was determined to be severely degraded.

[0175] S6: Knowledge Graph Construction S4 data showed that MTBF decreased by 25% at high temperatures.

[0176] S7: GAT Risk Reasoning Calculate the risk: .

[0177] S8: Maintenance Decision RUL = 722 hours =1000 hours Repair triggered.

[0178] Another embodiment of the present invention provides a highway train operation and maintenance decision-making method based on dynamic knowledge graph, comprising the following steps: Step 1: Collect historical fault records, operation monitoring data and maintenance logs of highway trains through the data acquisition module, and clean and preprocess the raw data to ensure data quality and consistency; Step 2: Based on the data processing module, analyze the hardware structure, operating environment, and working characteristics of the key components of the highway train to identify the main factors affecting equipment performance; Step 3: Use data mining and machine learning models to identify the degradation trends and potential failure modes of equipment under different operating conditions; Step 4: Based on the reliability assessment module, construct an equipment reliability assessment system and select key reliability indicators for quantitative analysis; Step 5: Based on the equipment health index calculation model, classify and assess the current status of the equipment to form a health status profile; Step 6: Construct an equipment knowledge graph, integrating multi-dimensional information such as equipment structure, operating status, degradation mode, and maintenance records to achieve structured representation and dynamic updating of knowledge; Step 7: Using the risk correlation analysis module, sort the failure risk level and the correlation between components to identify weak links and key risk points in the system; Step 8: Based on the maintenance timing determination module, set threshold conditions for health index and remaining service life, and automatically trigger the intelligent maintenance system to generate maintenance plans and maintenance strategies to achieve predictive maintenance and intelligent decision support.

[0179] According to another embodiment of the present invention, a highway train operation and maintenance decision-making system based on dynamic knowledge graph is provided. The highway train operation and maintenance decision-making method based on dynamic knowledge graph, which executes any of the aforementioned embodiments, may include: a data acquisition module, which collects and gathers historical fault records, operation monitoring data, and maintenance logs of highway trains, and cleans and preprocesses the raw data to ensure data quality and consistency; a data mining and machine learning module, which identifies the degradation trends and potential failure modes of equipment under different operating conditions; a reliability assessment module, which constructs an equipment reliability assessment system and selects key reliability indicators for quantitative analysis; an equipment health index calculation module, which performs a graded assessment of the current state of the equipment to form a health status profile; an equipment knowledge graph component module, which integrates multi-dimensional information such as equipment structure, operating status, degradation modes, and maintenance records to achieve structured representation and dynamic updating of knowledge; a risk correlation analysis module, which sorts the fault risk level and component correlation degree to identify weak links and key risk points in the system; and a maintenance timing determination module, which sets threshold conditions for the health index and remaining service life, automatically triggering the intelligent maintenance system to generate maintenance plans and maintenance strategies, thereby achieving predictive maintenance and intelligent decision support.

[0180] All of the above-mentioned optional technical solutions can be combined in any way to form optional embodiments of the present invention, and will not be described in detail here.

[0181] In the description of this application, it should be understood that if terms such as "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential" appear, these terms indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0182] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order and method of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0183] It should be understood that the foregoing only illustrates some embodiments, and changes, modifications, additions, and / or variations can be made without departing from the scope and spirit of the disclosed embodiments. These embodiments are illustrative and not restrictive. Furthermore, the described embodiments relate to those currently considered most practical and preferred, and should be understood as not being limited to the disclosed embodiments, but rather intended to cover different modifications and equivalent arrangements included within the spirit and scope of those embodiments. Moreover, the various embodiments described above can be used in conjunction with other embodiments; for example, an aspect of one embodiment can be combined with an aspect of another embodiment to achieve yet another embodiment. Additionally, individual features or components of any given component can constitute another embodiment.

[0184] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A highway train operation and maintenance decision-making method based on dynamic knowledge graph, characterized in that, include: S1: Acquire the raw data of the road train and preprocess it to obtain the model input data; the raw data includes historical fault data extracted from the equipment maintenance system of the road train, real-time operating data of the equipment collected from the sensor network as multi-source heterogeneous data, and maintenance logs of each equipment unit collected; the road train includes a lead car and at least one accompanying car following the lead car, the entire road train is represented by equipment, and each vehicle in the road train is represented by equipment unit; S2: Perform hardware structure characteristic analysis and operating environment analysis, decompose the structure of each equipment unit, identify key components and their connection methods; record the operating environment parameters of the equipment units; obtain the structural topology list of the equipment units and the initial values ​​of the performance degradation model parameters of the key components, and establish the performance degradation model of the key components; the key components include the power system, braking system and suspension system of each vehicle; S3: Establish a data mining model, receive multi-source heterogeneous data, use machine learning algorithms to extract the degradation curves or failure curves of equipment units under different operating conditions and time periods, calculate the comprehensive degradation index of key components of equipment units at each moment, identify the degradation mode of equipment units, and determine the degradation characteristics under different operating conditions and time periods. S4: Establish a comprehensive reliability assessment model for the equipment, calculate the reliability indicators of each key component of each equipment unit based on historical failure data, and generate a comprehensive reliability score for each key component of each equipment unit. The reliability metrics include mean time between failures, failure rate, and coefficient of variation of the failure interval. S5: Calculate the equipment health index using the dynamic Bayesian network method, and classify the equipment status into three levels: healthy, sub-healthy, and severely degraded; S6: Establish a set of entity nodes and a set of relationship edges, and construct a device knowledge graph in the form of entity-relation-entity; S7: Based on the device knowledge graph, a global reliability analysis is performed using a graph neural network algorithm to determine the failure risk ranking and weak points, which serve as the reliability assessment result; S8: Set dynamic thresholds for health index and prediction thresholds for remaining service life. Based on equipment status and reliability assessment results, determine and trigger the activation of the intelligent maintenance system, generate maintenance plans, and provide them to the maintenance team of the highway train.

