A turnout full-parameter comprehensive monitoring and AI prediction and early warning method and system

Abstract

The application discloses a turnout full-parameter comprehensive monitoring and AI prediction and early warning method and system, collects switch machine current data and close detection data, generates section division records and determines close elasticity margin by current slope inflection point; identifies section current slope batch drift to form section recession sequence, obtains elasticity critical record by counting margin continuous decline node; generates temperature decoupling index by section cross-temperature zone degradation comparison, determines wear accumulation level and calculates component health index; analyzes margin exhaustion rate slope change to establish exhaustion acceleration record, calculates residual life interval and outputs fault trigger timing; forms urgent early warning set based on fault trigger timing, infers each fault severity and maintenance response time limit by a prediction and early warning model to output graded early warning instruction, identifies the systematic risk of a single fault node in a low overall degradation state turnout, and improves the foresight of maintenance decision and the timing accuracy of gantlet resource allocation.

Description

Technical Field

[0001] This invention relates to the field of rail transit facility monitoring technology, and in particular to a method and system for comprehensive monitoring of all parameters of turnouts and AI prediction and early warning. Background Technology

[0002] Railway turnouts are core mechanical components in rail transit networks that enable train switching, and their operational status directly affects train safety. Existing turnout condition monitoring schemes mainly rely on switching completion signals and locking contact feedback, lacking continuous tracking of the force characteristics of each mechanical stage of the drive process and the contact state of elastic components, making it difficult to identify potential degradation trends of components before faults appear.

[0003] Meanwhile, ambient temperature fluctuations continuously interfere with commutation resistance. Existing methods lack effective means to separate temperature effects from mechanical wear, leading to an overlap between seasonal resistance changes and the actual degradation process, thus limiting the reliability of assessment conclusions. At the prediction and early warning level, existing systems generally trigger alarms with fixed thresholds, failing to combine degradation rate evolution and maintenance window resource constraints to estimate remaining usable life in advance. Maintenance decisions lack forward-looking timing basis, easily causing a mismatch between planned maintenance windows and actual failure risks. Summary of the Invention

[0004] This invention discloses a method and system for comprehensive monitoring and AI prediction and early warning of all parameters of a turnout. By collecting switch machine current and close-fitting detection data, a multi-stage degradation characteristic sequence is constructed. The wear status of components is determined by combining cross-temperature zone degradation comparison and health assessment. The remaining lifespan range is inferred based on the decay rate evolution. The graded early warning instructions are output by combining the fault triggering sequence and component health index through the prediction and early warning model, providing complete decision support for turnout operation and maintenance covering perception, assessment, prediction and early warning.

[0005] The first aspect of this invention proposes a method for comprehensive monitoring and AI-based prediction and early warning of all parameters of a turnout, comprising: Collect switch machine current data and contact detection data. Based on the switch machine current data, collect the inflection point time and duration of the current slope of each conversion stage to generate segment division records. Based on the contact detection data, track the rebound amount after contact is in place to determine the contact elasticity margin. Based on the segment division record, the segment current slope batch drift direction and rate are identified to form a segment decay sequence. The elastic critical record is obtained by using the closely attached elastic margin statistical margin batch continuously decreasing node. A temperature decoupling index is generated by comparing the segment degradation differences across temperature zones in the segment degradation sequence. The wear accumulation level is determined by identifying the degradation amount of thermally stable segments based on the temperature decoupling index. The component health index is obtained by calculating the score of each fault mode-driven segment based on the wear accumulation level. Based on the elastic critical record, analyze the slope change of the margin decay rate to establish a decay acceleration record. Use the decay acceleration record to calculate the remaining lifetime interval from the acceleration start point to the predicted critical time. Extract the remaining trigger time of each typical fault from the remaining lifetime interval and output the fault trigger timing sequence. Based on the fault triggering timing, fault nodes with insufficient triggering time to the nearest sunroof are identified to form an urgent warning set. The severity of each fault and the maintenance response time limit are inferred by the predictive warning model by combining the urgent warning set and the component health index, and graded warning instructions are output.

[0006] The second aspect of this invention proposes a turnout full-parameter integrated monitoring and AI prediction and early warning system, comprising: The data acquisition module is used to collect switch machine current data and contact detection data. Based on the switch machine current data, it collects the inflection point time and duration of the current slope of each conversion stage to generate segment division records. Based on the contact detection data, it tracks the rebound amount after the contact is in place to determine the contact elasticity margin. The segment analysis module is used to identify the segment current slope batch drift direction and rate based on the segment division record to form a segment decay sequence, and to obtain the elastic critical record by using the closely attached elastic margin statistical margin batch continuously decreasing node. The health assessment module is used to compare the segment degradation differences across temperature zones in the segment degradation sequence to generate a temperature decoupling index, identify the degradation amount of thermally stable segments based on the temperature decoupling index to determine the wear accumulation level, and calculate the score of each fault mode driven segment based on the wear accumulation level to obtain the component health index. The lifetime prediction module is used to establish a decay acceleration record based on the slope change of the margin decay rate according to the elastic critical record, and to calculate the remaining lifetime interval from the acceleration start point to the predicted critical time using the decay acceleration record. The remaining trigger time of each typical fault is extracted from the remaining lifetime interval and the fault trigger timing is output. The early warning output module is used to identify fault nodes whose trigger duration is insufficient for the nearest window based on the fault triggering sequence to form an urgent early warning set. The module combines the urgent early warning set with the component health index and uses a predictive early warning model to infer the severity of each fault and the maintenance response time limit, and outputs graded early warning instructions.

[0007] The beneficial effects of this invention are reflected in the following points: 1. By identifying the slope inflection point and collecting the segment duration of the switch machine current data, the switching process is decomposed into a sequence of independently traceable mechanical stage characteristics. The continuous identification of the drift direction and rate between batches allows for the quantitative recording of the degradation state at each stage. The gradual increase trend of the batch average of the close-fitting rebound amount and the continuous decrease node of the elastic margin are tracked synchronously, providing a complete basis for the elastic degradation history process for subsequent life estimation. 2. By comparing the segment degradation amounts of low-temperature and high-temperature batches, a temperature decoupling index is constructed. The degradation amount of the thermally stable segment after removing the temperature component directly maps the mechanical wear process, eliminating the interference of seasonal resistance fluctuations on degradation assessment. The component health index is jointly calculated by the wear accumulation level and the segment score driven by each fault mode, so that the degree of degradation of each mechanical stage of the turnout can be quantitatively expressed in the fault mode dimension. 3. Determine the starting point for the transition from pseudo-steady to accelerated based on the slope change of the margin decay rate, calculate the remaining life interval by combining the acceleration rate parameter, and output the trigger confidence interval and median trigger batch for each typical fault mode; form an urgent early warning set by combining the fault trigger timing and the track window time interval, and infer the fault severity and maintenance response time limit by combining the prediction and early warning model, and adjust the severity of turnouts with prominent overall degradation status, so that the early warning result reflects the overall risk of multi-mode degradation rather than the overrun of a single parameter. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a method for integrated monitoring of all parameters of a turnout and AI-based prediction and early warning, according to the present invention.

[0009] Figure 2 This is a structural block diagram of a turnout full-parameter integrated monitoring and AI prediction and early warning system according to the present invention. Detailed Implementation

[0010] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0011] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0012] The technical solutions of the embodiments of this application will be described below.

[0013] like Figure 1 As shown, this embodiment of the invention provides a method for comprehensive monitoring and AI prediction and early warning of all parameters of a turnout, including the following steps S110-S150: Step S110: Collect switch machine current data and contact detection data. Based on the switch machine current data, collect the inflection point time and duration of the current slope of each conversion stage to generate segment division records. Based on the contact detection data, track the rebound amount after contact is in place to determine the contact elasticity margin.

