A fault prediction and health management system for rail transit systems
By constructing comparable sample pairs and event masking, and combining causal closure degradation judgment and health baseline freeze-unlocking mechanism, the problems of fault judgment and inaccurate baseline update of non-homogeneous operating data in rail transit system are solved, improving the accuracy of fault prediction and health management and the rationality of maintenance plan.
Patent Information
- Application Number
- CN202611323790.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies in rail transit systems lack mechanisms for strictly comparable constraints on non-homogeneous operational data and for shielding against transient event interference. This leads to inaccurate fault degradation judgments and improper health baseline updates. In particular, when there are changes in train formation, control mode switching, line section changes, or version changes, transient deviations are easily misidentified as real faults or abnormalities, or health baseline updates are improper.
By constructing comparable sample pairs based on topology status identifiers, operational phases, and control command sequences, and combining event masking, causal closure degradation judgment, and a freeze-unfreeze mechanism for healthy baselines, transient disturbances during the transition process are masked, ensuring that fault determination and baseline updates are performed under a consistent operational topology context and control semantics.
It significantly improves the consistency and reliability of the basic data for fault diagnosis, reduces the false alarm rate, improves the accuracy and stability of fault degradation identification, avoids health baseline contamination and improper updates, and enhances the accuracy of predictive maintenance and the foresight of maintenance plans.
Smart Images

Figure CN123104731A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of rail transit operation monitoring and health management technology, and in particular to a rail transit system operation fault prediction and health management system. Background Technology
[0002] With the increasing intelligence, networking, and informatization of rail transit systems, health management technologies for online monitoring, fault early warning, and maintenance decision-making related to the operating status of trains and key equipment have been applied. Among existing technologies, CN111806516A discloses a health management device and method for intelligent train monitoring and maintenance. This solution includes a data acquisition module, a data processing module, an online status detection module, a data communication module, and a data storage module. The data processing module performs filtering, noise reduction, normalization, feature extraction, and cluster analysis on the acquired status data. The online status detection module compares the processed data with detection thresholds in a pre-established model library to determine the equipment's operating status and issue early warnings. Furthermore, this solution discloses data slicing processing of status data according to task requirements, and filtering for recurring faults, intermittent faults, and environmental factors that may cause faults. It also proposes constructing a whole-vehicle-level health management system that conforms to the network and system topology requirements of intelligent trains to achieve predictive maintenance for intelligent operation and maintenance.
[0003] While the aforementioned existing technologies can achieve status data acquisition, data processing, status detection, early warning filtering, and predictive maintenance at the vehicle level, their disclosed content is still mainly based on general processing, slicing, threshold comparison, and early warning filtering of the acquired data. For situations that may occur in the actual operation of rail transit systems, such as changes in train formation, control mode switching, line section changes, software version changes, parameter version changes, and adaptive trial runs after maintenance, the existing technologies do not disclose a specific mechanism for handling the data comparability distortion caused by such operational topology changes. In other words, data obtained under different operational topology states, even if they are continuous in time or similar in surface operating conditions, may not be strictly comparable due to differences in control command sequences, system configuration states, or operational semantics. If such non-homogeneous operational data is directly used for fault degradation judgment or health baseline updates, it is easy to misidentify transient deviations caused by topology changes or transitional operation processes as real fault anomalies, or to lead to improper health baseline updates. This problem has not been specifically addressed in the aforementioned closest existing technologies.
[0004] Therefore, the existing technology has at least the following technical problems: when the operation topology of the rail transit system changes, such as changes in train formation, control mode switching, line section changes or version changes, there is a lack of mechanisms to strictly constrain the comparability of non-homogeneous operation data and shield against interference from transitional events, which can easily lead to inaccurate fault degradation judgment and improper health baseline updates. Summary of the Invention
[0005] To overcome the aforementioned technical deficiencies, the present invention aims to provide a fault prediction and health management system for rail transit systems. This invention solves the inaccuracy problem caused by non-homogeneous operational data directly participating in fault determination and baseline updates under conditions of operational topology changes by constructing comparable sample pairs under the same topology state identifier, the same operational stage, and the same control command sequence, and by combining event masking, causal closure degradation judgment within unmasked intervals, and unfreezing and allowing updates or maintaining the frozen state of the health baseline based on verification results.
[0006] This invention discloses a fault prediction and health management system for rail transit systems, comprising: The data aggregation module is used to collect the operating data, control command data, event log data, maintenance history data, and component genealogy identifiers of the monitored objects. The topology status identifier generation module is used to generate topology status identifiers based on the operating grouping status, control mode, line section, software version, and equipment configuration. The comparable sample pair construction module is used to construct comparable sample pairs from current running data and historical running data under the same topology status identifier, according to the same running stage and the same control command sequence; The event mask generation module is used to generate event masking codes corresponding to comparable sample pairs based on event log data. The health baseline management module is used to store and retrieve the corresponding health baselines according to topology status identifiers and component lineage identifiers; The causal closure degradation judgment module is used to determine the deviation of comparable sample pairs within the unmasked data range determined by the event mask, based on the causal closure relationship between control commands, execution responses and physical state responses, and output fault degradation indicators. The module for determining the expected failure time interval is used to determine the expected failure time interval based on the fault deterioration identifier, maintenance history data, and executable maintenance window data. The health management task generation module is used to generate health management tasks based on fault deterioration indicators and expected failure time intervals. The health management task includes at least the target maintenance action, the target maintenance window, the target verification line section, and the target verification control command sequence. The verification unlock module is used to generate verification results based on the same topology status identifier and the same target verification control command sequence after the health management task is executed, and to unfreeze and allow updates to the health baseline according to the verification results, or keep it frozen.
[0007] Preferably, the topology status identifier generation module is used to generate topology status identifiers based on the operation group status identifier, control mode identifier, line section identifier, software version identifier, parameter version identifier, and equipment configuration verification value.
[0008] Preferably, the comparable sample pair construction module is used to perform time-series encoding on the control command data, generate command sequence verification values, and construct comparable sample pairs only when the topology status identifiers are consistent, the operation phases are consistent, the inter-station operation directions are consistent, and the command sequence verification values are consistent.
[0009] Preferably, it also includes a verification template library. The health management task generation module is used to call the corresponding target verification line segment and target verification control command sequence from the verification template library according to the topology status identifier and command sequence verification value.
[0010] Preferably, the event mask generation module is used to generate event masking masks for group reconnection events, line switching events, control mode switching events, software version switching events, parameter version switching events, and post-maintenance adaptive trial operation events.
[0011] Preferably, the event mask generation module is used to adaptively determine the masking start time and masking end time corresponding to each event based on the response stabilization time from the change of control command after the event occurs until the execution response reaches a stable state.
[0012] Preferably, the health baseline management module is used to establish a dual-index health baseline library based on topology status identifier and component lineage identifier, and each health baseline in the dual-index health baseline library corresponds to a unique baseline version number.
