A data analysis-based method and system for the maintenance management of rail transit vehicles

By constructing an estimated value of the impact of misalignment and an active risk period, and combining a recursive segmentation algorithm and a decision binary tree, the problem of identifying hidden faults of rail transit vehicles spanning the operation and rest periods in a hot and humid coastal environment was solved, achieving more accurate maintenance early warning and optimized maintenance cycle.

CN123118071APending Publication Date: 2026-11-17ZHEJIANG RAIL TRANSIT OPERATION MANAGEMENT GROUP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611345608.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-01
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing data analysis methods are insufficient to identify hidden fault risks of intercity rail transit vehicles during periods of operation and rest in hot and humid coastal environments, leading to a decrease in the accuracy of maintenance early warning.

Method used

By collecting vehicle operation data, environmental data, historical component status data, and virtual sensor data in hot and humid coastal areas, we construct estimated values ​​of misalignment impact and active risk periods. We then use a recursive segmentation algorithm to build a decision binary tree, mark crossover points, and optimize maintenance cycles.

Benefits of technology

It enables accurate identification and dynamic optimization of hidden fault risks, improving the accuracy of vehicle maintenance early warning and the ability to dynamically optimize maintenance cycles.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118071A_ABST
    Figure CN123118071A_ABST
Patent Text Reader

Abstract

This invention discloses a data analysis-based method for the maintenance and management of rail transit vehicles, comprising the following steps: S1, collecting vehicle operation data, vehicle operating environment data, component historical status data, and virtual sensor data from hot and humid coastal areas, and calculating the estimated value of misalignment impact; S2, extracting the virtual sensor data of the vehicle components corresponding to the estimated value of misalignment impact, identifying periodic risk periods, and determining active risk periods; S3, constructing a decision binary tree using the estimated value of misalignment impact and active risk periods as input features, and marking the crossover transition points; S4, dividing the lifespan intervals of vehicle components according to the crossover transition points, optimizing and outputting the recommended maintenance cycle for the current operating environment using a weighted index of maintenance cost and equipment availability. This invention overcomes the shortcomings of traditional rail transit vehicle maintenance and management methods that rely on operational data and ignore the evolution during rest periods, significantly improving the accuracy of vehicle maintenance early warning and the ability to dynamically optimize cycles under special environments.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of rail transit vehicle maintenance management technology, specifically, it relates to a rail transit vehicle maintenance management method and system based on data analysis. Background Technology

[0002] As a core component of urban agglomeration and metropolitan area transportation networks, rail transit systems play a vital role in the rapid transportation of intercity passengers. Intercity rail transit vehicle maintenance management refers to the entire process of ensuring vehicle operational safety and availability through the development of maintenance plans, implementation of condition monitoring, and organization of maintenance operations. In recent years, data-driven maintenance management methods have made significant progress in fault early warning and maintenance strategy optimization by collecting vehicle operation and environmental data and utilizing data mining and pattern recognition technologies. This has effectively improved vehicle maintenance efficiency and reduced operation and maintenance costs.

[0003] Existing data analysis methods primarily rely on real-time monitoring data collected during intercity vehicle operation, using preset thresholds and rule bases to identify anomalies and trigger fault alarms. Under normal operating conditions, these methods can meet basic vehicle maintenance needs. However, in the unique environment of hot and humid coastal areas, intercity vehicle components may exhibit normal data due to equipment heating during operation. Meanwhile, the causes of faults, such as salt spray deposition, occur during operation; fault evolution, such as condensation formation, occurs during rest periods; and fault manifestations, such as insulation breakdown, occur the following day upon power restoration. This results in a complete separation of fault causes, evolution processes, and fault manifestations on the timeline, creating data acquisition blind spots and causal misalignment issues.

[0004] The shortcomings of existing technologies are that they lack the ability to reverse the state evolution during rest periods and cannot establish a correlation model between cumulative damage and triggering conditions. This makes it difficult to identify hidden fault risks that span the operating and rest periods, resulting in a significant decrease in the accuracy of vehicle maintenance early warning in special environments. Summary of the Invention

[0005] To address the problem that existing data analysis methods for rail transit vehicle maintenance management lack the ability to invert the state evolution during rest periods and cannot establish a correlation model between cumulative damage and triggering conditions, making it difficult to identify hidden fault risks that span operating and rest periods, resulting in a significant decrease in the accuracy of vehicle maintenance early warning under special environments, this invention provides a data analysis-based method and system for rail transit vehicle maintenance management.

[0006] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows: A data analysis-based method for the maintenance management of rail transit vehicles includes the following steps: S1. Collect vehicle operation data, vehicle operating environment data, component historical status data and virtual sensor data in hot and humid coastal areas, and calculate the estimated value of misalignment impact; S2. Extract virtual sensor data of vehicle components corresponding to the estimated value of misalignment impact, identify periodic risk periods and determine active risk periods; S3. Construct a decision binary tree using the estimated impact of misalignment and the period of proactive risk as input features, and mark the crossover points. S4. Divide the lifespan range of vehicle components based on the crossover point, optimize and output the recommended maintenance cycle for the current operating environment using a weighted index of maintenance cost and equipment availability.