[0014] Specifically, switch machine current data and contact detection data are collected. The power consumption of the drive motor varies with the phased travel of the switch rail. If the data acquisition starting point is not aligned with the moment the reversing command is issued, the slope of the starting segment of the switch machine current data will be truncated. Occasional premature or delayed triggering of data acquisition due to contact jitter in the field control cabinet is addressed by using the moment when the current value first surges above the threshold from the stationary baseline as the valid starting point. The initial offset caused by jitter is corrected by adding a time correction label to the switch machine current data. Contact detection data is continuously output at high frequency by the displacement sensor after the contact signal is triggered, fully covering the entire rebound attenuation process. Fluctuations in the sensor power supply cause spikes in the contact detection data frames, resulting in sudden drops in amplitude followed by resets. These spike frames are filled with the average of adjacent normal frames, and interpolation labels are added to the filled frames. When these interpolation labels appear in multiple batches, it indicates an intermittent fault in the power supply circuit. The sampling frequency of the two data streams is set as a configuration parameter based on the shortest transition phase duration of this type of switch machine, ensuring that the number of switch machine current data sampling points in each transition phase is not less than the lower limit required for slope estimation. When the contact detection data is transmitted back via the wireless link, packet loss is detected according to the frame sequence number. Batches with consecutive frame loss exceeding the threshold are marked with link anomalies and excluded from rebound extraction. Both the switch machine current data and contact detection data acquisition terminals retain batch-level timestamps. If the timestamp accuracy is lower than the sampling interval, the correlation between the preceding and following batches of reversing will be misaligned. During the track maintenance operation window, the track bed is subject to external force intervention, and the switch machine current data is mixed with non-reversing load components. At the same time, the steady-state value of the contact detection data deviates from the normal range due to track vibration. Both sets of data are uniformly marked with operation interference during the maintenance window period. The marked batches are not included in segment division and rebound extraction.

[0015] Based on the current data of the switch machine, the inflection points and durations of the current slopes at each transition stage are collected to generate segment division records. During turnout reversal, the switch rail sequentially undergoes three different stress processes: disengaging from locking constraints, overcoming lateral resistance of the track bed, and tightening the locking hook at the end. The mechanical transition between these three stages leaves identifiable slope bends on the drive motor load. A sliding differential window is used to scan the slope sequence of the switch machine current data. When the slope direction reverses within the window, it is determined as the inflection point. The duration of each segment at the inflection point constitutes the duration element of the segment division record. Grid harmonics superimpose on the switch machine current data during periods of low power factor. Brief slope abrupt changes can easily be misjudged as inflection points. The shortest allowable duration between adjacent inflection points is used as a filtering threshold. Candidate inflection points below the threshold are merged into the preceding segment to prevent a single mechanical stage from being segmented into multiple false records by harmonics. The number of segments for each reversal is fixed at three segments corresponding to the three stress processes. When more than two candidate inflection points are obtained from the scan, the two inflection points with the largest slope change are used as the boundary to prevent redundant segments caused by local jitter. Changes in ambient temperature cause the internal resistance of the motor windings to drift, resulting in a systematic shift in the overall amplitude of the switch machine current data between batches. The inflection point determination is based on the slope rather than the absolute amplitude, and is not sensitive to amplitude drift. When the duration of a certain segment in the segment division record monotonically increases within consecutive batches, it corresponds to the continuous accumulation of mechanical resistance in that stage. Batches with missing durations are filled with the average duration of the most recent consecutive normal batches, and the filled batches are marked with an anomaly in the segment division record. During peak operating periods, the frequency of turnout reversals increases significantly. The segment division record is arranged based on the batch number rather than the absolute time. Sections with consecutive batch numbers but abnormally compressed time intervals are simultaneously marked with high-density reversal labels to prevent the high-frequency reversal batches from being out of order due to clock accuracy errors.

[0016] In some embodiments, determining the fit elasticity margin by tracking the rebound amount after the fit is in place based on the fit detection data includes: using the fit detection data to statistically analyze the steady-state value of the displacement sensor after the fit is in place to generate a steady-state displacement record; analyzing the difference between the instantaneous value and the steady-state value in the steady-state displacement record to obtain the single rebound amount; identifying the gradually increasing range of the batch average rebound amount based on the single rebound amount to form a rebound growth trend; and determining the fit elasticity margin by aggregating the average difference in growth rate under various load conditions based on the rebound growth trend.

[0017] Steady-state displacement records are generated by statistically analyzing the steady-state values ​​of displacement sensors after the contact patch detection data is applied. After contact patching is completed, the elastic component continuously applies pressure to the switch rail. The sensor readings gradually converge from the peak value at the moment of contact patching through elastic decay. Steady state is considered to have been reached only when the difference between consecutive frames in the contact patch detection data is lower than the convergence threshold and the number of consecutive frames exceeds the shortest stable frame number. If the convergence threshold is too tight, slight vibrations will delay steady-state recognition; if the threshold is too loose, the system will lock prematurely before elastic decay is complete. Steady-state displacement records are annotated for batches with both types of deviations. The steady-state value is represented by the average of several consecutive frame readings after steady-state determination, rather than taking the instantaneous value of a single frame. Multi-frame averaging suppresses the interference of random noise on the steady-state benchmark. If the number of frames participating in the averaging is lower than a set lower limit, the steady-state displacement record is annotated with insufficient steady-state samples. During periods of high-density operation, trains pass immediately after reversing direction. Track vibration continuously accumulates as disturbance readings during the close-fit steady-state phase. The frame-by-frame difference in the close-fit detection data fails to determine the steady-state state due to excessive vibration. Instead, the inter-window difference of the short window mean is used to replace the frame-by-frame difference. The window length is written into the configuration parameters based on the longest duration of train vibration on the line. A larger window length enhances the suppression of single-frame disturbances but correspondingly reduces the accuracy of identifying the start of the steady-state state. Seasonal changes in ambient temperature cause systematic shifts in sensor zero drift between batches. Before collecting the steady-state values ​​of each batch of close-fit detection data, the compensation amount is calculated using the sensor temperature coefficient and the current ambient temperature. After temperature drift compensation, the long-term outward shift trend of the steady-state displacement record between batches truly reflects the continuous change of the close-fit position with elastic degradation. When the rate of outward shift of the steady-state value monotonically increases within multiple consecutive batches, elastic degradation has entered an accelerated phase. An accelerated degradation label is added to the steady-state displacement record for such batches.

[0018] The difference between the instantaneous value and the steady-state value in the steady-state displacement record is used to obtain the single rebound amount. The displacement reading at the moment of maximum compression is extracted from the steady-state displacement record as the instantaneous value d_peak, and the reading at the new equilibrium position of the switch rail after elastic decay convergence is extracted as the steady-state value d_s. The single rebound amount δ_r is calculated as δ_r = d_peak - d_s, with both d_peak and d_s being the readings after temperature drift compensation. δ_r quantifies the additional rebound amplitude released due to stiffness decay during a single action of the elastic component. The single rebound amount monotonically increases with the continuous deterioration of the elastic component across batches. When the same batch contains multiple reversals, the single rebound amount is taken as the median of the rebound values ​​within the batch, rather than the mean, as the median is more robust to batches experiencing sporadic impacts. When the rebound amplitude is lower than the sensor resolution, this action is not included in the single rebound amount statistics to avoid quantization noise being treated as actual rebound accumulation. When the sampling rate is insufficient, the instantaneous frame of the steady-state displacement record already contains some rebound amount, so the single rebound amount is systematically low. The instantaneous value of the arrival time is estimated by extrapolating the maximum value of the arrival time using the reading slope of several adjacent frames. The extrapolation result is marked with low precision in the single rebound amount, and the weight of low precision batches is reduced accordingly in the incremental interval identification stage. When the switch rail is jammed or the connecting parts are loose, the elastic compression amount exceeds the normal range, and the single rebound amount of the corresponding batch is abnormally high while the batches before and after are normal. The batches with a change in the amount between batches exceeding three times the global standard deviation are identified as isolated abnormal batches, and isolated abnormal batches are not included in the incremental interval judgment. In the steady-state displacement record, the batches whose steady-state value drops sharply due to the maintenance or replacement of elastic components correspond to a single rebound amount that drops sharply to below the historical level. The batches with sharp drops are replaced by linear interpolation of the normal batches before and after. There are no artificial breakpoints in the time sequence of single rebound amounts at the component replacement node, and the degradation pattern of slowly rising between batches is continuously displayed.