[0013] Preferably, the health baseline management module is used to, when a component replacement event occurs, call the initial inherited baseline from the verified health baselines of the same model and configuration based on the component lineage identifier, equipment configuration verification value and topology status identifier of the replaced target component, and establish an association between the initial inherited baseline and the replaced target component.
[0014] Preferably, the verification unlocking module is used to apply a freeze mark to the health baseline after a component replacement event, software version switch event, or parameter version switch event occurs, and to prohibit updating the baseline version number before the verification of the target verification line segment and the target verification control command sequence is completed.
[0015] Preferably, the causal closure degradation judgment module is used to establish a causal closure relationship based on the sequential correspondence between changes in control commands, changes in actuator action feedback, changes in power channel status, and changes in temperature response, vibration response, or pressure response, and to identify causal closure relationship anomalies when the sequential correspondence does not meet preset conditions.
[0016] Preferably, the causal closure degradation determination module determines the causal closure anomaly as a valid anomaly only when the same causal closure anomaly occurs repeatedly in at least two independent complete operating cycles.
[0017] Preferably, the causal closure degradation determination module is used to output a fault degradation indicator when the amplitude deviation, duration deviation and recovery lag deviation corresponding to the valid anomaly simultaneously meet preset conditions.
[0018] Preferably, the expected failure time interval determination module is used to determine the expected failure time interval based on the threshold crossing order corresponding to the fault deterioration indicator, the next executable maintenance window, and the spare parts preparation lead time.
[0019] Preferably, it also includes an on-board edge processing unit, a vehicle-to-ground communication unit, and a ground center processing unit. The on-board edge processing unit is used to generate a topology status identifier, an event mask, and a causal closure relationship determination index that characterizes the anomaly determination result of the causal closure relationship. The vehicle-to-ground communication unit is used to upload the topology status identifier, the event mask, and the causal closure relationship determination index. The ground center processing unit is used to execute the expected failure time interval determination module, the health management task generation module, and the verification and unlocking module based on the uploaded results.
[0020] Preferably, the verification unlocking module only updates the baseline version number and removes the freeze mark when, after maintenance, under the same topology status identifier as described above, the entire target verification line segment is run according to the target verification control command sequence, the causal closure relationship determination index indicates that no causal closure relationship abnormality is detected, and there are no continuous abnormal segments except for the abnormal segments masked by the event masking mask.
[0021] Compared with existing technologies, the above technical solution has the following advantages: 1. This invention constructs comparable sample pairs based on the common constraints of topology status identifiers, operating phases, and control command sequences, so that the data involved in fault degradation determination are under the same operating topology context and control semantics. This can effectively avoid the direct comparison of non-homogeneous operating data under conditions of changes in train formation, control mode switching, line section changes, software version changes, or parameter version changes, thereby significantly improving the consistency and reliability of the basic data for fault determination.
[0022] 2. By setting an event mask generation mechanism and adaptively determining the masking interval in conjunction with the response stabilization duration, this invention can exclude transient disturbance data during transition processes such as group reconnection, line switching, control mode switching, software version switching, parameter version switching, and post-maintenance adaptive trial operation from fault degradation judgment, thereby effectively suppressing false fault interference and significantly reducing the false alarm rate.
[0023] 3. By establishing a causal closed relationship between control commands, execution responses, and physical state responses, and combining an effective anomaly judgment mechanism that recurs across a complete operating cycle with a joint judgment mechanism for amplitude deviation, duration deviation, and recovery lag deviation, this invention can more accurately distinguish between occasional fluctuations and true degradation trends, thereby improving the accuracy and stability of fault degradation identification.
[0024] 4. By setting an initial inheritance, freezing, verification, and unlocking update mechanism for the health baseline, this invention ensures that data from newly replaced components, new software versions, or new parameter versions during the transition phase will not directly enter the health baseline. This effectively avoids health baseline contamination and improper updates, thereby improving the continued effectiveness of subsequent fault prediction and health management results.
[0025] 5. By incorporating the threshold crossing order corresponding to the fault degradation identifier, the next executable maintenance window, and the spare parts preparation lead time into the process of determining the expected failure time interval, this invention enables the expected failure time interval to be closer to the actual operation and maintenance schedule requirements, thereby improving the foresight and rationality of maintenance plan formulation, spare parts preparation, and maintenance window selection.
[0026] 6. By setting up a health management task generation module and a verification template library, this invention can automatically output target maintenance actions, target maintenance windows, target verification line sections, and target verification control command sequences based on different topology status identifiers and different command sequence verification values, thereby improving the standardization of maintenance execution and post-maintenance verification and reducing reliance on manual experience.
[0027] 7. By setting up an on-board edge processing unit, a vehicle-to-ground communication unit, and a ground-based central processing unit, this invention enables key results such as topology status identifiers, event masking codes, and causal closure relationship determination indexes to be preprocessed on the on-board side and uploaded to the ground-based central processing unit for comprehensive decision-making. This ensures the accuracy of fault prediction and health management while also taking into account the efficiency of vehicle-to-ground collaborative processing and communication load control.
[0028] 8. This invention is applicable to door systems, braking systems, traction systems, and other key equipment in rail transit. It has good system versatility and engineering applicability, and can improve the level of fault prediction and health management of rail transit systems under complex operating conditions. It also helps to reduce unplanned downtime, improve operation and maintenance efficiency, and enhance operational safety assurance capabilities. Attached Figure Description
[0029] Figure 1 This is a schematic diagram of the structure of the rail transit system operation fault prediction and health management system of the present invention; Figure 2 This is a schematic diagram of the fault prediction and health management process for rail transit systems according to the present invention; Figure 3 This is a schematic diagram illustrating the construction of comparable sample pairs and event masking in this invention; Figure 4 This is a schematic diagram illustrating the freezing, verification, unlocking, and updating of the health baseline in this invention; Figure 5 This is a schematic diagram illustrating the freezing, verification, unlocking, and updating of the health baseline in this invention; Figure 6 This is the execution response delay variation curve under multiple running cycles of the present invention; Figure 7 This is a curve showing the change in the accuracy of fault degradation identification over observation time in this invention. Figure 8 This is a schematic diagram of the process of freezing, unlocking and updating the health baseline in this invention. Detailed Implementation
[0030] The following is combined Figures 1 to 8 The present invention will be further described in detail below. It should be understood that the following specific embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, equivalent substitutions, modifications, combinations or deletions made to the technical features of the embodiments without departing from the concept of the present invention should all fall within the scope of the disclosure of the present invention.
[0031] like Figures 1 to 4As shown, this embodiment provides a rail transit system operation fault prediction and health management system. This system is deployed between rail transit trains, station-side data acquisition nodes, and a ground-based operation and maintenance center. It is used to predict and manage the operation faults of the traction system, braking system, door system, auxiliary power supply system, network control system, and other monitored objects related to safe operation. The system includes a data aggregation module, a topology status identifier generation module, a comparable sample pair construction module, an event mask generation module, a health baseline management module, a causal closure degradation judgment module, an expected failure time interval determination module, a health management task generation module, and a verification unlocking module. In this embodiment, to meet the collaborative needs of online train processing and centralized ground processing, the system also includes an onboard edge processing unit, a vehicle-to-ground communication unit, and a ground-based central processing unit. For standardized management of subsequent verification conditions, the system also includes a verification template library. The definition of a complete operating cycle, various thresholds, various verification templates, and line section configuration relationships can be pre-stored in the system configuration file and the operation and maintenance database of the ground-based central processing unit, and can be maintained by operation and maintenance personnel according to the actual conditions of the line.