[0007] Furthermore, the detailed steps of step S1 include: Using abnormal operation data in vehicle operation data and abnormal environment data in vehicle operation environment data as initial analysis data pairs, record the abnormal operation time and abnormal environment time in each initial analysis data pair; The initial analysis data pairs corresponding to the same abnormal running time and abnormal environment time are marked as instantaneous correlation data pairs; For each instantaneous correlated data pair, extract the sequence of component state changes within a preset time window before and after the corresponding instantaneous correlated data pair in the component's historical state data, extract the sequence of simulated state changes within a preset time window before and after the corresponding instantaneous correlated data pair in the virtual sensor data, and align the component state change sequence with the simulated state change sequence in time.

[0008] Furthermore, the sequence of component state changes is time-aligned with the sequence of simulated state changes, including: The absolute value of the difference between the sequence of component state changes and the sequence of simulated state changes is compared at each time step. If the absolute value of the difference is less than the preset consistency threshold, the corresponding time step is marked as a consistent time step. If the absolute value of the difference is greater than or equal to the preset consistency threshold, the corresponding time step is marked as an inconsistent time step and the time difference value is recorded. The number of inconsistent moments within a preset time window and the time difference value corresponding to each inconsistent moment are counted. The time difference values ​​are accumulated to obtain the total inconsistency deviation. The ratio of the number to the total number of moments in the preset time window is used as the proportion of inconsistent moments. The initial analysis data pairs corresponding to different abnormal running times and abnormal environment times are marked as misaligned candidate data pairs.

[0009] Furthermore, the step of marking the initial analysis data pairs corresponding to different abnormal running times and abnormal environment times as misaligned candidate data pairs includes: For each misaligned candidate data pair, the time difference between the abnormal operation time and the abnormal environment time is extracted. The total inconsistency deviation and the proportion of inconsistency time of the instantaneous associated data pair with the same fault type as the misaligned candidate data pair are retrieved. The product of the total inconsistency deviation and the proportion of inconsistency time is used as the influence intensity benchmark. The influence intensity benchmark is multiplied by the time difference to obtain the estimated value of the misalignment influence.

[0010] Furthermore, the detailed steps of step S2 include: The virtual sensor data is divided according to the vehicle operation cycle, and each vehicle operation cycle includes a complete running period and a complete rest period. Within each vehicle operation cycle, using the component state simulation curves in the virtual sensor data as input, the system identifies periods in which the component state simulation curves exhibit monotonous changes during rest periods and the magnitude of these changes exceeds a preset change threshold, and marks these periods as cycle risk periods.

[0011] Furthermore, the periodic risk period marked includes: Calculate the length of the periodic risk period for N consecutive vehicle operation cycles, and then calculate the difference in the length of the periodic risk period between two adjacent vehicle operation cycles. If N consecutive differences are all positive, then the length of the periodic risk assessment period is continuously increasing. If N consecutive differences are all negative, the length of the risk period is continuously shortened. The periodic risk period for the next vehicle operation cycle, which is determined to be continuously increasing or continuously decreasing, and the subsequent vehicle operation cycles, will be marked as the active risk period. By using proactive risk periods as input for maintenance management, inspection plans are developed for vehicle components corresponding to the estimated impact of misalignment.

[0012] Furthermore, the formulation of an inspection plan for the vehicle components corresponding to the estimated misalignment impact includes: A special testing plan is developed based on the vehicle components and risk characteristics involved in the active risk period. The special testing plan includes measuring the degree of contamination in the standard area or reinforced area, the depth of microcracks in the load-bearing components, and the insulation performance value of the protective items.

[0013] Furthermore, the detailed steps of step S3 include: The estimated value of the misalignment effect is used to measure the degree of separation between the time when the cause of the failure occurs and the time when the failure manifests on the time axis. Misalignment means that the root cause of the failure and the final manifestation of the failure occur at different times, and there is a time interval between the two. The active risk period is used to measure the persistence and periodicity of risk evolution within the blind zone. A continuous increase or decrease in the length of the periodic risk period indicates a continuous increase in the risk level within the blind zone. The blind zone refers to a time period or physical dimension in which the fault evolution process occurs outside the monitoring range of the data acquisition system, causing the data collected during operation to appear normal while actual hidden dangers have accumulated. The estimated value of the misalignment impact of each vehicle component calculated in each vehicle operation cycle and the active risk period are used as sample point features. The actual failure time of the vehicle component or the actual detection life end of the vehicle component is used as the target variable. A decision binary tree is constructed using a recursive segmentation algorithm. Each internal node of the decision binary tree corresponds to a splitting condition of a sample point feature, and each leaf node corresponds to a group of sample points with similar lifetime features. Extract the split threshold of all internal nodes from the decision binary tree, and mark the nodes on the branch path that involve the characteristics of both the misalignment impact estimate and the active risk period as crossover points.

[0014] Furthermore, the detailed steps of step S4 include: The crossover point represents the critical condition at which the combined effects of misalignment and active risk periods cause a significant change in the lifespan of vehicle components. Based on the feature thresholds corresponding to the crossover points, the sample points under different operating environments are divided into K life intervals. The division is based on the actual fault time or the actual detection life end point corresponding to the sample point. The average remaining life of the vehicle components is calculated within each life interval. Based on the average remaining lifespan, M candidate maintenance cycles are generated according to a preset ratio set. Using the weighted comprehensive index of maintenance cost and equipment availability as the optimization objective, the recommended maintenance cycle that optimizes the weighted comprehensive index is selected from the M candidate maintenance cycles as the recommended maintenance cycle under the current operating environment.