[0019] The increasing interval of the average rebound amount across batches is identified based on the single rebound amount, forming a rebound growth trend. The fatigue degradation process of elastic components is slow, and the inter-batch elasticity loss is usually much smaller than the batch disturbance of the single rebound amount caused by track bed resistance and commutation condition fluctuations. A fixed batch window average is used instead of batch-by-batch values. Intervals where the difference between adjacent window averages is continuously positive are identified as increasing intervals. Smoothing the window average separates the degradation direction signal from noise. The number of batches with continuously positive differences between adjacent window averages must exceed a set number of confirmed batches for the increasing interval to be considered valid. The number of confirmed batches is set based on the shortest identifiable continuous batch degradation for this type of component, preventing the brief upward movement of several batches from being mistakenly included in the rebound growth trend. Isolated abnormal batches are excluded during window average calculation to prevent isolated high values ​​from artificially inflating the window average and shifting the increasing interval boundary forward. Seasonal temperature variations cause the average rebound amount to be systematically higher in summer batches. The residual sequence obtained by subtracting the historical baseline average for the same season from the current batch window average is used to determine the gradual increase interval. After eliminating the seasonal effect, the boundary of the gradual increase interval in the rebound growth trend more accurately reflects the true start and end time of elasticity degradation. When the historical baseline batches are insufficient, the seasonal slope fitted by the existing batches of the current year is used instead, and the fitting results are labeled with low samples. If there is a brief period of stagnation or a slight decline between multiple gradual increase intervals, adjacent intervals with a spacing smaller than the set fusion window are merged into a single trend segment. When the slope of the gradual increase interval in the later stage of the rebound growth trend is significantly higher than that in the early stage, it indicates that the elasticity decay has entered an accelerated degradation stage. In the time series of single rebound amounts, segments where the change direction between batches frequently reverses without a continuous direction, and the window average remains stable in the segment, the rebound growth trend does not include such stable segments in the gradual increase interval, to avoid misjudging short-term disorderly disturbances as a stage interruption of the degradation process.

[0020] The close-fitting elastic margin is determined by aggregating the average difference in growth rates under various load conditions based on the rebound growth trend. The rebound growth rate of the same elastic component exhibits a systematic difference under heavy-load, high-frequency and light-load, low-frequency conditions. The same incremental interval in the rebound growth trend typically covers batches under multiple load conditions. For each incremental interval, the high- and low-load graded growth rates β_{j,h} and β_{j,l} are extracted, and the difference in growth rates Δβ_j = β_{j,h} - β_{j,l} is calculated. When the average Δβ_j in later intervals is generally smaller than that in earlier intervals, it indicates that the load tolerance of the elastic component continuously decreases with each batch. During high-density heavy-load periods, the elastic component experiences greater compressive force in a single action, and the slope of the rebound growth trend in the incremental interval of this period is higher than that in light-load periods. The smaller the difference in growth rates between the two types of conditions, the more significant the trend of degradation under light-load conditions catching up with heavy-load conditions. When the difference narrows rapidly over multiple consecutive batches, the component's load adaptation reserve is approaching the lower limit of its design tolerance. When the number of batches for a certain load range is less than the statistical lower limit, the growth rate of that range is estimated by linear interpolation of adjacent ranges. When the interpolation result is used in the calculation, a low-sample label is added. If the proportion of low-sample labels exceeds the set proportion, a low-confidence label is added to the close-fitting elastic margin. The load range boundary is set in the initialization stage based on the quantile of the historical reversing frequency and load distribution of the line, so that the same load range corresponds to comparable working conditions in different time periods. The close-fitting elastic margin is normalized with the growth rate difference level in the early stage of new component commissioning as the baseline. After normalization, the close-fitting elastic margin of different models of elastic components has cross-component comparability. The average value of the close-fitting elastic margin M_b for each batch is obtained by weighting the difference in growth rate Δβ_j between each incremental interval by the batch number n_j and then normalizing it to the baseline. That is, M_b=Σ(n_j×Δβ_j) / (Σn_j×Δβ_0), where n_j is the batch number of the j-th incremental interval and Δβ_0 is the baseline of the difference in growth rate in the early stage of the new component's operation. The incremental interval with more batches has a higher contribution weight. The lower the close-fitting elastic margin value, the closer the rebound growth rate is under each load condition. The elastic component can no longer maintain the elasticity reserve higher than that under heavy load conditions under light load conditions. When the close-fitting elastic margin drops to the set critical value, it indicates that the elastic component has entered the high-risk degradation range.

[0021] Step S120: Based on the segment division record, identify the segment current slope batch drift direction and rate to form a segment decay sequence, and use the close-fitting elastic margin statistical margin batch continuously decreasing node to obtain the elastic critical record.

[0022] Specifically, based on the segment division records, the drift direction and rate of segment current slope batches are identified to form a segment decay sequence. The current slope k_nb of each segment is calculated by dividing the segment current envelope range R_nb by the segment duration T_nb given in the segment division records, i.e., k_nb = R_nb / T_nb. When a mid-range trough appears in the waveform within a segment, the envelope slope is used instead of the single range to prevent waveform morphology differences from causing incomparability in slope estimation between batches for the same mechanical stage. During the spring snowmelt season, the track bed moisture content rises sharply, and the lateral resistance of the switch rail increases abnormally during this period. In the segment division records, the slope of the advancing segment is isolatedly higher in the corresponding batch while the preceding and following batches are normal. Isolated high values ​​are replaced by the average slope of adjacent normal batches. The original values ​​are retained for manual verification and are not included in the trend calculation of drift direction and rate. In the segment division record, the slope of each segment is smoothed using a fixed window according to the batch time sequence. The sign sequence of the difference between the mean values ​​of adjacent windows determines the drift direction. The mean of the difference between the mean values ​​of consecutive windows in the same direction is used as the drift rate. When the drift direction is continuously negative, the resistance of the corresponding mechanical stage continues to accumulate. The higher the drift rate, the faster the degradation progresses. The reversal frequency is concentrated during peak periods. The drift rate in the segment decay sequence is normalized and corrected for the reversal frequency during high-density periods. After correction, the drift rates of each period are comparable across time periods. If the drift rate of a certain segment in the segment decay sequence drops sharply and then rebounds, it usually corresponds to the component having undergone temporary maintenance, but the effect was not sustained. The segment decay sequence adds a maintenance intervention mark to such batches. When maintenance intervention marks appear densely, it indicates that temporary treatment cannot stop the degradation process, and the component needs to be included in the planned replacement.

[0023] In some embodiments, obtaining the elastic critical record using the continuously decreasing nodes of the closely attached elastic margin statistical margin batches includes: establishing a difference sign record from the positive and negative sign sequence of the margin difference between consecutive batches of the closely attached elastic margin statistical margin; identifying the number of consecutive batches with negative signs exceeding the threshold in the difference sign record to determine the number of batches with continuous decline; identifying the first batch exceeding the threshold based on the number of batches with continuous decline to form a critical starting batch; and collecting the batch margin mean and the standard deviation within the batch according to the critical starting batch to obtain the elastic critical record.