[0032] In this embodiment, the data aggregation module is used to collect the operating data, control command data, event log data, maintenance history data, and component genealogy identifiers of the monitored object. Operating data may include traction current, traction voltage, inverter switching status, brake cylinder pressure, brake command execution time, door opening / closing displacement, door opening / closing duration, auxiliary converter output temperature, fan speed, network delay, vibration amplitude, shaft temperature, and other data characterizing operating and physical state responses. Control command data may include traction level commands, brake level commands, door opening / closing commands, air conditioning start / stop commands, auxiliary power supply switching commands, network reset commands, and other command sequences issued by the upper-level control system to the object being executed. Event log data is used to record events occurring during operation, such as train reconnection events, line switching events, control mode switching events, software version switching events, parameter version switching events, component replacement events, and post-maintenance adaptive trial operation events. Maintenance history data is used to record fault codes, fault occurrence time, disassembly and inspection results, maintenance work orders, component replacement information, maintenance duration, trial operation records, retest records, and recovery confirmation records. The component lineage identifier is used to uniquely identify the source, model, batch, supplier, manufacturing date, installation location, last maintenance date, and cumulative mileage of a target component, so as to accurately associate it during the health baseline recall, inheritance, and freeze unlocking processes.
[0033] To ensure that operational data, control command data, and event log data can be compared on the same timeline, this embodiment uses a unified clock reference for time synchronization. The unified clock reference can be provided by the master clock signal output by the train network control system, or it can be corrected by the onboard edge processing unit based on the standard time provided by the timing module. For data sources with inconsistent sampling frequencies, such as control command data recorded based on event triggers while operational data is recorded at a fixed sampling period, they can be mapped to the same timeline through timestamp alignment, resampling, or interpolation. In this embodiment, the standard sampling period for operational data can be 1 second, and for some rapidly changing signals, it can be 0.1 seconds. Both control command data and event log data retain their original timestamps and are uniformly aligned before constructing comparable sample pairs. Through the above processing, it can be ensured that subsequent command sequence checksum generation, event mask determination, and causal closure relationship determination are all based on a unified timeline.
[0034] The topology status identifier generation module generates topology status identifiers based on the train formation status, control mode, track section, software version, and equipment configuration. In this embodiment, the train formation status can be single-train formation, multiple-unit formation, separate formation, or temporary rescue formation; the control mode can be automatic operation mode, manual driving mode, restricted operation mode, section maintenance mode, and trial operation mode; the track section can be divided according to the up / down direction, station intervals, gradient level, curve level, and power supply zone; the software version and parameter version represent the software version number and parameter set version number currently in operation of the corresponding controller, respectively; the equipment configuration represents the configuration relationship of the traction control unit, braking control unit, door control unit, auxiliary power supply unit, and network control unit under the current train formation. The topology status identifier generation module can concatenate the train formation status identifier, control mode identifier, track section identifier, software version identifier, parameter version identifier, and equipment configuration verification value in a predetermined order to form the topology status identifier. The equipment configuration verification value can be generated after consistency verification of the current carriage composition, equipment installation position, equipment model, and configuration relationship. By using the above methods, the runtime data can be placed within the same strictly defined runtime topology before comparison, thereby reducing the risk of non-homogeneous runtime data being directly compared from the source.
[0035] To clarify the formation process of the topology status identifier and device configuration verification value, this embodiment provides a specific example. Suppose a train is currently a six-car formation, in manual driving mode, located in the third section of the uphill direction, with software version V3, parameter version P2, and the current device configuration relationship, after verification, corresponds to device configuration verification value A17. Then, a topology status identifier "6-M2-U3-V3-P2-A17" can be formed. Here, "6" represents a six-car formation, "M2" represents manual driving mode, "U3" represents the third section of the uphill direction, "V3" represents software version V3, "P2" represents parameter version P2, and "A17" represents device configuration verification value A17. The device configuration verification value itself can be calculated as follows:
[0036] In the formula, This indicates the device configuration verification value. This indicates the number of device entries participating in the verification. Indicates the first The vehicle model or equipment model code of each device. Indicates the first The installation location code for each device Indicates the first The connection relationship code for each device. Using the above expression, device model, installation location, and connection relationship can be combined into a device configuration check value that can be used for consistency assessment. The component lineage identifier can be formed by sequentially concatenating the supplier code, batch number, installation date, installation location number, and number of repairs, for example, "G05-B12-20260108-D3-02," which corresponds to supplier code G05, batch number B12, installation date January 8, 2026, installation location D3, and number of repairs 02, respectively. In this way, the component lineage identifier can not only uniquely identify the target component but also provide a stable foundation for subsequent health baseline inheritance and history tracking.
[0037] The comparable sample pair construction module is used to construct comparable sample pairs from current and historical operating data under the same topology status identifier, according to the same operating stage and the same control command sequence. In this embodiment, the operating stage can be further subdivided into the station stop and wait-to-depart stage, the start-up traction stage, the constant speed cruise stage, the coasting stage, the braking and entering stage, the station stop and door opening stage, the station stop and door closing stage, and the intra-section trial operation stage. The control command sequence records the command type, issuance time, duration, target level, and cancellation time of each control command in chronological order. The comparable sample pair construction module first performs time-series encoding on the control command data to generate a command sequence check value, and then uses the consistency of topology status identifier, operating stage, inter-station operating direction, and command sequence check value as the sample pairing conditions. In order to achieve consistency determination of the control command sequence, the command sequence check value in this embodiment is calculated using the following expression:
[0038] In the formula, This represents the command sequence check value. Indicates the number of commands. Indicates the first The command type encoding of the command. Indicates the first The target level code of the command, Indicates the first The duration of the command, in seconds.
[0039] For example, in a standard verification process for a vehicle door system, the control command sequence is as follows: open door command, hold command, and close door command. If the open door command has a command type code of 2, a target level code of 3, and a duration of 12 seconds; the hold command has a command type code of 5, a target level code of 2, and a duration of 8 seconds; and the close door command has a command type code of 1, a target level code of 0, and a duration of 5 seconds, then the command sequence check value for this control command sequence is calculated as follows:
[0040]
[0041] If the command sequence checksum corresponding to the historical samples is also 15743, and the topology status identifier, operating stage, and inter-station operating direction are consistent, then the current sample and the historical sample can be constructed as a comparable sample pair; if the command sequence checksums are different, then even if the two are located in the same inter-station segment, they do not constitute a comparable sample pair. Therefore, this embodiment does not directly compare data with similar surface conditions, but rather ensures sample comparability through constraints formed by the topology status identifier, operating stage, and control command sequence.