[0015] A data analysis-based rail transit vehicle maintenance management system, including The data acquisition and misalignment analysis module is used to collect vehicle operation data, vehicle operating environment data, component historical status data and virtual sensor data in hot and humid coastal areas, and calculate the estimated value of misalignment impact; The proactive risk identification module is used to extract virtual sensor data of vehicle components corresponding to the estimated value of misalignment impact, identify periodic risk periods, and determine proactive risk periods. The decision tree construction module is used to construct a decision binary tree using the estimated value of the misalignment impact and the active risk period as input features, and to mark the crossover points. The maintenance cycle optimization module is used to divide the lifespan range of vehicle components based on the crossover point, optimize and output the recommended maintenance cycle for the current operating environment using a weighted index of maintenance cost and equipment availability.

[0016] Compared with the prior art, the present invention has the following advantages: By collecting vehicle operation data, environmental data, historical component status data, and virtual sensor data from hot and humid coastal areas, this study integrates and constructs estimates of misalignment impacts and proactive risk periods, quantifying the temporal separation between fault causes and manifestations, as well as the evolution trend of blind zone risks. A recursive segmentation algorithm is used to construct a decision binary tree, extracting the crossover points between misalignment impacts and proactive risk periods, enabling accurate identification of hidden fault risks spanning operating and idle periods. Based on these crossover points, lifespan intervals are divided, and a weighted comprehensive index of maintenance cost and equipment availability is used to optimize recommended maintenance cycles. This overcomes the shortcomings of traditional methods that rely on operating period data and ignore idle period evolution, significantly improving the accuracy of vehicle maintenance early warning and the ability to dynamically optimize cycles under special environments. Attached Figure Description

[0017] Figure 1 This is a flowchart of a data analysis-based rail transit vehicle maintenance management method in an embodiment of the present invention; Figure 2 This is a structural block diagram of a rail transit vehicle maintenance management system based on data analysis, as described in an embodiment of the present invention. Detailed Implementation

[0018] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0019] like Figure 1 As shown, this embodiment provides a data analysis-based method for the maintenance and management of rail transit vehicles, such as... Figure 1 As shown, it includes the following steps: S1. Collect vehicle operation data, vehicle operating environment data, component historical status data, and virtual sensor data in hot and humid coastal areas, and calculate the estimated value of misalignment impact, including: Using abnormal operation data in vehicle operation data and abnormal environment data in vehicle operation environment data as initial analysis data pairs, record the abnormal operation time and abnormal environment time in each initial analysis data pair; The initial analysis data pairs corresponding to the same abnormal running time and abnormal environment time are marked as instantaneous correlation data pairs; For each instantaneous correlated data pair, extract the sequence of component state changes within a preset time window before and after the corresponding instantaneous correlated data pair in the component historical state data, extract the sequence of simulated state changes within a preset time window before and after the corresponding instantaneous correlated data pair in the virtual sensor data, and align the component state change sequence with the simulated state change sequence in time. The absolute value of the difference between the sequence of component state changes and the sequence of simulated state changes is compared at each time step. If the absolute value of the difference is less than the preset consistency threshold, the corresponding time step is marked as a consistent time step. If the absolute value of the difference is greater than or equal to the preset consistency threshold, the corresponding time step is marked as an inconsistent time step and the time difference value is recorded. The number of inconsistent moments within a preset time window and the time difference value corresponding to each inconsistent moment are counted. The time difference values ​​are accumulated to obtain the total inconsistency deviation. The ratio of the number to the total number of moments in the preset time window is used as the proportion of inconsistent moments. The initial analysis data pairs corresponding to different abnormal running times and abnormal environment times are marked as misaligned candidate data pairs; For each misaligned candidate data pair, the time difference between the abnormal operation time and the abnormal environment time is extracted. The total inconsistency deviation and the proportion of inconsistency time of the instantaneous associated data pair with the same fault type as the misaligned candidate data pair are retrieved. The product of the total inconsistency deviation and the proportion of inconsistency time is used as the influence intensity benchmark. The influence intensity benchmark is multiplied by the time difference to obtain the estimated value of the misalignment influence.

[0020] It should be noted that intercity rail transit vehicles in hot and humid coastal areas operate for extended periods in environments with high salt spray, high humidity, and significant diurnal temperature variations. Salt spray particles adhere to the surfaces of roof insulators and undercarriage junction box components via airflow. During the day, when temperatures rise or equipment heats up, the surfaces remain dry, and insulation performance and electrical parameters appear normal. However, after trains return to the depot and stop operating at night, the equipment cools below ambient temperature, causing condensation on the surfaces that dissolves salt and forms conductive channels. Insulation performance drops sharply during periods without monitoring, leading to flashover tripping when power is restored the following morning. Existing methods only collect real-time monitoring data during operation, completely missing the insulation evolution process during periods of inactivity, making it difficult to identify such hidden fault risks that span both operational and inactive periods.

[0021] To address the aforementioned deficiencies, this step collects four types of data to construct a complete data foundation. Vehicle operation data, obtained through onboard sensors, tracks component temperature, current, voltage, and status changes, reflecting actual operating conditions during operation. Vehicle operating environment data, including ambient temperature, relative humidity, and salt spray concentration, is obtained from meteorological stations along the route or from grid-based forecasts, characterizing external stress input. Component historical status data, derived from insulation resistance measurements, salt density tests, and cleaning times in maintenance records, provides a measured basis for the degree of cumulative contamination. Virtual sensor data, based on thermodynamic principles, constructs component surface temperature change curves, and based on the principle of component surface thermal balance, predicts the condensation initiation time, filling monitoring gaps during periods of inactivity.