[0024] A difference sign record is established by statistically analyzing the positive and negative sign sequences of the margin differences between consecutive batches of closely fitted elastic margin. The difference in the mean margin between adjacent batches is Δ_b = M_b - M_{b-1}, where M_b and M_{b-1} are the mean values ​​of the closely fitted elastic margin of the b-th batch and the preceding batch, respectively. After assigning positive or negative signs to Δ_b, the basic sequence of difference sign records is formed by arranging the batches accordingly. This compresses the margin time series from the amplitude dimension to the direction dimension, allowing the trend of elastic degradation direction to be directly separated from the random fluctuations between batches. A negative difference value means that the elasticity reserve of the elastic component in this batch is further narrowed compared to the previous batch. The continuous negative sign segment is a temporal confirmation that the degradation direction remains unchanged. The difference sign record, with the signs of each batch arranged in chronological order, constitutes the finest-grained process record of the degradation direction. When the absolute value of Δ_b in adjacent batches of the close-fitting elastic margin is extremely small but the sign is negative, the decrease in margin is within the measurement accuracy range, and the sign itself is not statistically significant. The difference sign record adds a low-amplitude label to the negative signs of such batches. The low-amplitude negative signs are treated with half weight in the subsequent counting of consecutive batches exceeding the threshold, preventing the accumulation of slight degradation that is long within the measurement accuracy boundary from being misjudged as a continuous degradation trend. After the overhaul and replacement of elastic components, the margin slightly decreases during the initial break-in stage of the new components. The continuous negative sign sequence during the break-in period is highly similar in morphology to the normal degradation period. The concentrated maintenance period at the end of the year is often a high-incidence period for component replacement. After the replacement, the consecutive negative signs of the first few batches are very easy to be misjudged by the identification system as a continuation of degradation. The difference sign record uses maintenance intervention labeling to identify the start and end batches of the break-in period. Negative signs within the break-in period are not included in the continuous decrease judgment, preventing normal break-in from being misjudged as the start signal of a degradation trend.

[0025] The number of consecutive negative symbols exceeding the threshold in the difference symbol record is used to determine the number of continuous descent batches. After removing low-confidence and maintenance intervention labels, the length of the consecutive negative symbol segment in the difference symbol record is the candidate number of continuous descent batches. Candidate segments exceeding the set minimum continuous batch threshold are considered valid descent segments. The threshold is written into the configuration parameters during the initialization phase based on the shortest continuous batch of historical degradation of the line's elastic components. During the winter and spring seasons, passenger flow on the line rebounds intensively, and the frequency of reversing increases significantly in a short period. The density of consecutive negative symbols in the difference symbol record is high during this period. The actual time span corresponding to the same number of continuous descent batches is much shorter during high-frequency periods than during regular periods. The calibration of the continuous batch threshold needs to be combined with the setting of differentiated parameters based on the operational density. In the difference sign record, when a brief positive sign is followed by multiple negative signs, the consecutive negative sign statistics are not interrupted when the number of consecutive positive sign batches is lower than the set fusion tolerance. The fusion tolerance is written into the configuration parameters based on the longest consecutive batch of short-term rebound margin in historical batches. When the fusion tolerance is too large, the actual slight rebound is also included in the continuous decline segment. The impact on early warning sensitivity and false alarm rate needs to be jointly calibrated with maintenance personnel. When there are multiple independent over-threshold segments in the difference sign record, the number of consecutive decline batches for each segment is determined separately. The appearance of a stable period between segments usually corresponds to maintenance intervention or a temporary improvement in operating conditions. The longer the number of consecutive stable period batches, the more lasting the suppression effect of the intervention on elastic degradation. The comparison of the number of consecutive decline batches between segments reflects the difference in the evolution rate of each stage of the degradation process.

[0026] The first batch exceeding the threshold in the continuous decline segment is identified as the critical starting batch. The first batch in the threshold-exceeding section to meet the continuous batch threshold is the candidate critical starting batch. The reason for choosing the first batch rather than the middle or last batch of the segment is that the first batch is the earliest verifiable starting point for the elastic degradation to transition from fluctuation to continuous unidirectional decline. Locating the critical state with the earliest starting point allows for the maximum lead time for subsequent maintenance. When the threshold is small, the critical starting batch is earlier, triggering an early warning in the early stage of degradation, which is beneficial for early intervention but has a higher false alarm rate. When the threshold is large, the identification is conservative. The choice between these two types needs to be determined jointly by the maintenance personnel based on the response margin for different fault types. After the candidate critical starting batch is confirmed, the subsequent margin trend is back-checked to verify whether it truly maintains a continuous decline. If the candidate batch barely reaches the continuous batch threshold and the subsequent trend turns upward again within the observation window, the candidate is withdrawn and re-identified in the next threshold-exceeding segment to prevent batches that just touched the threshold and then reversed immediately from being mistakenly identified as the critical starting point. If there are low-confidence markers before the first batch in the continuous decline batch number exceeding the threshold, the critical starting batch identification is postponed to the first high-confidence batch to ensure that the time series positioning is based on reliable data. For some lines, due to short construction time and insufficient historical data accumulation, the continuous batch threshold lacks sufficient statistical support. For the critical starting batch identification of such lines, the shortest continuous batch number of degradation on the industry average for this type of component is used as a substitute threshold. The source of the substitute threshold is marked in the critical starting batch record. After sufficient historical data is accumulated for this line, the measured value is used to replace it and a retrospective evaluation is carried out. In the multiple continuous decline batch number exceeding the threshold, the batch corresponding to the first segment forms the primary critical starting batch, and the starting batches of subsequent segments are marked as secondary critical batches. The shorter the batch interval between the two types of starting batches, the faster the elastic degradation enters a secondary acceleration after the initial continuous decline, and the overall remaining life of the elastic component is thus significantly compressed.

[0027] The elastic critical record is obtained by aggregating the mean margin and within-batch standard deviation of the critical starting batch. The mean margin of this batch directly reflects the remaining elastic reserve level when the elastic component enters the continuous degradation stage. The lower the mean, the more severe the available margin is at the critical starting batch. The within-batch standard deviation reflects the consistency of the elastic component response during multiple commutations within the same batch. A large standard deviation indicates that the elastic response within the critical starting batch has become significantly dispersed, and the elastic decay has shown non-uniformity within the same batch. When the commutation frequency corresponding to the critical starting batch is high, the batch contains more commutations, and the statistical basis of the margin mean is more sufficient. Conversely, when the number of commutations is very low, the elastic critical record adds a low sample label to the mean of this batch. When there are multiple sets of parallel elastic components, the mean margin and standard deviation of each component are independently aggregated at the critical starting batch. The critical state parameters of each component are displayed in parallel in the elastic critical record. The overall degradation risk of a multi-component system with a high concentration at the critical moment is significantly higher than that of a system where the critical moments of each component are dispersed. The elastic critical record adds a high synchronous degradation label to concentrated critical batches. When the elastic critical record synchronously collects the percentage of the difference between high and low load margins in the critical starting batch, an additional margin uniformization warning label is added when the percentage of the difference is lower than the set lower limit. For batches where the average margin value is low and the margin uniformization warning label meet the triggering conditions at the same time, the elastic critical record is distinguished from the batches triggered by a single condition by a double warning label. The decay rate threshold corresponding to the double triggering situation must be tightened.

[0028] Step S130: Compare the segment degradation differences across temperature zones to generate a temperature decoupling index. Identify the degradation amount of thermally stable segments based on the temperature decoupling index to determine the wear accumulation level. Calculate the score of each fault mode-driven segment based on the wear accumulation level to obtain the component health index.

[0029] In some embodiments, the step of generating a temperature decoupling index by comparing the segmental degradation differences across temperature zones in the segmental degradation sequence includes: extracting the average segment degradation amount from low-temperature batches and high-temperature batches of the segmental degradation sequence to obtain the average temperature zone degradation; using the average temperature zone degradation to screen outlier segments by temperature difference and determine the segmental temperature difference response; identifying a set of segments with response amounts exceeding the average of all segments based on the segmental temperature difference response and constructing a set of thermosensitive segments; and generating a temperature decoupling index by aggregating the proportion of non-thermally sensitive segment degradation to the total degradation amount according to the set of thermosensitive segments.