[0042] In certain special scenarios, the historical database may not contain historical samples that simultaneously meet the criteria of consistent topology status identifiers, consistent operational phases, consistent inter-station operational directions, and consistent command sequence checksums. In such cases, this embodiment does not forcibly construct comparable sample pairs but instead implements a fallback mechanism. This fallback mechanism may include any one or a combination of the following: First, only output observation prompts, without outputting fault degradation indicators; second, maintain the current health baseline frozen and do not perform baseline version number updates; third, filter candidate samples only based on topology status identifiers and operational phases, and use them as reference samples rather than comparable samples; fourth, record the current sample as a sample to be accumulated, and participate in the comparison only after samples with the same command sequence checksums are formed later. Through this processing method, new judgment biases can be avoided by forcibly comparing samples when there are no truly comparable sample pairs.
[0043] The event mask generation module generates event masking masks corresponding to comparable sample pairs based on event log data. The event masking masks are primarily generated for group reconnection events, line switching events, control mode switching events, software version switching events, parameter version switching events, and post-maintenance adaptive trial operation events, used to mask transient data during the transition process. The boundaries of the event masking masks are not fixed but adaptively determined based on the response stabilization time from the change in control command after the event to the execution response reaching a stable state. The response stabilization time can be determined using the following method: when the difference between the execution responses at adjacent sampling times is not greater than a preset stabilization threshold for several consecutive times, the execution response is considered to have entered a stable state. To avoid overlapping symbol meanings, the response stabilization time in this embodiment uses the following expression:
[0044] In the formula, Indicates the duration of a stable response. This indicates the moment when a steady state is reached. Indicates the time when the event occurred.
[0045] In this embodiment, taking the door motor current as an example, the event occurrence time is set to 100s, and the motor current sampling values after the event are shown in Table 1.
[0046] Table 1 shows examples of motor current sampling values after the event. ; In this embodiment, the response stabilization threshold is set to 0.02A, and a stable state is defined as when the difference between three consecutive adjacent samples is no greater than 0.02A. Table 1 shows that the current difference between 107s and 108s is 0.01A, and the current difference between 108s and 109s is also 0.01A. Furthermore, the current has entered a small convergence phase starting from 106s. Therefore, 109s can be considered the moment when the execution response reaches a stable state. Substituting this data into the above expression, we obtain:
[0047] Therefore, the masking interval corresponding to the event masking mask can be set to 100s to 109s. Figure 5 The shaded area represents the masking region for this type of event. Figure 5 It can be seen that within the event shielding interval, the current temperature response has a large disturbance relative to the healthy baseline temperature response. However, this disturbance is a transitional deviation caused by mode switching and version switching and should not be directly used as the basis for judging fault degradation. Within the unshielded data interval, the current temperature response deviates from the healthy baseline continuously after 70 seconds, which better reflects the true degradation trend.
[0048] To further improve the completeness of event masking processing, when the masking intervals of multiple events overlap, this embodiment can perform a merge and aggregate operation on the multiple masking intervals to form a continuous masking interval. For example, when the masking interval for a control mode switching event is 100s to 109s and the masking interval for a software version switching event is 106s to 114s, the final effective masking interval can be merged into 100s to 114s. When multiple events have different importance levels, the masking interval with the later end time can also be used as the final effective masking interval to ensure that the entire transition phase is excluded from the fault degradation determination.
[0049] The health baseline management module stores and retrieves corresponding health baselines based on topology status identifiers and component lineage identifiers. In this embodiment, the health baseline management module establishes a dual-index health baseline library. The first index is used to locate the set of health baselines under the same operating topology state, and the second index is used to locate the health baseline instance corresponding to the historical evolution of the same target component. Each health baseline corresponds to a unique baseline version number, which is used to distinguish health baselines formed at different times under the same topology state and the same component lineage identifier. Health baselines may include command response delay baselines, power channel establishment time baselines, temperature rise curve baselines, vibration waveform characteristic baselines, pressure establishment slope baselines, and sequential relationship baselines between key signals, etc. When comparing, the health baseline management module does not indiscriminately call global baselines, but first locates the same operating topology environment by topology status identifier, and then locates the historical health baseline of the corresponding target component by component lineage identifier, thereby improving the accuracy of baseline retrieval.
[0050] When a component replacement event occurs, the health baseline management module also retrieves an initial inherited baseline from a list of verified health baselines of the same model and configuration, based on the component lineage identifier, device configuration verification value, and topology status identifier of the replaced target component. This initial inherited baseline is then associated with the replaced target component. In other words, after replacing the target component, the system does not immediately restart baseline accumulation from scratch. Instead, it inherits a set of initial inherited baselines from a list of verified health baselines of the same model and configuration to ensure that health management can quickly return to a usable state after maintenance. For example, when a door control unit is replaced with a product of the same model and version, the system can retrieve an initial inherited baseline from a verified door control unit health baseline with the same wiring, control mode, and device configuration, and establish a mapping relationship with the replaced target component. This initial inherited baseline is not immediately considered the final stable baseline but enters the subsequent freezing, verification, and unlocking process.
[0051] The verification and unlocking module applies a freeze flag to the health baseline after a component replacement event, software version switch event, or parameter version switch event. It prohibits updating the baseline version number until the verification of the target verification line section and the corresponding target verification control command sequence is completed. The freeze flag prevents initial transitional data after maintenance or version changes from directly entering the health baseline. After the health management task is completed, the verification and unlocking module generates verification results based on the same topology status identifier and the same target verification control command sequence. Based on the verification results, it either unfreezes the health baseline and allows updates, or maintains the frozen state. In other words, this embodiment adopts a "freeze-verify-unfreeze" health baseline governance mechanism, rather than an open governance mechanism of "update as soon as data arrives." This mechanism is particularly suitable for scenarios where rail transit systems still need to undergo a stable reconstruction process after mode switching, version switching, and component replacement.
[0052] To make the verification and unlocking process clearer, this embodiment further specifies that the verification pass criteria in the verification template library include at least one or more of the following: First, the causal closure relationship determination index indicates that no causal closure relationship anomaly was detected; second, the execution response delay does not exceed the corresponding health baseline allowable threshold; third, there are no continuous abnormal segments except for abnormal segments masked by the event masking mask; fourth, the key operating data within the target verification line segment returns to the health baseline allowable range. By writing the above criteria into the verification template library, different device objects can adopt configurable, traceable, and consistent pass standards during the verification and unlocking phase.