[0022] Abnormal operational data and abnormal environmental data are used as initial analysis data pairs. Abnormal operational data refers to data points where the state change exceeds the normal operating fluctuation range, and abnormal environmental data refers to environmental parameter points where the salt spray concentration, relative humidity, or temperature change rate exceeds the climatological statistical threshold. The abnormal operational time and abnormal environmental time are recorded for each initial analysis data pair. Data pairs with the same occurrence time are marked as instantaneously correlated data pairs, and data pairs with different occurrence times are marked as misalignment candidate data pairs. Instantaneously correlated data pairs represent fault types where the cause and manifestation occur simultaneously, used to establish the baseline impact intensity. Misalignment candidate data pairs represent fault types where the cause and manifestation are temporally separated, and are the core objects of subsequent misalignment analysis.

[0023] For each instantaneous correlated data pair, the sequence of component state changes and the sequence of simulated state changes within a preset time window before and after that moment are extracted. The preset time window is set according to the characteristics of the fault type. For faults such as insulation failure, where the cause and manifestation occur simultaneously, the time window covers the entire period before and after the fault to capture instantaneous change characteristics. For cumulative faults such as stress corrosion cracking, the time window needs to cover a longer accumulation process, usually set to a time range of several weeks to several months before the fault occurs. The specific length is determined comprehensively based on material properties, stress levels, and historical maintenance cycles. The sequence of component state changes comes from actual measurement data, and the sequence of simulated state changes comes from virtual sensor extrapolation data. The two sequences represent the actual state evolution and the theoretical expected evolution, respectively. After aligning the two sequences in time, the absolute value of the difference is compared moment by moment. When the absolute value of the difference is less than a preset consistency threshold, it is marked as a consistent moment; when the absolute value of the difference is greater than or equal to the preset consistency threshold, it is marked as an inconsistent moment, and the difference value at that moment is recorded. The preset consistency threshold is determined based on a combination of the accuracy of the measuring instrument and the calculation error of the physical model upon which the virtual sensor is based. The accuracy of the measuring instrument is provided by the sensor manufacturer, and the model calculation error is obtained by statistically analyzing the deviations between historical simulated data and measured data from the same period. For example, the root mean square error of multiple comparisons can be used as the error estimate. Consistency indicates that the actual state matches the theoretical expectation, while inconsistency indicates that the actual state deviates from the theoretical expectation. The greater the deviation, the more severe the cumulative damage or contamination.

[0024] The number of inconsistent moments within a preset time window and the corresponding time difference value for each inconsistent moment are counted. These time difference values ​​are then summed to obtain the total inconsistency deviation. The total inconsistency deviation is the cumulative deviation magnitude of all deviation moments within the time window; a larger value indicates a more severe overall deviation between the actual state and the expected state within that time window. The ratio of the number of inconsistent moments to the total number of moments in the preset time window is used as the inconsistency moment percentage, representing the distribution density of deviation moments within the time window. A larger percentage indicates more frequent or more persistent deviations.

[0025] For each misaligned candidate data pair, the time difference between the abnormal operating time and the abnormal environment time is extracted. This time difference represents the interval between the occurrence of the fault cause and the occurrence of the fault manifestation; a larger time difference indicates a more significant misalignment. The total inconsistency deviation and the proportion of inconsistency times of the instantaneously correlated data pairs with the same fault type as the misaligned candidate data pairs are retrieved, and their product is used as the impact strength benchmark. The impact strength benchmark is a comprehensive measure of the cumulative magnitude and distribution density of the deviation of the actual state from the expected state within a unit time window under conditions without time misalignment; a larger value indicates a greater impact strength for the same type of fault under instantaneous correlation conditions. The impact strength benchmark is multiplied by the time difference to obtain the estimated misalignment impact value. This estimate quantifies the contribution of the cumulative cause to the fault manifestation after a time delay, even with time misalignment. A larger estimated misalignment impact value indicates a higher fault risk corresponding to the misaligned candidate data pair, requiring priority inclusion in subsequent proactive risk identification processes.

[0026] This step establishes an impact intensity model based on the instantaneous correlated data pairs and misaligned candidate data pairs by separating the instantaneous correlated data pairs. The time difference is multiplied by the impact intensity benchmark to quantify the degree of misalignment impact. For the first time, a quantitative assessment of the hidden fault risk spanning the operation period and the rest period has been achieved, providing a reliable input for subsequent proactive risk identification and maintenance cycle optimization.

[0027] S2: Extract virtual sensor data of vehicle components corresponding to the estimated impact of misalignment, identify periodic risk periods, and determine active risk periods, including: The virtual sensor data is divided according to the vehicle operation cycle, and each vehicle operation cycle includes a complete running period and a complete rest period. Within each vehicle operation cycle, using the component state simulation curves in the virtual sensor data as input, the system identifies periods in which the component state simulation curves exhibit monotonous changes during rest periods and the magnitude of these changes exceeds a preset change threshold, and marks these periods as cycle risk periods. Calculate the length of the periodic risk period for N consecutive vehicle operation cycles, and then calculate the difference in the length of the periodic risk period between two adjacent vehicle operation cycles. If N consecutive differences are all positive, then the length of the periodic risk assessment period is continuously increasing. If N consecutive differences are all negative, the length of the risk period is continuously shortened. The periodic risk period for the next vehicle operation cycle, which is determined to be continuously increasing or continuously decreasing, and the subsequent vehicle operation cycles, will be marked as the active risk period. Using proactive risk periods as input for maintenance management, we develop inspection plans for vehicle components corresponding to the estimated impact of misalignment. A special testing plan is developed based on the vehicle components and risk characteristics involved in the active risk period. The special testing plan includes measuring the degree of contamination in the standard area or reinforced area, the depth of microcracks in the load-bearing components, and the insulation performance value of the protective items.