[0030] The mean value of segment degradation for each temperature zone is obtained by extracting the mean values ​​of segment degradation for low-temperature and high-temperature batches in the segmental degradation sequence. The segment degradation is calculated as the drift rate of each segment in the segmental degradation sequence. The low-temperature group includes batches with ambient temperatures below the lower quartile, and the high-temperature group includes batches with ambient temperatures above the upper quartile. Batches in the intermediate temperature zone are not included in the mean calculation. The comparison between the extreme value groups at both ends maximizes the manifestation of the temperature difference effect in the mean value of temperature zone degradation. The exclusion of intermediate transitional temperature zones prevents gradient ambiguity from masking the systematic differences between high and low temperatures. In northern lines, the duration of low temperatures in winter is long, and the high-temperature batches are relatively concentrated. The low-temperature group usually has more batches than the high-temperature group in the segmental degradation sequence. When the number of batches in the two groups is significantly different, the mean value of the temperature zone with a smaller sample size is more significantly affected by a few extreme batches. For temperature zones with a sample size below a set lower limit, a low-sample label is added to the mean value of the temperature zone. The mean value of the low-sample temperature zone is supplemented by the cumulative mean value of batches in the same temperature zone in history. The source of the supplement is labeled in the mean value record of the temperature zone degradation. Batches with lower supplementation accuracy have reduced weight in the subsequent outlier screening stage. In the segment decay sequence, the average degradation amount of each segment in the low-temperature group and the average degradation amount in the high-temperature group constitute the temperature zone degradation mean pair for that segment. Segments with a systematically higher low-temperature mean than the high-temperature mean indicate a strong resistance response to temperature changes, and the degradation amount contains a large temperature component. Segments with an absolute difference between the two means close to zero are dominated by mechanical wear, and segments with a high-temperature mean higher than the low-temperature mean are marked with additional directional anomalies. In mixed lines crossing tunnel and ground sections, the micro-environmental temperature fluctuation amplitude of each segment of the same turnout is different. The difference between the high and low temperature means in the ground-exposed segments is systematically higher than that in the tunnel segments. In the initialization stage, the segment decay sequence establishes an independent benchmark temperature difference range for each segment according to the installation location type, so that the temperature zone degradation mean comparison is comparable under different installation environments. When a batch of environmental temperature records is missing, meteorological station records are used to supplement it, and the degree of accuracy limitation when the distance between the station and the site is far is marked simultaneously.

[0031] For example, the step of using the temperature zone degradation mean to screen outlier segments and determine the segment temperature difference response includes: calculating the absolute value of the difference between the high and low temperature degradation mean values ​​of each segment to obtain the absolute value of the segment temperature difference; using the absolute value of the segment temperature difference to statistically analyze the mean and standard deviation of the absolute value distribution of the difference across all segments to obtain the difference distribution characteristics; screening outlier segments based on the difference distribution characteristics to construct a set of outlier segments; and aggregating the absolute value of the segment temperature difference according to the set of outlier segments to determine the segment temperature difference response.

[0032] The absolute value of the segmental temperature difference is obtained by calculating the absolute value of the difference between the high and low temperature degradation means of each segment based on the temperature zone degradation mean. The absolute value is obtained by subtracting the high temperature group mean from the low temperature group mean of each segment. Differences in positive and negative directions have already been handled separately during the directional anomaly labeling stage of the temperature zone degradation mean. After taking the absolute value, the absolute value of the segmental temperature difference only reflects the magnitude of the difference without distinguishing direction, ensuring the directional independence of subsequent outlier statistics at the calculation source. For segments where both the high and low temperature group means are small in the temperature zone degradation mean, the small absolute value of the segmental temperature difference is not due to low temperature sensitivity, but rather because the degradation baseline of each temperature zone in that segment is inherently low. The small difference is a magnitude effect rather than a reflection of weak response. For these segments, an additional low baseline label is added to the absolute value of the segmental temperature difference. Segments with low baselines are grouped separately during the difference distribution statistics to prevent low-base segments from suppressing the distribution mean and underestimating the outlier degree of truly heat-sensitive segments. When the sample size difference between two temperature zones in a certain segment is significant, the accuracy of mean estimation is limited. The mean of a small sample temperature zone is greatly affected by a few extreme batches. The absolute value of the segment temperature difference is marked with a limited accuracy label for the segment with limited accuracy. During the outlier screening stage, the judgment threshold for this segment is correspondingly relaxed to prevent estimation errors caused by sample size asymmetry from being judged as thermosensitive outlier signals. When the absolute value of the segment temperature difference is right-skewed across all segments, it usually indicates that a few highly thermosensitive segments dominate the temperature component of the current batch. When high-value segments are concentrated in a certain mechanical stage, the mean of temperature zone degradation is marked with a thermosensitive concentration label. This label corroborates the stage-concentrated thermosensitive label of subsequent thermosensitive segment sets.

[0033] The difference distribution characteristics are obtained by statistically analyzing the mean and standard deviation of the absolute difference distribution across all segments using the absolute value of segment temperature differences. Segments with limited accuracy are excluded from the statistical sample. Segments with low baseline values ​​are grouped separately and also excluded from the main distribution statistics. This ensures that the mean and standard deviation of the difference distribution characteristics are calculated based on a set of segments with comparable magnitudes. The exclusion operation prevents segments with large estimation errors from increasing the standard deviation and raising the subsequent outlier detection threshold. A large standard deviation of the difference distribution characteristics indicates significant differentiation in the temperature sensitivity of segments in the current batch. Some segments show extremely strong degradation response to temperature, while others show extremely weak response. This polarization usually corresponds to significant differences in the installation position or stress characteristics of each segment of the turnout within this batch. When the historical mean of the standard deviation of the absolute difference distribution characteristics of segment temperature differences is consistently higher than that of all turnouts of the same type along the entire line, it suggests a long-term abnormality in the thermal sensitivity distribution among the turnout segments, requiring verification of whether there is a systematic deviation in the sensor installation. When a sudden drop in the frequency of reversal in a batch leads to insufficient samples of segment degradation, the overall absolute value of segment temperature difference is small. Consequently, the mean of the difference distribution characteristic systematically shifts downward. This downward shift in the mean lowers the absolute value of the outlier detection threshold, causing a large number of mildly heat-sensitive segments to cross the threshold and enter the outlier set. The difference distribution characteristic adds a low-frequency batch label to batches with abnormally low reversal frequencies, and the outlier detection of low-frequency batches continues from the difference distribution characteristic of the preceding normal batches. A sudden increase in the mean of the absolute value of segment temperature difference in a batch usually corresponds to that batch experiencing extreme temperature difference weather rather than a synchronous increase in the actual heat sensitivity of each segment. The difference distribution characteristic adds a global temperature wave label to batches with a sudden increase in the mean, and outlier screening for global temperature wave batches is suspended, continuing from the difference distribution characteristic of the previous batch to prevent misjudgment of a large number of heat-sensitive segments caused by a single extreme weather event.

[0034] A set of outlier temperature difference segments was constructed by screening outlier segments based on the difference distribution characteristics and standard deviation. Outlier determination was based on the boundary that the absolute value of the segment temperature difference exceeded the mean of the difference distribution characteristics plus one standard deviation. This one standard deviation boundary statistically corresponds to approximately the 84th percentile, ensuring that the set of outlier temperature difference segments covers a minority of segments with significantly higher thermosensitivity. An overly wide boundary would include mildly thermosensitive segments, weakening the representativeness of degradation in the non-thermally sensitive group; an overly narrow boundary would introduce more temperature noise into the non-thermally sensitive group. For batches with an overall high mean of the difference distribution characteristics, the absolute value of the one standard deviation boundary was correspondingly high. Some mildly thermosensitive segments, whose absolute values ​​did not exceed the raised boundary, fell into the non-thermally sensitive group. For such batches, the boundary of the set of outlier temperature difference segments was lowered to the mean plus 0.5 standard deviations to prevent systematic omission of mildly thermosensitive segments from high-mean batches. Batch-specific outlier segment sets with global temperature wave annotations will pause updates, continuing from the previous batch with a continuation annotation. A warning will be triggered when the number of continued batches exceeds a set limit. When outlier segment sets continuously expand between batches, a warning will appear indicating a sustained increase in the range of thermistor segments. If the expansion rate exceeds a set limit, a warning for intensified temperature interference will be triggered. The warning limit is written into the configuration parameters based on the historical maximum thermistor expansion rate of the line. Continuous contraction of outlier segment sets typically corresponds to a recent narrowing of the temperature difference or maintenance intervention that homogenizes local resistance. Contraction batches are marked with a set reduction annotation. Historical thermistor records for outlier segment sets are continuously retained in the segment-level archive and are not deleted due to set contraction.