[0053] The causal closure degradation judgment module is used to determine the deviation of comparable sample pairs from the unmasked data range defined by the event mask, based on the causal closure relationship between control commands, execution responses, and physical state responses, and outputs a fault degradation indicator. The causal closure relationship refers to a repeatable and verifiable sequential correspondence between changes in control commands, actuator action feedback, power channel state, and changes in temperature, vibration, or pressure responses under normal operating conditions. For example, in a vehicle door system, the normal sequence can be defined as "door opening command issued—door actuator started—door leaf displacement established—door opening signal appears—vibration decays and enters a stable state." If "the door actuator's action delay increases significantly after the door opening command is issued, the door leaf displacement establishment is delayed, and the time of the opening signal continues to be delayed," it can be identified as an abnormal causal closed relationship. In a braking system, the normal sequence can be defined as "brake command issued—solenoid valve action feedback—brake cylinder pressure established—vehicle deceleration established—pressure drops after release." If "the brake command has been issued but the brake cylinder pressure establishment is delayed, the deceleration establishment is delayed, or the pressure drops abnormally slowly after release," it can be identified as an anomaly. In a traction system, the normal sequence can be defined as "traction stage command issued—inverter enabled—power channel current rises—motor torque established—temperature rise gradually enters a steady state." If "the traction command has been issued but the inverter enable is delayed, the current rise is abnormally slow, or the temperature rise curve deviates from the healthy baseline," it can be identified as an anomaly. By predefining and template-based storage of causal closure relationships in different systems, this embodiment enables the causal closure degradation judgment module to use a consistent logical framework for fault degradation identification on different objects.
[0054] To improve the robustness of anomaly identification, this embodiment does not directly identify a single causal closure anomaly as a fault degradation. Instead, it requires that the same causal closure anomaly recur within at least two independent complete operating cycles before it is determined to be a valid anomaly. A complete operating cycle can be defined according to different equipment objects. For example, for a door system, it can be defined as a complete door opening-stopping-closing cycle; for a traction system, it can be defined as a start-tractioning-cruising-braking-stopping cycle; and for a braking system, it can be defined as a brake establishment-holding-release cycle. Furthermore, the definition of a complete operating cycle can be pre-stored in the system configuration and maintained by maintenance personnel according to equipment category, thereby avoiding temporary manual judgment during actual operation. After a valid anomaly is established, the causal closure degradation judgment module further comprehensively judges the amplitude deviation, duration deviation, and recovery lag deviation corresponding to the valid anomaly. Only when all three simultaneously meet preset conditions is a fault degradation flag output. To avoid using the same basic symbol for different technical names, the three deviations in this embodiment are calculated using the following expressions:
[0055]
[0056]
[0057] In the formula, Indicates amplitude deviation. This represents the target response amplitude of the current sample. This represents the target response magnitude at the health baseline; Indicates a deviation in duration. Indicates the duration of the current sample. Indicates the duration of the health baseline; Indicates recovery of lag deviation, This indicates the time required for the current sample to recover to a steady state.
[0058] This indicates the time required for a healthy baseline to return to a stable state.
[0059] For example, in a certain car door opening action, the steady-state current amplitude corresponding to the healthy baseline is 1.25A, and the steady-state current amplitude corresponding to the current sample is 1.40A; the door opening duration corresponding to the healthy baseline is 1.80s, and the door opening duration corresponding to the current sample is 2.05s; the vibration recovery time corresponding to the healthy baseline is 0.35s, and the vibration recovery time corresponding to the current sample is 0.52s. Therefore, we can obtain:
[0060]
[0061]
[0062] If the amplitude deviation threshold is set to 8%, the duration deviation threshold is set to 0.08s, and the recovery hysteresis deviation threshold is set to 0.10s, then all three deviations in this sample exceed the thresholds. If the same anomaly occurs repeatedly in at least two independent complete operating cycles, then the anomaly is considered a valid anomaly, and a fault degradation indicator is output. Figure 6 The curves showing the variation of the gantry crane's execution response delay over multiple operating cycles are presented. Figure 6 It can be seen that the execution response delay in the comparative sample fluctuates but remains generally stable across multiple operating cycles, while the execution response delay in the sample identified by the present invention gradually increases with the operating cycle, reflecting a continuous deterioration trend; by combining cross-cycle repetition and joint judgment of three types of deviation, it is possible to effectively distinguish between occasional fluctuations and real deterioration.
[0063] To clarify the sources of various thresholds and weights, this embodiment further explains their determination methods. The amplitude deviation threshold, duration deviation threshold, and recovery hysteresis deviation threshold can be determined by statistical analysis of the distribution of historical healthy samples and historical fault samples, or by equipment factory calibration values, line operation experience values, or preset strategies of the maintenance department. In a specific implementation, at least three months of healthy and fault samples can be collected first, and the percentile distribution of each deviation in the healthy samples and the mean distribution in the fault samples can be calculated. Then, the threshold that effectively separates healthy and fault samples is used as the formal threshold. Furthermore, the thresholds corresponding to different equipment objects can be configured and stored separately in the system configuration. For example, the door system uses one set of thresholds, the braking system uses another set, and the traction system uses yet another set, thereby ensuring that the system-level solution can adapt to the operating characteristics of different objects. The weights in determining the expected failure time interval can also be obtained through historical regression analysis, cross-validation, or calibration based on maintenance experience. In this embodiment, 0.4, 0.4, and 0.2 are used, which is only an example for illustrative purposes and does not constitute a limitation on the implementation method.
[0064] The estimated failure time interval determination module is used to determine the estimated failure time interval based on the fault degradation identifier, maintenance history data, and executable maintenance window data. To make this calculation process more directly correlated with the technical factors of the threshold crossing order corresponding to the fault degradation identifier, the next executable maintenance window, and the spare parts preparation lead time, this embodiment first introduces a threshold crossing order index. The threshold crossing order index indicates which preset threshold stage the current fault degradation identifier has crossed, such as the first threshold crossing, the second threshold crossing, or the third threshold crossing. For different threshold crossing order indices, the corresponding historical average remaining days, historical median remaining days, and the average remaining days of the most recent similar faults are calculated. Using the threshold crossing order index as... The expected failure center time can be calculated using the following expression:
[0065] In the formula, This indicates that the threshold crossing order index is The number of remaining days corresponding to the expected failure center time. This indicates the historical average remaining days corresponding to the order in which the threshold is crossed. This indicates the number of remaining historical median days corresponding to the order in which the threshold is crossed. This represents the average remaining days for the three most recent faults of the same type that crossed the same threshold.
[0066] Based on this, the next executable maintenance window and spare parts preparation lead time are then incorporated into the process of determining the expected failure time interval. The expected failure time interval can be expressed using the following interval expression:
[0067] In the formula, Indicates the expected failure time range. Indicates time margin. This indicates the time point corresponding to the next executable maintenance window. This indicates the lead time for spare parts preparation.
[0068] Therefore, the lower limit of the expected failure time interval is related not only to the expected failure center time and time margin, but also to the latest preparation time corresponding to the spare parts preparation lead time; the upper limit of the expected failure time interval is related not only to the expected failure center time and time margin, but also directly constrained by the next executable maintenance window. Thus, in this embodiment, the determination of the expected failure time interval is the result of the combined effect of the threshold crossing order corresponding to the fault degradation indicator, the next executable maintenance window, and the spare parts preparation lead time.