[0028] It should be noted that after obtaining the misalignment impact estimate in step S1, this step focuses on quantifying the risk evolution trend within the blind zone. The misalignment impact estimate only reveals the degree of separation between the cause of the failure and its temporal manifestation, but it cannot reflect how the risk dynamically accumulates during the rest period. Therefore, virtual sensor data of the vehicle components corresponding to the misalignment impact estimate is extracted. This type of data is used to calculate the surface temperature and humidity change curves during the rest period through the component thermal balance principle. The virtual sensor data is divided according to the vehicle operation cycle. Each vehicle operation cycle includes one complete daytime operation segment and one nighttime rest period. The vehicle operation cycle starts at the time of the train's first departure from the depot each day and ends at the time before the first departure from the depot the following day. Each cycle includes one complete daytime operation segment and one nighttime rest period. The operation segment refers to the actual passenger-carrying operation time from the train's departure from the depot to its return to the depot, and the rest period refers to the time of the train's stoppage from its return to the depot to its first departure from the depot the following day. For trains operating across days, one complete departure from the depot to its return to the depot is still considered as one operation cycle, and it is not divided due to crossing midnight. The operating period refers to the time during which a train carries passengers, while the rest period refers to the time from when the train returns to the depot and stops operating until it leaves the depot the next day.

[0029] Within each vehicle operating cycle, a component condition simulation curve, generated from virtual sensor data, is used as input to reflect the continuous changes in component surface temperature, humidity, or insulation performance during rest periods. Periods where the curve exhibits a monotonically increasing or decreasing trend during rest periods, with the magnitude of change exceeding a preset threshold, are marked as periodic risk periods. The preset threshold is determined based on component material properties and environmental stress limits, such as the minimum temperature difference required for the insulator surface temperature to drop to the dew point temperature. Periodic risk periods represent potential condensation or corrosion risk windows within those rest periods.

[0030] The length of the cyclical risk period is statistically analyzed for N consecutive vehicle operation cycles, and the difference between the lengths of the cyclical risk period between two adjacent operation cycles is calculated. A continuously positive difference indicates that the cyclical risk period is lengthening cycle by cycle, while a continuously negative difference indicates that the cyclical risk period is shortening cycle by cycle. A lengthening period implies accelerated accumulation of surface contamination on components or deterioration of environmental conditions, while a shortening period indicates the emergence of cleaning or self-healing effects. The cyclical risk period of the next vehicle operation cycle, and all subsequent vehicle operation cycles, determined to be continuously lengthening, is marked as a high-risk proactive risk period, requiring focused inspection measures. If the period is determined to be continuously shortening, the subsequent cyclical risk periods are marked as observational proactive risk periods, used to record and verify the effectiveness of risk mitigation, but not as a basis for mandatory inspection.

[0031] Active risk periods are used as input for maintenance management and incorporated into the maintenance plan development stage. Standard areas refer to untreated or unreinforced areas on component surfaces, typically the baseline measurement locations marked on design drawings. Reinforced areas refer to areas with special protection or reinforcement, requiring separate measurement to assess protection effectiveness. Specific inspection plans are developed based on the vehicle component types and risk characteristics involved in the active risk periods. These plans differ from routine inspections and are scheduled within appropriate time windows based on the severity of the active risk period and component type. They allow the use of dedicated testing equipment or power outage and disconnection operations to obtain accurate measurement data. For example, the contamination level is measured in the standard area of ​​roof insulators, the microcrack depth is measured in bogie frame load-bearing components, and the insulation performance value is measured for high-voltage cable joint protection items. These measurement data are used to verify the accuracy of the active risk periods and provide sample points for subsequent decision binary tree construction. Verifying the accuracy of proactive risk periods is achieved by statistically analyzing the significant differences between the measured contamination levels, microcrack depths, or insulation performance values ​​corresponding to high-risk proactive risk periods and their corresponding values ​​during normal periods. For example, if the average measured value during high-risk proactive risk periods is significantly higher than that during non-risk periods, it indicates that the proactive risk period identification is accurate. This step, by quantifying the risk evolution trend within rest periods, achieves for the first time the proactive identification of hidden fault risks spanning operating and rest periods, overcoming the shortcomings of existing methods that rely solely on operating period data. The parameter N is used to determine the number of consecutive periods for trend assessment. It is determined through optimization based on historical fault data, for example, selecting the value that maximizes the accuracy of trend assessment within the range of N=3 to N=10; or setting it to a fixed value, such as 5 or 7, based on business experience.