[0035] The segmental temperature difference response is determined by aggregating the absolute values ​​of segmental temperature differences from a set of outlier segments. The sum of the absolute values ​​of segmental temperature differences in each segment within the outlier set forms the basis for calculating the segmental temperature difference response. The batch mean is used to represent the response amplitude of each segment, quantifying the average response intensity of the thermosensitive segment to temperature changes. When the absolute value of the segmental temperature difference fluctuates significantly across batches within the same segment, the reliability of single-batch values ​​is low due to the influence of extreme reversal conditions. Replacing single-batch values ​​with batch means provides a more robust response estimate. Fluctuations in sufficiently sampled batches originate from actual dynamic degradation changes, while fluctuations in insufficiently sampled batches stem from mean estimation errors. The segmental temperature difference response is calculated using a weighted average with batch reversal frequency as the weight; batches with higher frequencies have a larger weight and smaller estimation errors. When outlier segments in a ensemble change in or out of adjacent batches, the resulting change in the segment's temperature difference response causes a jump across batches. If the jump direction is consistent with the direction of the member change, it is considered a normal jump caused by the ensemble effect. If the jump amplitude exceeds the historical normal range, an abnormal jump label is added, indicating that the change in the batch's response is not entirely derived from the actual change in thermal sensitivity. If the absolute value of a segment's temperature difference has been stable for a long time across historical batches and then suddenly jumps, it usually corresponds to a loose installation connection causing sensor reading drift rather than an increase in the actual temperature response. For such segments with a jump in temperature difference response, linear interpolation is used to replace the normal batches before and after. The original values ​​are retained in the segment file for manual verification. Drift replacement labels are added to the interpolated batches. Batches with a high proportion of drift replacement labels in the segment's temperature difference response are treated as low-confidence batches.

[0036] A set of thermosensitive segments is constructed by identifying segments whose response values ​​exceed the overall segment mean based on segment temperature difference response. The overall segment mean is calculated as the average of all segment responses in the segment temperature difference response. Using the mean as a threshold provides adaptability. Since the overall level of segment temperature difference response varies significantly across different lines and seasons, a fixed absolute threshold cannot accommodate the benchmark differences between lines and seasons. Using the mean as the boundary ensures that the coverage of the thermosensitive segment set maintains a relatively consistent statistical meaning across different lines and seasons. Segments whose response values ​​are exactly near the overall segment mean have ambiguous boundaries. The thermosensitive segment set treats these critical segments with half weight. The degradation amount is included in both the total thermosensitive degradation and the total non-thermal degradation amount separately when calculating the temperature decoupling index, preventing non-physical abrupt changes in the temperature decoupling index caused by changes in boundary node assignments. For lines with long-term uneven distribution of high and low temperature batches, the average value of the temperature difference response of a segment is systematically shifted due to batch skew. For such periods, the thermally sensitive segment set replaces the original average with the equally weighted sample average of each temperature zone batch in the judgment, eliminating the systematic bias of sample imbalance on the threshold. When the segments covered in the thermally sensitive segment set are highly concentrated in the mechanical stage dimension, it indicates that the overall resistance characteristics of that mechanical stage are sensitive to temperature fluctuations. The greater the difference in thermal density compared to adjacent mechanical stages, the more prominent the temperature dependence of the force mechanism in that stage. The thermally sensitive segment set adds a stage-concentrated thermal label for such cases. When consecutive batches of labels appear, it is necessary to check whether there is a systematic thermal drift deviation in the sensor installation of that stage to avoid sensor thermal drift being misjudged as true component thermal sensitivity.

[0037] The temperature decoupling index is generated by grouping the proportion of degradation of non-thermal segments to the total degradation based on the set of thermally sensitive segments. Non-thermally sensitive segments are those segments not included in the set of thermally sensitive segments. These segments exhibit higher stability in degradation slope between batches, and the changes in degradation as batches progress more directly reflect the mechanical wear process. Let the degradation amount of segment i be q_i, which is its segment drift rate. The temperature decoupling index ρ = Σ(i∉thermal)q_i / Σq_i, where the numerator is the sum of the degradation amounts of non-thermal segments and the denominator is the total degradation amount of all segments. The more segments the thermal segment set covers, the lower the value of ρ. In the extreme case, when all segments are included in the thermal segment set, the non-thermal degradation amount is zero, the temperature decoupling index drops to zero, and the degradation signal of the entire batch is completely masked by the temperature effect. At this time, the temperature full masking label is triggered, and the wear accumulation level continues the rating of the normal batch with the most recent temperature decoupling index. The continued batch label is retained in the wear rating record to distinguish it from the normal rating batch. When the set of thermally sensitive segments contains critical half-weight segments, the degradation of boundary segments in both the numerator and denominator of the temperature decoupling index is included at half the value, maintaining a consistent caliber. When the members of the thermally sensitive segment set change between batches, the entry and exit of segments cause the temperature decoupling index to jump across batches. Batches with jump amplitudes exceeding a set threshold are marked with an additional set change label, indicating that the jump is due to member changes rather than actual wear and structural changes. A continuous decrease in the temperature decoupling index between batches indicates a continuously expanding proportion of thermally sensitive segments and a deepening of the obstruction of turnout degradation signals batch by batch. When the continuous decrease exceeds a set number of batches, a warning of intensified temperature interference is triggered. The batch number threshold is written into the configuration parameters based on the typical number of decreasing batches before the occurrence of full temperature obstruction on similar historical lines. If the threshold is too small, the warning will be triggered too early; if the threshold is too large, the obstruction problem will only be identified when it is already severe. The trade-off between these two types needs to be calibrated in conjunction with the operation and maintenance response cost.

[0038] The wear accumulation level is determined by identifying the degradation amount of thermally stable segments based on the temperature decoupling index. Thermally stable segments are non-thermally sensitive segments not included in the thermally sensitive segment set. After excluding the temperature component, the slope of their degradation amount changes continuously between batches, mapping the wear accumulation rate of each mechanical stage. A wear rate slope k_s is fitted to the mean degradation amount of thermally stable segments with a fixed batch window. k_s is combined with the cumulative degradation amount C_acc for joint classification. Both reflect the current degradation rate and the historical total consumption, respectively. Classification based solely on rate may underestimate the risk of highly worn components, and classification based solely on cumulative amount may underestimate the risk of rapid deterioration in the near future. The wear accumulation level is divided into four levels: low, medium, high, and critical. The boundary of each level is written into the configuration parameters based on the typical k_s and C_acc parameters at the end of the design life of this type of component. A batch with a low temperature decoupling index has an insufficient number of thermally stable segments, weakening the statistical basis of k_s. The wear accumulation level adds a low reliability label to such batches, continuing the reliability of previous batches. When the temperature decoupling index remains consistently low for more than a set number of batches, a wear rating interruption warning is triggered, indicating that the degradation signal is continuously masked by temperature. During the regular track maintenance window, the turnout switching resistance temporarily decreases, and the degradation of thermally stable segments shows a temporary decline in the first batch after maintenance. k_s temporarily turns negative in this batch, and the wear accumulation level marks this batch with a maintenance intervention label, excluding the declining batches from the slope estimation. If the k_s recovery slope after maintenance is significantly higher than before maintenance, it indicates limited maintenance effectiveness and that the wear dynamic mechanism has not been effectively intervened. The wear accumulation level adds a label indicating insufficient maintenance effectiveness. When the dispersion of thermally stable segment degradation in the wear accumulation level is high, a non-uniform wear label is added. Non-uniform wear means that the force distribution of each segment is unbalanced, and the degradation rate of local segments exceeds the overall average level. Accelerated degradation of a single segment may trigger local failure before the overall lifespan is exhausted. The wear accumulation level rating is adjusted to upgrade the level of the segment with the highest degradation in the non-uniform wear batch by one level to prevent the overall average from masking the risk of extreme local degradation.