[0069] For example, if a brake control unit has already crossed the second threshold stage, its threshold crossing order index is 2. If historical statistics show that the historical average remaining days corresponding to the second threshold stage is 7.0 days, the historical median remaining days is 8.0 days, and the average remaining days corresponding to the second threshold stage for the three most recent similar faults is 6.0 days, then:
[0070] If the time margin is 1 day, the next feasible maintenance window is on the 7th day, and the spare parts preparation lead time is 2 days, then:
[0071]
[0072] Converted to a calendar interval convenient for operation and maintenance, this can be recorded as day 6 to day 7. Therefore, the expected failure time interval not only reflects the fault degradation trend itself but also directly constrains the maintenance window and spare parts preparation conditions, better meeting the actual needs of operation and maintenance decisions. Based on this, the health management task generation module can select day 7 as the optimal maintenance window and provide the target maintenance action, target maintenance window, target verification line section, and target verification control command sequence in the health management task.
[0073] The health management task generation module generates health management tasks based on fault degradation indicators and expected failure time intervals. A health management task includes at least a target maintenance action, a target maintenance window, a target verification line segment, and a target verification control command sequence. The target maintenance action can be cleaning, tightening, calibration, software rollback, parameter resetting, component replacement, or assembly replacement. The target maintenance window specifies the specific time range for maintenance. The target verification line segment specifies the line segment where verification should be performed after maintenance. The target verification control command sequence specifies which control command sequence should be executed during verification. To ensure verification consistency, this embodiment also includes a verification template library. The verification template library stores various standard verification templates according to topology status identifiers and command sequence check values. The verification template library may include at least the following fields: template number, topology status identifier, command sequence check value, target verification line segment, target verification control command sequence, verification pass criterion, allowed anomaly segment types to be masked, and number of verification executions. For example, for replacing the door control unit, a target verification control command sequence of "stop and open door—stop and close door—repeat three times" can be configured and associated with the target verification line sections corresponding to straight sections and low wind pressure platform environments; for switching the software version of the brake control unit, a target verification control command sequence of "start—medium speed cruise—level braking—emergency braking simulation" can be configured and associated with verification sections with gentle gradients. When outputting tasks, the health management task generation module can call the corresponding template from the verification template library according to the current topology status identifier and command sequence verification value.
[0074] The vehicle-mounted edge processing unit, vehicle-to-ground communication unit, and ground-based central processing unit together constitute the distributed deployment structure of this embodiment. The vehicle-mounted edge processing unit is mainly used to generate topology status identifiers, event masking masks, and causal closure relationship determination indexes representing the results of causal closure relationship anomaly determinations. The causal closure relationship determination index is not the original anomaly data itself, but a structured expression of information such as unmasked data intervals, anomaly types, anomaly duration intervals, and anomaly levels, used to reduce vehicle-to-ground communication bandwidth consumption. Further, in this embodiment, the causal closure relationship determination index may include at least the following fields: anomaly type field, anomaly start time field, anomaly end time field, anomaly level field, corresponding topology status identifier field, and corresponding command sequence verification value field. Through the above structured field settings, the ground-based central processing unit can not only know whether a causal closure relationship anomaly exists, but also know the operating topology context in which the anomaly occurred, how long the anomaly lasted, and the degree of the anomaly. The vehicle-to-ground communication unit is used to upload topology status identifiers, event masking masks, and causal closure relationship determination indexes, and to receive health management tasks and verification templates issued by the ground-based central processing unit. The ground-based processing unit is used to determine the expected failure time interval, generate health management tasks, and perform verification and unlocking processes based on the uploaded results. It also maintains a global health baseline database, a global maintenance history database, and a verification template database. This distributed structure reduces the processing load on the vehicle and facilitates unified scheduling of spare parts, work orders, and verification resources by the ground operations and maintenance center.
[0075] To clarify the meaning of "continuous abnormal segments," this embodiment further defines it as follows: a continuous abnormal segment refers to an abnormal segment within a data interval not masked by an event mask, where multiple consecutive sampling points satisfy the causal closure relationship anomaly determination condition, and the duration of this continuous abnormal state reaches a preset minimum duration. The preset minimum duration can be configured for different devices; for example, it can be 0.5s for a door system, 1.0s for a braking system, and 2.0s for a traction system. This definition ensures that the statement "there are no continuous abnormal segments other than those masked by the event mask" has clear engineering implications and operability.
[0076] The present invention will be further described below through three parallel embodiments.
[0077] Example 1: This embodiment focuses on a train door system. In a certain urban rail transit line, a train transitions from automatic to manual operation mode in the upstream section, simultaneously completing a software version switch for a door control unit. The data aggregation module continuously collects operational data, control command data, event log data, maintenance history data, and component family identifiers for this door control unit. The topology status identifier generation module generates a corresponding topology status identifier based on the train's six-car formation, manual operation mode, upstream section between stations, new software version, predetermined parameter set, and unchanged equipment configuration. Due to the existence of control mode switching and software version switching events, the event mask generation module first generates an event mask based on the event logs and marks the transition period from the switching moment until the door control command change and the door operator's response stabilize again as the masked interval. Subsequently, the comparable sample pair construction module selects historical samples from historical operational data that share the same topology status identifier, the same station stop and door opening phase, the same control command sequence, and the same direction, and constructs comparable sample pairs with the current operational data. The causal closure degradation judgment module only checks whether the sequential correspondence of "door opening command issuance - door actuator start-door leaf displacement establishment - door opening signal appearance - vibration decay to stability" meets preset conditions within the unshielded data range. If the same anomaly occurs repeatedly in multiple complete door opening-stopping-closing cycles, and the amplitude deviation, duration deviation, and recovery lag deviation all exceed the threshold, a fault degradation indicator is output. The expected failure time interval determination module comprehensively considers historical data, the threshold crossing sequence corresponding to the second threshold crossing, the next executable maintenance window, and the spare parts preparation lead time to give the expected failure time interval. Based on this, the health management task generation module outputs a health management task: "Inspect and recalibrate the door control unit at the night window the following day, and perform three standard door opening and closing verifications in the specified flat section." After maintenance is completed, the verification unlocking module, under the same topology status identifier, calls the target verification line segment and the target verification control command sequence, keeping the healthy baseline frozen. Only when no causal closure anomalies are detected in three verifications, and no consecutive abnormal segments exist except for those masked by the event masking mask, is the freeze lifted and the baseline version number updated. Thus, this embodiment fully supports the application logic of the system solution on the vehicle door system.