[0032] S3: Construct a decision binary tree using the estimated impact of misalignment and the active risk period as input features, and mark the crossover points, including: The estimated value of the misalignment effect is used to measure the degree of separation between the time when the cause of the failure occurs and the time when the failure manifests on the time axis. Misalignment means that the root cause of the failure and the final manifestation of the failure occur at different times, and there is a time interval between the two. The active risk period is used to measure the persistence and periodicity of risk evolution within the blind zone. A continuous increase or decrease in the length of the periodic risk period indicates a continuous increase in the risk level within the blind zone. The blind zone refers to a time period or physical dimension in which the fault evolution process occurs outside the monitoring range of the data acquisition system, causing the data collected during operation to appear normal while actual hidden dangers have accumulated. The estimated value of the misalignment impact of each vehicle component calculated in each vehicle operation cycle and the active risk period are used as sample point features. The actual failure time of the vehicle component or the actual detection life end of the vehicle component is used as the target variable. A decision binary tree is constructed using a recursive segmentation algorithm. Each internal node of the decision binary tree corresponds to a splitting condition of a sample point feature, and each leaf node corresponds to a group of sample points with similar lifetime features. Extract the split threshold of all internal nodes from the decision binary tree, and mark the nodes on the branch path that involve the characteristics of both the misalignment impact estimate and the active risk period as crossover points. S4. Based on the crossover point, divide the vehicle component lifespan range, optimize and output the recommended maintenance cycle for the current operating environment using a weighted index of maintenance cost and equipment availability. The steps include: The crossover point represents the critical condition at which the combined effects of misalignment and active risk periods cause a significant change in the lifespan of vehicle components. Based on the feature thresholds corresponding to the crossover points, the sample points under different operating environments are divided into K life intervals. The division is based on the actual fault time or the actual detection life end point corresponding to the sample point. The average remaining life of the vehicle components is calculated within each life interval. Based on the average remaining lifespan, M candidate maintenance cycles are generated according to a preset ratio set. Using the weighted comprehensive index of maintenance cost and equipment availability as the optimization objective, the recommended maintenance cycle that optimizes the weighted comprehensive index is selected from the M candidate maintenance cycles as the recommended maintenance cycle under the current operating environment.

[0033] It should be noted that the misalignment impact estimate quantifies the temporal separation between the cause and manifestation of a fault. For example, a misalignment impact estimate of 0.8 for an insulator indicates a significant time interval between salt spray deposition and insulation breakdown. The active risk period reflects the trend of risk evolution within the blind zone. For instance, if the length of the periodic risk period increases from 2 hours to 3 hours over three consecutive operating cycles, it is considered a continuous increase, indicating a rising risk level in the future. The purpose of constructing the decision binary tree is to automatically learn the nonlinear relationship between these two features and component lifespan from historical data, identifying the critical conditions that lead to significant changes in lifespan.

[0034] The estimated misalignment impact value and active risk period calculated for each vehicle component in each operating cycle are used as sample point features. The detection lifespan endpoint is determined based on measurement data obtained from the special inspection plan in S2, combined with a preset safety threshold determined according to industry standards or equipment technical specifications. For example, when the insulator contamination level exceeds the safety threshold, that moment is determined as the lifespan endpoint. The actual failure time of the component or the detection lifespan endpoint is used as the target variable. When both the actual failure time and the detection lifespan endpoint exist for a sample point, the actual failure time is used as the target variable; if the component does not experience an actual failure during the observation period, the detection lifespan endpoint is used as the target variable. A recursive segmentation algorithm is used to construct a decision tree. It should be noted that during the decision tree construction process, the splitting condition of each internal node is determined by a critical value on the corresponding feature. This critical value is referred to as the feature threshold or splitting threshold in this invention; both have the same meaning, referring to the feature value boundary point used to segment the samples. The algorithm steps include: traversing all features and their values, selecting the feature and threshold that maximizes the decrease in the variance of the partitioned subset's lifespan as the internal node splitting condition; recursively proceeding until the stopping criterion is met. Each internal node corresponds to a splitting condition for a feature, and each leaf node contains a set of sample points with similar lifetime features. Similarity is quantified by the variance or standard deviation of lifetime values ​​within the subset. The splitting threshold is determined by calculating the midpoint of adjacent values ​​for each feature and selecting the value that minimizes the impurity of the resulting subset.

[0035] Nodes on branch paths involving both the estimated impact of misalignment and the active risk period are marked as crossover points. These points represent the critical conditions under which the component's lifespan changes significantly due to the combined effect of these two features. Since crossover points reveal the critical conditions that lead to significant changes in lifespan due to the combined effect of these two features, using their corresponding feature thresholds as the basis for segmentation allows for more accurate differentiation of sample groups with different risk levels, providing a reasonable foundation for recommending differentiated maintenance cycles. For example, if a node's splitting condition is that the estimated impact of misalignment is greater than 0.6 and the active risk period length is greater than 3 hours, then this node is a crossover point, meaning that when both conditions are met simultaneously, the average component lifespan drops sharply from 10 years to 5 years. Based on the feature thresholds corresponding to the crossover points, sample points under different operating environments are divided into K lifespan intervals. The average remaining lifespan of components within each lifespan interval is statistically analyzed to generate candidate maintenance cycles and optimize their selection.

[0036] The optimization objective is a weighted composite index of maintenance cost and equipment availability. Maintenance cost includes spare parts, labor, and downtime losses, while equipment availability refers to the proportion of vehicle uptime. Weighting coefficients are set using expert scoring or the analytic hierarchy process (AHP) to reflect managerial preferences. Based on the average remaining lifespan within each interval, M candidate maintenance cycles are generated according to a preset set of proportions. The recommended maintenance cycle that optimizes the weighted composite index is selected from these candidate cycles as the recommended cycle for the current operating environment. For example, if a component in a specific environment generates M candidate cycles (e.g., M=3) with weighted indices of 0.8, 0.7, and 0.9 respectively, the cycle corresponding to 0.7 is selected.