2. The highway train operation and maintenance decision-making method based on dynamic knowledge graph as described in claim 1, characterized in that, In the original data of S1: Historical fault data includes the number of faults, fault time, fault type, and maintenance measures. Real-time operational data includes temperature, pressure, and vibration data, presented as multi-source heterogeneous data; The maintenance log is a detailed record of all maintenance procedures, including maintenance time, replaced parts, and maintenance personnel.

3. The highway train operation and maintenance decision-making method based on dynamic knowledge graph as described in claim 2, characterized in that, The preprocessing of the raw data in S1 includes: Missing values ​​in the real-time running data of the original data are filled using the mean imputation method to obtain the sample mean of the numerical data in the original data, which is used as the original data after imputation. The DBSCAN clustering algorithm is used to identify and process outliers in the original data after filling, resulting in DBSCAN-processed data. The data processed using DBSCAN will be standardized using the Min-Max standardization algorithm to obtain the model input data.

4. The highway train operation and maintenance decision-making method based on dynamic knowledge graph as described in claim 1, characterized in that, The comprehensive degradation indices of key components of the computing device unit in S3 at various time points include: Based on the structural topology list of equipment units, the preprocessed real-time operating data is grouped according to the vehicle-key component granularity; The grouped real-time running data is time-aligned according to a unified time base to form a synchronized multidimensional time series. Based on the statistical correlation strength between each variable in historical fault records and maintenance logs and braking system failure events, fixed weighting coefficients are pre-set; By combining the set fixed weight coefficients, the comprehensive degradation index of the key components of the equipment unit at each time moment is calculated, which is used to characterize the overall performance degradation degree of the key components of the equipment unit at each time moment.

5. The highway train operation and maintenance decision-making method based on dynamic knowledge graph according to claim 1, characterized in that, The degradation modes of the identified device unit in S3 include: For the comprehensive degradation index of each key component of the equipment unit at each time, the least squares method is used to fit the linear degradation curve. The degradation curves of all key components of all equipment units are used as input to perform K-means clustering analysis to obtain clustering results and identify degradation patterns. The types of degradation patterns include slow linear degradation pattern, accelerated degradation pattern, step degradation pattern and periodic modulation degradation pattern.

6. The highway train operation and maintenance decision-making method based on dynamic knowledge graph according to claim 1, characterized in that, In S4, a comprehensive reliability score is generated for each key component of each device unit, including: Based on historical failure data, the average failure interval time is calculated, which is the sample mean of the normal operating time of each equipment unit or key component between two consecutive independent failures. Calculate the failure rate based on the mean time between failures (MTBF). Calculate the coefficient of variation of the failure interval based on the mean time between failures (MTBF). Based on the obtained mean time between failures, failure rate, and coefficient of variation of the failure interval, normalization is performed, and combined with preset weights, to obtain the comprehensive reliability score of the equipment unit or key component.

7. The highway train operation and maintenance decision-making method based on dynamic knowledge graph according to claim 1, characterized in that, S6 include: By treating key equipment components, degradation modes, and degradation factors as entity nodes, a set of entity nodes is obtained; Establish a set of relation edges to determine the connection relationships between components, the correlation between degradation modes and degradation factors, and treat each correlation relationship as a relation edge to obtain the set of relation edges; Based on the set of entity nodes and the set of relation edges, entities and relations are organized into a graph structure to obtain a device knowledge graph.

8. The highway train operation and maintenance decision-making method based on dynamic knowledge graph according to claim 1, characterized in that, S7 includes: S7.1: Establish a graph attention network, receive the device knowledge graph, process it to obtain the final embedded representation of each node, so as to capture its structural and attribute information in the device knowledge graph; S7.2: Based on the final embedded representation of the obtained nodes, calculate the failure risk of each node, sort the key components, and use a linear classifier to identify high-risk components; S7.3: Analyze the correlation between key components, calculate the attention weight between each node, and locate potential weak links.

9. The highway train operation and maintenance decision-making method based on dynamic knowledge graph according to claim 1, characterized in that, S8 includes: S8.1: Based on the equipment status and reliability assessment results, use the statistical distribution characteristics based on historical operating data to set dynamic thresholds for health index and prediction thresholds for remaining service life; S8.2: Trigger maintenance decision: Set a threshold for the predicted remaining useful life of the equipment unit or critical component. When the predicted remaining useful life is less than the predicted threshold, a maintenance decision is triggered. The predicted remaining useful life of the equipment unit or critical component is the time required from the current moment until the performance of the equipment unit or critical component degrades to the failure threshold. S8.3: Establish an optimization model with the goal of minimizing maintenance costs. Based on the equipment status and the failure risk of each node, generate a maintenance task list, allocate maintenance resources, and schedule maintenance time.

10. A highway train operation and maintenance decision-making system based on dynamic knowledge graphs, executing the highway train operation and maintenance decision-making method based on dynamic knowledge graphs as described in any one of claims 1 to 9, comprising: The system includes a data acquisition module, a data mining and machine learning module, a reliability assessment module, an equipment health index calculation module, an equipment knowledge graph component module, a risk correlation analysis module, and a maintenance timing determination module.

Images