[0078] Example 2: This embodiment focuses on the braking system. A train undergoes nighttime maintenance, during which the brake control unit is replaced, and the train is put into operation the following day. The data aggregation module collects braking command data, brake cylinder pressure data, valve action feedback data, vehicle deceleration data, event log data, and maintenance history data from the brake control unit. The topology status identifier generation module generates a topology status identifier based on the current train formation, operating mode, section information, software version, and equipment configuration relationships. The health baseline management module, based on the component family identifier, equipment configuration verification value, and topology status identifier of the replaced target component, retrieves the initial inherited baseline from the verified health baselines of the same model and configuration and applies a freeze flag. The causal closure degradation judgment module determines whether the sequential correspondence of "brake command issuance—solenoid valve action feedback—brake cylinder pressure establishment—vehicle deceleration establishment—pressure drop after release" meets preset conditions within the unshielded data range. If the brake cylinder pressure establishment delay and deceleration establishment lag are detected to repeatedly occur in at least two complete brake establishment-hold-release cycles, and the amplitude deviation, duration deviation, and recovery lag deviation all exceed thresholds, a fault degradation identifier is output. The estimated failure time interval determination module determines the estimated failure time interval based on the threshold crossing sequence corresponding to the current failure degradation stage of the braking system, the executable maintenance window, and the spare parts preparation lead time. The health management task generation module outputs the task "Check the brake control unit parameter matching status during the night window of the seventh day and perform level braking and emergency braking verification in the specified smooth section." If, after completing the operation of the entire target verification line section, no anomalies are detected in the causal closure relationship determination index representation, and there are no continuous abnormal segments except for the abnormal segments masked by the event masking mask, the verification unlocking module unfreezes and allows the baseline version number to be updated.
[0079] Example 3: This embodiment focuses on the traction system. A train is put back into operation after completing a traction inverter parameter version switch. The data aggregation module continuously collects traction level commands, inverter enable status, power channel current, drive motor temperature rise, vibration amplitude, event log data, and maintenance history data. The topology status identifier generation module generates a topology status identifier based on the current train formation status, control mode, up or down line section, software version, parameter version, and equipment configuration relationship. Due to the existence of parameter version switch events, the event mask generation module generates corresponding masked intervals based on the event logs. The comparable sample pair construction module selects historical samples with the same topology status identifier, the same starting traction stage, the same control command sequence, and the same direction from historical operating data, and constructs comparable sample pairs with the current operating data. The causal closure degradation judgment module determines the sequential correspondence of "traction level command issuance—inverter enable—power channel current rise—motor torque establishment—temperature rise gradually enters steady state" within the unmasked data interval. If slow current rise and temperature rise curve deviation occur in multiple consecutive complete start-to-cruise-brake-stop cycles, they are identified as valid anomalies, and a fault degradation flag is output based on amplitude deviation, duration deviation, and recovery lag deviation. The expected failure time interval determination module determines the expected failure time interval based on the threshold crossing sequence corresponding to the current fault degradation stage of the traction system, the next executable maintenance window, and the spare parts preparation lead time. The health management task generation module outputs a task to "perform traction parameter verification in the specified window and perform traction level step verification in the specified section." If no causal closure anomaly is detected after verification, and there are no continuous abnormal segments except for those masked by the event masking mask, the freeze is lifted and the baseline version number can be updated.
[0080] Let's take a component replacement event as an example to illustrate the general application process of the initial inherited baseline and the freeze-unlock mechanism. A brake control unit is replaced due to a field fault. The replaced target component has a new component family identifier, but its model, software version, parameter version, and equipment configuration relationship are consistent with other trains of the same type on the line. The health baseline management module, based on the component family identifier, equipment configuration verification value, and topology status identifier of the replaced target component, calls the initial inherited baseline from the verified health baselines of the same model and configuration and establishes a new association. The verification-unlock module then applies a freeze mark to this health baseline. During this period, even if new operational data is collected, it is only used for fault determination and task generation, and not for updating the baseline version number. After the train completes the specified start, cruise, level braking, and brake release verification conditions, if the causal closure relationship determination index shows that no causal closure relationship abnormality was detected, and the abnormality only occurs in the brief transition segment covered by the event mask, then the verification-unlock module unfreezes and allows the baseline version number to be updated; if there are still continuous abnormal segments that are not masked, the frozen state is maintained, and a new health management task is generated. Therefore, this embodiment can not only perform health management under normal operating conditions, but also perform steady-state reconstruction after component replacement, software version switching, and parameter version switching.
[0081] In this embodiment, the following standardized processing logic can also be adopted: the verification unlocking module only updates the baseline version number and removes the freeze mark when, after maintenance, under the same topology status identifier as described above, it completes the operation of the entire target verification line segment according to the target verification control command sequence, the causal closure relationship determination index indicates that no causal closure relationship anomaly is detected, and there are no continuous abnormal segments except for the abnormal segments masked by the event mask. Through this supplementary processing logic, it can be further clarified that the update of the healthy baseline is not based on local verification segments, but on a comprehensive judgment based on the complete verification line segment, the complete verification command sequence, and the complete anomaly determination result.
[0082] To verify the improvement effect of the present invention's solution compared to existing rail transit vehicle health management approaches, a comparative experiment is set up in this embodiment. Comparative Example 1 adopts the processing flow of the existing solution, namely, using operational data slicing, model library threshold comparison, repeated fault filtering, intermittent fault filtering, and environmental condition influence filtering. It does not set a comparable sample pair construction mechanism under topological state identifier constraints, does not set an event mask adaptive boundary mechanism, and does not set a health baseline freezing, verification, and unlocking update mechanism. Example 1 adopts the solution of the present invention. The test objects are the operation and maintenance data of twelve trains on the same line for ninety consecutive days, including door system fault samples, braking system fault samples, traction system fault samples, and complex operation scenarios such as software version switching, parameter version switching, component replacement, and post-maintenance adaptive trial operation. To illustrate the source and acquisition method of the test data, the operational data in this embodiment is mainly provided jointly by the train network control system, on-board sensor nodes, and ground maintenance database. The basic sampling period for operational data is 1 second, while the sampling period for some fast response signals is 0.1 seconds. Event log data is automatically recorded by the train network control system, maintenance history data is exported from the maintenance work order system and disassembly / inspection record system, and fault labels are jointly determined by fault codes, maintenance work orders, disassembly / inspection results, and recovery confirmation records. The twelve trains used in the test all came from the same line and the same train model platform, and the continuous observation period was ninety days to ensure the consistency of test conditions and the comparability of results. Furthermore, it can be stated that the test data all come from the actual operational data and corresponding maintenance history data of the same train model on the same line, and the fault labels are jointly verified by maintenance work orders, disassembly / inspection results, and recovery confirmation records to enhance the authenticity of the test results. Test conditions are shown in Table 2.
[0083] Table 2 of the test conditions settings ; Table 2 is used to illustrate the differences between this embodiment and the comparative example in terms of experimental condition settings, and in particular to illustrate the key processing steps added in this embodiment in terms of topology state identification, construction of comparable sample pairs, event masking, causal closure degradation determination, and freezing, verification and unlocking of health baseline.
[0084] Under the above test conditions, the false alarm rate of false faults, the accuracy of fault degradation identification, the failure rate of health baseline update, the hit rate of the expected failure time interval, and the rate of reduction of unplanned downtime were used as evaluation indicators, and the results are shown in Table 3.
[0085] Table 3 compares the test results of Comparative Example 1 and Example 1. ; Table 3 illustrates the overall improvement of this embodiment compared to the comparative example in terms of fault degradation identification accuracy, false fault false alarm rate, health baseline update error rate, expected failure time interval hit rate, and unplanned downtime reduction rate.
[0086] Furthermore, in three high-interference scenarios—after the door control unit software version was switched, after the brake control unit was replaced, and after maintenance and adaptive trial operation—the effective anomaly identification and subsequent baseline treatment effects were statistically analyzed, and the results are shown in Table 4.