[0037] This step automatically identifies key thresholds for hidden risks across timeframes using decision trees and dynamically optimizes maintenance cycles, overcoming the limitation of existing methods that cannot accurately formulate maintenance strategies for specific environments. Parameter K represents the number of lifespan intervals, determined using cross-validation. The optimal K value is selected based on the recommended maintenance cycle accuracy on the validation set; alternatively, it can be set to 3 or 4 based on experience with component lifespan stages. Parameter M represents the number of candidate maintenance cycles, and its value is related to the method of determining the preset proportion set. A preferred method is to generate a set of proportion coefficients based on the average remaining lifespan, following an arithmetic or geometric sequence. For example, the proportion set could be set to {1 / 4, 1 / 3, 1 / 2} corresponding to M=3, or {0.2, 0.3, 0.4, 0.5} corresponding to M=4. The selection of the proportion set is determined based on the refinement requirements of the maintenance strategy; a larger M value results in a denser candidate cycle pool, but also increases the computational load. Alternatively, it can be set based on business experience, such as determining the M value based on the number of maintenance levels recommended by the equipment manufacturer, typically between 3 and 5.

[0038] It should be noted that the method proposed in this invention is further integrated with Reliability-Centered Maintenance (RCM) theory. In traditional RCM analysis, Failure Mode and Effects Analysis (FMEA) is usually based on the direct correlation between observable failure phenomena and causes. However, for hidden failures with misaligned causes and conditions and spanning data blind spots in hot and humid coastal environments, traditional RCM struggles to accurately assess their risk level and maintenance requirements. This invention quantifies the temporal separation degree and blind spot evolution trend of such failures by constructing estimated values ​​of misaligned impacts and active risk periods, providing crucial quantitative input for RCM analysis. Based on this, maintenance decision-makers can formulate more precise maintenance tasks, such as condition monitoring, periodic replacement, and fault location, based on the failure consequence classification of RCM, such as safety, operational, and economic consequences, combined with the lifespan intervals and recommended maintenance cycles defined by the crossover points output by this invention. This allows for the optimization of maintenance resource allocation while ensuring reliability.

[0039] A data analysis-based rail transit vehicle maintenance management system, such as Figure 2As shown, it includes: The data acquisition and misalignment analysis module is used to collect vehicle operation data, vehicle operating environment data, component historical status data and virtual sensor data in hot and humid coastal areas, and calculate the estimated value of misalignment impact; The proactive risk identification module is used to extract virtual sensor data of vehicle components corresponding to the estimated value of misalignment impact, identify periodic risk periods, and determine proactive risk periods. The decision tree construction module is used to construct a decision binary tree using the estimated value of the misalignment impact and the active risk period as input features, and to mark the crossover points. The maintenance cycle optimization module is used to divide the lifespan range of vehicle components based on the crossover point, optimize and output the recommended maintenance cycle for the current operating environment using a weighted index of maintenance cost and equipment availability.

[0040] Compared with the prior art, the present invention has the following advantages: By collecting vehicle operation data, environmental data, historical component status data, and virtual sensor data from hot and humid coastal areas, this study integrates and constructs estimates of misalignment impacts and proactive risk periods, quantifying the temporal separation between fault causes and manifestations, as well as the evolution trend of blind zone risks. A recursive segmentation algorithm is used to construct a decision binary tree, extracting the crossover points between misalignment impacts and proactive risk periods, enabling accurate identification of hidden fault risks spanning operating and idle periods. Based on these crossover points, lifespan intervals are divided, and a weighted comprehensive index of maintenance cost and equipment availability is used to optimize recommended maintenance cycles. This overcomes the shortcomings of traditional methods that rely on operating period data and ignore idle period evolution, significantly improving the accuracy of vehicle maintenance early warning and the ability to dynamically optimize cycles under special environments.

[0041] The foregoing has provided a detailed description of a data analysis-based method and system for the maintenance and management of rail transit vehicles provided in this application. The specific embodiments described are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A data analysis-based rail transit vehicle inspection management method, characterized in that, Including the following steps: S1. Collect vehicle operation data, vehicle operating environment data, component historical status data and virtual sensor data in the hot and humid coastal areas, and calculate the estimated value of misalignment impact; S2. Extract virtual sensor data of vehicle components corresponding to the estimated value of misalignment impact, identify periodic risk periods and determine active risk periods; S3. Construct a decision binary tree using the estimated impact of misalignment and the period of active risk as input features, and mark the crossover points. S4. Divide the lifespan range of vehicle components based on the crossover point, optimize and output the recommended maintenance cycle for the current operating environment using a weighted index of maintenance cost and equipment availability.

2. The data analysis-based rail transit vehicle inspection management method according to claim 1, wherein, The detailed steps of step S1 include: Using abnormal operation data in vehicle operation data and abnormal environment data in vehicle operation environment data as initial analysis data pairs, record the abnormal operation time and abnormal environment time in each initial analysis data pair; The initial analysis data pairs corresponding to the same abnormal running time and abnormal environment time are marked as instantaneous correlation data pairs; For each instantaneous correlated data pair, extract the sequence of component state changes within a preset time window before and after the corresponding instantaneous correlated data pair in the component's historical state data, extract the sequence of simulated state changes within a preset time window before and after the corresponding instantaneous correlated data pair in the virtual sensor data, and align the component state change sequence with the simulated state change sequence in time.