[0039] The component health index is obtained by calculating the score of each fault mode-driven segment based on the wear accumulation level. The dependence of each fault mode on the degradation amount of different segments varies significantly. Mechanical jamming faults are preceded by accelerated wear of the propulsion segment, while elastic failure faults are mainly driven by the cumulative degradation of the locking segment. The wear accumulation level is summarized according to the fault mode to which each segment belongs. The score of each fault mode-driven segment S_i is calculated as the weighted average of the wear accumulation level of the corresponding segment. The weights are preset based on the contribution rate of each segment degradation to the historical triggering of the fault. The component health index H is calculated as H=1-Σ(w_i×S_i) / S_max, where S_max is the reference value for the full load score of this type of component, w_i is the weight of each fault mode, and the closer H is to zero, the closer the total degradation of each fault mode-driven segment is to the design limit. When the component health index is lower than the set warning threshold, a health warning is triggered. The threshold is written into the configuration parameters based on the lower quartile of the health index distribution of the last batch of components of the same type before the historical fault. Batches with non-uniform wear labels in the wear accumulation level have peak scores for each segment involved in the corresponding fault mode that are much higher than the mean. When calculating the component health index, the weighted sum of the peak score and the mean of the fault mode score is taken as S_i to prevent local extreme degradation from being systematically underestimated in mean smoothing. When a power distribution fault causes a stop in the commutation process, the propulsion segment is subjected to an instantaneous impact force exceeding the normal range. The corresponding batch has an abnormally high score for the drive segment, but it is unrelated to wear accumulation. The wear accumulation level adds an impact interference label to such batches. When calculating the component health index, the S_i of this batch is replaced by interpolation from the normal batches before and after it to prevent impact interference batches from lowering the component health index and triggering false alarms. When the component health index is below the threshold for multiple consecutive batches and the rate of decline continues to accelerate, an accelerated deterioration label is added. The accelerated deterioration label is used to improve the prediction update frequency of the remaining life range in the subsequent lifespan estimation stage. When the scores S_i of each fault mode are synchronously high in consecutive batches, it indicates the concurrent risk of multiple types of faults. The component health index adds a multi-mode concurrency label to batches with synchronously high scores for multiple modes. This label triggers the highest priority response.

[0040] Step S140: Based on the change in slope of the margin decay rate in the elastic critical record, establish a decay acceleration record. Use the decay acceleration record to calculate the remaining lifetime interval from the acceleration start point to the predicted critical time. Extract the remaining trigger time of each typical fault from the remaining lifetime interval and output the fault trigger timing sequence.

[0041] In some embodiments, establishing a depletion acceleration record based on the slope change of the margin depletion rate according to the elastic critical record includes: extracting the time series of batch margin difference from the elastic critical record to form a margin change time series; performing a jump correction window mean analysis on the margin change time series to determine the rate trend sequence; identifying the starting batch of pseudo-steady deceleration turning into accelerated growth based on the rate trend sequence to form the slope acceleration starting point; and collecting the mean difference of the rates before and after the acceleration starting point to establish a depletion acceleration record.

[0042] The margin change time series is constructed by extracting the batch margin difference time series from the elastic critical record. The margin mean difference between adjacent batches is ΔE_b = E_b - E_{b-1}, where E_b and E_{b-1} are the elastic critical record margin mean values ​​of the b-th batch and the previous batch, respectively. ΔE_b is arranged batch by batch to form the margin change time series, transforming the margin absolute value time series into a batch-by-batch change sequence. The direction signal of the degradation rate is directly separated from the margin amplitude change. The maintenance intervention batch has a sudden increase in margin due to component replacement, and the corresponding ΔE_b is an abnormally large positive value. If this batch is not removed, the margin change time series will form a false signal of direction reversal at the replacement node, and the boundary judgment between the false stable segment and the acceleration segment will drift. The margin change time series adds a maintenance label to the maintenance intervention batch. After component replacement, the direction of ΔE_b in the break-in segment is random and the amplitude is small. The break-in segment is also added with a break-in label. The maintenance label and the break-in label batches are not included in the rate slope estimation to prevent normal break-in fluctuations from interfering with the degradation trend judgment. During peak holiday operating periods, reversing batches accumulate intensively in a short period, and the batch numbering rate is significantly higher than during normal periods. The margin change time series uses the batch number rather than the absolute time as the horizontal axis. In high-density periods, frequency normalization correction is added when estimating the rate to ensure the comparability of the margin change time series across different operating density periods. The margin change time series is also labeled with temperature batches according to the batch ambient temperature for subsequent seasonal fluctuation identification. In the elastic critical record, the margin mean of low-sample labeled batches is estimated based on a small number of reversing times, and ΔE_b is greatly affected by estimation error. The margin change time series adds low-precision labels to low-sample batches, and the weight of low-precision batches is reduced in the subsequent window mean calculation stage. When the acquisition is interrupted in the margin change time series, resulting in discontinuous batch numbers, the time series on both sides of the gap are processed independently. It is not allowed to calculate the difference across the gap. The segment boundary is labeled with an acquisition gap in the margin change time series, and the sequence after the gap is re-established as a baseline to participate in accelerated judgment.

[0043] For example, the step of performing abrupt correction window mean analysis on the margin change time series to determine the rate trend sequence includes: generating a window mean sequence based on the average batch difference within each fixed window according to the margin change time series statistics; analyzing the continuous change pattern of the difference between adjacent window means in the window mean sequence to obtain the window mean change rate; separating the smoothed and suppressed abrupt residuals based on the window mean change rate to obtain the abrupt residual sequence; and correcting the window mean sequence according to the abrupt residual sequence to determine the rate trend sequence.

[0044] A window mean sequence is generated based on the average batch difference within each fixed window according to the time series statistics of margin changes. Using several consecutive batches with consecutive batch numbers as fixed window units, the mean ΔE_b of each batch within the window is calculated by sliding the window. Each point in the window mean sequence corresponds to the average rate of margin consumption within a sliding window. The sequence smooths out the interference of single-batch disturbances on rate estimation. For lines with significant differences in operating density during morning and evening peak hours, the actual time length covered by the fixed batch window during peak hours is much shorter than that during nighttime. The rate time scale reflected by the window mean sequence in the peak dense batch segment is much shorter than that in the trough segment. When the window width is too narrow, the acceleration start point identification is sensitive but the false alarm rate is high. When the width is too wide, the response to the real acceleration signal is lagging. The choice between these two types needs to be jointly calibrated based on the maintenance personnel's tolerance for false alarms and missed alarms. The minimum number of batches that typically last for the historical degradation acceleration segment of this component model is used as the reference lower limit for the window width. In margin variation time series, the ΔE_b error of low-precision labeled batches is relatively large. Directly including them in the window mean calculation will cause uncontrollable shifts in the mean within the window. The window mean sequence adds low-precision window labels to windows containing low-precision batches. The reliability of the window mean sequence for time periods with concentrated low-precision window labels is limited. The results of the previous normal window are continued and the number of continued batches is labeled. In margin variation time series, when the acquisition interruption gap width is less than the window width, it is filled by interpolation of adjacent batches. When the gap width exceeds the window width, the window mean sequences on both sides of the gap are segmented independently, and discontinuous labels are added at the gap to prevent the mean of the cross-gap window from masking the rate difference on both sides of the gap. The rate estimation of the window mean sequence returns to normal in the first complete window after acquisition recovery. The recovered batch is labeled with a restart label to distinguish it from the continued batches before and after the gap.

[0045] The window mean change rate is obtained by analyzing the continuous variation of the difference between adjacent window means in the window mean sequence. The difference between adjacent window means reflects the change in the margin consumption rate between batches. Extracted from the first difference of the window mean sequence, a sustained negative value indicates an accelerating consumption rate, while a sustained positive value indicates a slowing consumption rate. This is a direct signal source for identifying the transition from pseudo-steady state to acceleration. The temperature rise in late winter and early spring causes changes in the track bed expansion coefficient. The switching resistance of the switch rail first increases and then decreases within the warming batch. During this period, the window mean sequence forms a periodic positive and negative alternating window mean change rate, the shape of which is highly similar to the continuous same-direction segment during degradation acceleration. Seasonal alternation can be identified by autocorrelation analysis and temperature batch labeling of the margin change time series, preventing misjudgment of the acceleration start triggered by seasonal fluctuations. The window mean sequence itself contains residual low-frequency fluctuations after smoothing. Therefore, the window mean change rate contains random positive and negative alternations caused by low-frequency fluctuations. Consecutive batches with the same sign have trend identification value, while a change in direction in a single batch is not used as a basis for determining the rate turning point. In the window mean sequence, the window mean estimation error of the low-precision window-labeled batch is relatively large. The reliability of the corresponding window mean change rate is limited due to the influence of error propagation. The low-precision propagation batch has a reduced weight in the slope acceleration start identification stage. The window mean change rate usually changes from random fluctuations close to zero to a continuous negative value near the elastic degradation acceleration start batch. The starting batch of the continuous negative value segment provides a time series reference for subsequent accurate positioning of the slope acceleration start point. The accuracy gain of the deviation correction process between the reference batch and the finally confirmed slope acceleration start point batch is quantified.