[0087] Comparison Table 4 for Special Projects in High-Interference Scenarios ; Table 4 illustrates the suppression effect of this embodiment on false alarms and baseline contamination in high interference scenarios, and particularly demonstrates the applicability of this embodiment in scenarios such as version switching, component replacement, and transitional operation after maintenance.
[0088] As shown in Tables 3 and 4, this invention excludes heterogeneous operational data from comparison through a comparable sample pair construction mechanism under topological state identifier constraints; it shields transient deviations caused by mode switching, version switching, and maintenance transition processes from fault determination through an event masking mask and a masking boundary determination mechanism driven by response stability duration; it establishes fault identification based on the sequential correspondence between control commands, execution responses, and physical state responses through a causal closure degradation determination mechanism; and it avoids the direct contamination of the health baseline by new components, software versions, or parameter versions during the transition phase through a health baseline freezing, verification, and unlocking update mechanism. Therefore, under complex operational topology changes, this invention can not only significantly reduce the false alarm rate of spurious faults but also improve the accuracy of fault degradation identification, the hit rate of predicted failure time intervals, and reduce subsequent misjudgments caused by improper health baseline updates, demonstrating superior operational fault prediction and health management effects. Figure 5 It is mainly used to explain that event masking can remove data in transitional intervals. Figure 6 It is mainly used to illustrate that valid exceptions are repetitive across multiple complete runtime cycles. Figure 7 This invention is primarily used to illustrate that the solution presented in this invention has a higher accuracy rate in identifying fault degradation under long-term operation. Figure 8 It is primarily used to illustrate the process and timing relationships of freezing, verifying, and unlocking / updating health baselines. Figure 7 As can be seen, with the extension of continuous observation time, the recognition accuracy of the present invention solution continues to improve and stabilizes at a high level, while the comparative solution remains at a low level, further proving the authenticity and stability of the present invention solution under complex working conditions.
[0089] The beneficial effects of this embodiment are as follows: First, it can implement strict comparability constraints on non-homogeneous operating data in complex scenarios such as changes in train formation, control mode switching, line section changes, software version changes, parameter version changes, and adaptive trial operation after maintenance, thereby improving the consistency of the comparison basis; Second, it can suppress false fault interference caused by transitional events through event masking and its adaptive boundary mechanism, significantly reducing false alarms; Third, it can improve the accuracy of identifying real fault degradation through causal closure degradation judgment and effective anomaly judgment mechanism; Fourth, it can reduce the risk of improper health baseline updates after maintenance and version changes through the inheritance, freezing, verification, and unlocking update mechanism of health baseline; Fifth, it can improve the foresight of spare parts preparation, maintenance window selection, and verification organization through the determination of expected failure time intervals and the generation of health management tasks, thereby reducing unplanned downtime and improving operation and maintenance efficiency.
[0090] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.
Claims
1. A rail transit system operation failure prediction and health management system, characterized in that, include: The data aggregation module is used to collect the operating data, control command data, event log data, maintenance history data, and component genealogy identifiers of the monitored objects. The topology status identifier generation module is used to generate topology status identifiers based on the operating grouping status, control mode, line section, software version, and equipment configuration. The comparable sample pair construction module is used to construct comparable sample pairs from current running data and historical running data under the same topology status identifier, according to the same running stage and the same control command sequence; An event mask generation module is used to generate an event mask corresponding to the comparable sample pair based on the event log data. The health baseline management module is used to store and retrieve the corresponding health baselines according to the topology status identifier and the component lineage identifier; The causal closure degradation judgment module is used to perform deviation judgment on the comparable sample pair within the unmasked data range determined by the event mask, based on the causal closure relationship between the control command, execution response and physical state response, and output a fault degradation indicator. The expected failure time interval determination module is used to determine the expected failure time interval based on the fault deterioration identifier, the maintenance history data, and the executable maintenance window data. The health management task generation module is used to generate a health management task based on the fault deterioration identifier and the expected failure time interval. The health management task includes at least a target maintenance action, a target maintenance window, a target verification line section, and a target verification control command sequence. The verification unlocking module is used to generate a verification result based on the same topology status identifier and the same target verification control command sequence after the health management task is executed, and to unfreeze and allow updates to the health baseline according to the verification result, or to keep it frozen.
2. The rail transit system operations prognostics and health management system of claim 1, wherein, The topology status identifier generation module is used to generate the topology status identifier based on the operation group status identifier, control mode identifier, line section identifier, software version identifier, parameter version identifier, and equipment configuration verification value.
3. The rail transit system operations fault prediction and health management system of claim 2, wherein, The comparable sample pair construction module is used to perform time-series encoding on the control command data, generate command sequence verification values, and construct the comparable sample pair only when the topology status identifier is consistent, the operation stage is consistent, the inter-station operation direction is consistent, and the command sequence verification values are consistent.
4. The rail transit system operations fault prediction and health management system of claim 3, wherein, It also includes a verification template library. The health management task generation module is used to call the corresponding target verification line segment and the target verification control command sequence from the verification template library according to the topology status identifier and the command sequence verification value.
5. The rail transit system operations prognostics and health management system of claim 1, wherein, The event mask generation module is used to generate the event mask for group reconnection events, line switching events, control mode switching events, software version switching events, parameter version switching events, and post-maintenance adaptive trial operation events.
6. The rail transit system operations fault prediction and health management system of claim 5, wherein, The event mask generation module is used to adaptively determine the masking start time and masking end time corresponding to each event based on the response stabilization time from the change of control command after the event occurs until the execution response reaches a stable state.
7. The rail transit system operation fault prediction and health management system according to claim 2, characterized in that, The health baseline management module is used to establish a dual-index health baseline library based on the topology status identifier and the component lineage identifier. Each health baseline in the dual-index health baseline library corresponds to a unique baseline version number.
8. The rail transit system operation fault prediction and health management system according to claim 7, characterized in that, The health baseline management module is used to, when a component replacement event occurs, retrieve an initial inherited baseline from the verified health baselines of the same model and configuration based on the component lineage identifier, the device configuration verification value, and the topology status identifier of the replaced target component, and establish an association between the initial inherited baseline and the replaced target component.
9. The rail transit system operation fault prediction and health management system according to claim 8, characterized in that, The verification unlocking module is used to apply a freeze mark to the health baseline after a component replacement event, software version switch event, or parameter version switch event occurs, and to prohibit updating the baseline version number before the verification of the target verification line segment and the target verification control command sequence is completed.
10. The rail transit system operation fault prediction and health management system according to claim 1, characterized in that, The causal closure degradation judgment module is used to establish the causal closure relationship based on the sequential correspondence between changes in control commands, changes in actuator action feedback, changes in power channel status, and changes in temperature response, vibration response, or pressure response, and to identify causal closure relationship anomalies when the sequential correspondence does not meet preset conditions.
Citation Information
Patent Citations
Health management device and method for intelligent train monitoring, operation and maintenance
CN111806516A