3. The data analysis-based rail transit vehicle inspection management method according to claim 2, characterized in that, The steps for time-aligning the component state change sequence with the simulated state change sequence include: The absolute value of the difference between the sequence of component state changes and the sequence of simulated state changes is compared at each time step. If the absolute value of the difference is less than the preset consistency threshold, the corresponding time step is marked as a consistent time step. If the absolute value of the difference is greater than or equal to the preset consistency threshold, the corresponding time step is marked as an inconsistent time step and the time difference value is recorded. The number of inconsistent moments within a preset time window and the time difference value corresponding to each inconsistent moment are counted. The time difference values ​​are accumulated to obtain the total inconsistency deviation. The ratio of the number to the total number of moments in the preset time window is used as the proportion of inconsistent moments. The initial analysis data pairs corresponding to different abnormal running times and abnormal environment times are marked as misaligned candidate data pairs.

4. The data analysis-based rail transit vehicle inspection management method according to claim 3, characterized in that, The steps for marking the initial analysis data pairs corresponding to different abnormal running times and abnormal environment times as misaligned candidate data pairs include: For each misaligned candidate data pair, extract the time difference between the abnormal running time and the abnormal environment time; The total inconsistency deviation and the proportion of inconsistency times for instantaneous correlated data pairs with the same fault type as the misaligned candidate data are retrieved; The product of the total inconsistency deviation and the proportion of inconsistency time is used as the influence intensity benchmark. The influence intensity benchmark is multiplied by the time difference to obtain the estimated value of the misalignment influence.

5. The data analysis-based rail transit vehicle inspection management method according to claim 4, characterized in that, The detailed steps of step S2 include: The virtual sensor data is divided according to the vehicle operation cycle, and each vehicle operation cycle includes a complete running period and a complete rest period. Within each vehicle operation cycle, using the component state simulation curves in the virtual sensor data as input, the system identifies periods in which the component state simulation curves exhibit monotonous changes during rest periods and the magnitude of these changes exceeds a preset change threshold, and marks these periods as cycle risk periods.

6. The data analysis-based rail transit vehicle inspection management method according to claim 5, wherein, The steps for marking a period as a cyclical risk period include: Calculate the length of the periodic risk period for N consecutive vehicle operation cycles, and then calculate the difference in the length of the periodic risk period between two adjacent vehicle operation cycles. If N consecutive differences are all positive, then the length of the periodic risk assessment period is continuously increasing. If N consecutive differences are all negative, the length of the risk period is continuously shortened. The periodic risk period for the next vehicle operation cycle, which is determined to be continuously increasing or continuously decreasing, and the subsequent vehicle operation cycles, will be marked as the active risk period. By using proactive risk periods as input for maintenance management, inspection plans are developed for vehicle components corresponding to the estimated impact of misalignment.

7. The data analysis-based rail transit vehicle inspection management method according to claim 6, characterized in that, The inspection plan for the vehicle components corresponding to the estimated impact of misalignment is as follows: A special testing plan is developed based on the vehicle components and risk characteristics involved in the active risk period. The special testing plan includes measuring the degree of contamination in the standard area or reinforced area, the depth of microcracks in the load-bearing components, and the insulation performance value of the protective items.

8. The data analysis-based rail transit vehicle inspection management method according to claim 7, characterized in that, The detailed steps of step S3 include: The misalignment effect estimate is used to measure the degree of separation on the time axis between the time when the cause of the failure occurs and the time when the failure manifests. The active risk period is used to measure the persistence and periodicity of risk evolution within the blind zone. A continuous increase or decrease in the length of the periodic risk period indicates a continuous increase in the risk level within the blind zone. The estimated value of the misalignment impact of each vehicle component calculated in each vehicle operation cycle and the active risk period are used as sample point features. The actual failure time of the vehicle component or the actual detection life end of the vehicle component is used as the target variable. A decision binary tree is constructed using a recursive segmentation algorithm. Each internal node of the decision binary tree corresponds to a splitting condition of a sample point feature, and each leaf node corresponds to a group of sample points with similar lifetime features. Extract the split threshold of all internal nodes from the decision binary tree, and mark the nodes on the branch path that involve the characteristics of both the misalignment impact estimate and the active risk period as crossover points.

9. The data analysis-based rail transit vehicle inspection management method according to claim 8, wherein, The detailed steps of step S4 include: The crossover point represents the critical condition at which the combined effects of misalignment and active risk periods cause a significant change in the lifespan of vehicle components. Based on the feature thresholds corresponding to the crossover points, the sample points under different operating environments are divided into K life intervals. The division is based on the actual fault time or the actual detection life end point corresponding to the sample point. The average remaining life of the vehicle components is calculated within each life interval. Based on the average remaining lifespan, M candidate maintenance cycles are generated according to a preset ratio set. Using the weighted comprehensive index of maintenance cost and equipment availability as the optimization objective, the recommended maintenance cycle that optimizes the weighted comprehensive index is selected from the M candidate maintenance cycles as the recommended maintenance cycle under the current operating environment.

10. A rail transit vehicle inspection management system based on data analysis, characterized in that, It includes a data acquisition and misalignment analysis module, a proactive risk identification module, a decision tree construction module, and a maintenance cycle optimization module; The data acquisition and misalignment analysis module is used to collect vehicle operation data, vehicle operating environment data, component historical status data, and virtual sensor data in hot and humid coastal areas, and to calculate the estimated value of misalignment impact. The proactive risk identification module is used to extract virtual sensor data of vehicle components corresponding to the estimated value of misalignment impact, identify periodic risk periods, and determine proactive risk periods. The decision tree construction module is used to construct a decision binary tree using the estimated value of the misalignment impact and the active risk period as input features, and to mark the crossover points. The maintenance cycle optimization module is used to divide the lifespan range of vehicle components based on the crossover point, optimize and output the recommended maintenance cycle for the current operating environment using a weighted index of maintenance cost and equipment availability.