[0046] The sudden jump residual sequence is obtained by separating the smoothed and suppressed sudden jump residuals based on the window mean change rate. After smoothing by the window mean, a residual signal is still left in the window mean change rate of a single batch sudden jump. The sudden jump residual sequence is constructed by subtracting the mean of the window corresponding to each batch from the original ΔE_b of each batch. Batches with larger absolute residual values ​​are sudden jump candidates, and three times the standard deviation of the full-time residual is used as the identification threshold. When a train with wheelset abrasion passes through a turnout, it generates an instantaneous impact force. The elastic components of the corresponding reversing batch are abnormally compressed, and the margin change time sequence shows isolated high values ​​in this batch. The sudden jump residual sequence accurately locates such isolated impact batches, marks them as sudden jump candidates, and removes them from the window mean sample. This ensures that the subsequent smoothing of the window mean sequence accurately reflects the degradation trend before and after the impact, preventing a single impact batch from forming a false acceleration signal in the window mean change rate. When consecutive jump candidates appear, the residual mean within the continuous segment is used to replace the residuals of each batch to prevent multiple batches of consecutive high values ​​from being split into several isolated jumps and losing continuous information. When the length of a continuous jump segment exceeds a set threshold, it is not treated as a jump but is retained as a high-value segment to prevent the real elastic acceleration segment from being misjudged as a jump residual sequence and removed from the window mean sequence. If a dense jump occurs in the jump residual sequence within a continuously negative window mean rate of change, the jump residual sequence adds a source verification label to the dense jump segment. When compared with the label of the elastic critical record, if the time sequence matches, it tends to be caused by mechanical anomalies; if they do not match, it tends to be caused by sensor drift. If a jump is missed in the jump residual sequence, it causes the window mean sequence to shift. If a jump is misidentified, the window mean sequence will have insufficient sample size after the normal batches are removed. Both types of errors are quantitatively evaluated by comparing the difference between the window mean before and after correction during the rate trend sequence correction stage.

[0047] The rate trend sequence is determined by correcting the window mean sequence based on the jump residual sequence. The correction operation removes the jump batches marked in the jump residual sequence from the calculation samples of each affected window in the window mean sequence. The window mean is recalculated using the remaining batches. After jump removal, the window mean sequence is no longer affected by extreme single-batch values, resulting in a higher signal-to-noise ratio for the accelerated signal. The corrected sequence is the rate trend sequence. Maintenance personnel use the rate trend sequence during routine inspections to determine whether to trigger a maintenance window application. If the sequence morphology before and after correction differs significantly, but the acceleration starting batches are consistent, it indicates that jump removal has limited impact on trend direction judgment. If the starting batch deviation exceeds a set threshold, the correction result directly determines whether the time window for maintenance resource allocation is sufficient. The rate trend sequence adds a correction sensitivity label for such situations, indicating that maintenance personnel's current lifespan prediction is highly dependent on the jump removal strategy and should be manually verified in conjunction with the original sequence. In the abrupt batch removal of the abrupt residual sequence, the number of effective samples in some windows may fall below the set minimum sample limit. Window with insufficient samples is replaced by interpolation of the mean of adjacent normal windows. The interpolated windows are marked with sample completion annotations in the rate trend sequence. When the completion annotations appear densely, it indicates that the overall reliability of the margin change time series is low. The difference in trend slope before and after correction quantifies the contribution of the abrupt residual sequence correction to the manifestation of the acceleration signal. A large contribution indicates that abrupt removal is a key prerequisite for the clear presentation of the acceleration signal, while a small contribution indicates that the acceleration signal is also clear in the uncorrected window mean sequence. The robustness of the rate trend sequence shape does not depend on the accuracy of the abrupt removal strategy. In both cases, the contribution magnitude is marked when the decay acceleration record is established for operation and maintenance personnel to assess the necessity of correction.

[0048] Based on the rate trend sequence, the initial batches that transition from pseudo-stable deceleration to accelerated growth are identified, forming the starting points of slope acceleration. In the rate trend sequence, pseudo-stable segments are characterized by multiple batches where the absolute value of the window mean remains at a low level and the direction is random and discontinuous. Accelerated segments are characterized by a continuously increasing absolute value of the window mean and a continuously negative direction. The boundary between these two types of segments is the temporal positioning target for the starting points of slope acceleration. During the spring maintenance window, maintenance personnel perform local lubrication repairs on elastic components. After the repair, the rate trend sequence shows a brief deceleration in several batches. However, the local repair fails to eliminate the accumulation effect of microcracks within the material, and the rate trend sequence accelerates again after the brief deceleration. The initial batches that accelerate again after the repair form new starting points of slope acceleration. Both the old and new starting points are retained in the attenuation acceleration record. The smaller the rate difference before and after the repair, the more limited the effect of the local repair on extending the remaining life. The maintenance department can quantify the actual life extension effect of the local repair based on the rate difference between the two slope acceleration starting points. A slope acceleration start point is confirmed only when the proportion of consecutive batches with window averages exceeding the absolute value of the pseudo-stable phase average exceeds a set confirmation threshold. This prevents short-term fluctuations in the rate trend sequence from being mistaken for acceleration initiation. When multiple transition candidates exist in the rate trend sequence, the transition batch with the largest difference in absolute value average is designated as the primary slope acceleration start point, and the remaining candidates are marked as secondary acceleration start points. The shorter the interval between the primary and secondary slope acceleration start point batches, the faster the elastic component enters secondary acceleration after the initial acceleration. When the difference between the secondary slope acceleration start point batch and the primary start point exceeds the upper limit of the typical interval for similar historical components, multi-stage acceleration labels are added, indicating that the degradation process has obvious staged breaks rather than continuous gradual progression.

[0049] The decay acceleration record is established by collecting the rate mean difference before and after the acceleration point based on the slope acceleration point. Before the slope acceleration point, the mean absolute rate v_pre is taken within a fixed batch window, and after the slope acceleration point, the mean absolute rate v_post is taken within a window of the same batch size. The same batch size is used before and after the acceleration point to ensure that the statistical basis of the two means is comparable and to avoid the bias caused by the difference in sample size due to inconsistent window lengths. The rate difference Δv = v_post - v_pre quantifies the rate jump from pseudo-steady to acceleration in elastic degradation. The decay acceleration record uses Δv and the linear slope r_acc of the acceleration segment to jointly drive the subsequent lifetime estimation. Δv describes the rate step height from pseudo-steady to acceleration, and r_acc describes the continuous upward trend of the rate after entering the acceleration segment. The two correspond to the jump magnitude and acceleration persistence, respectively, and are jointly input into the downstream uniform acceleration lifetime model. The short intervals between adjacent maintenance windows on high-speed mainlines result in a tight window for allocating maintenance resources. An underestimation of Δv will lead to a delay in maintenance intervention, causing components to fail functionally before the planned maintenance window. If the batches covered in the window before the slope acceleration start point contain maintenance interventions or low-precision annotations, the corresponding batches are excluded in the v_pre calculation to prevent the maintenance effect from raising the baseline rate and underestimating Δv. If the window after the slope acceleration start point contains a large number of low-precision batches, the v_post estimation error will be too large. The decay acceleration record adds low-precision annotations to the back window for this situation. r_acc is obtained by linear regression slope of the absolute value of acceleration segment rate over time and the regression goodness of fit is recorded simultaneously. A low goodness of fit indicates that the acceleration segment rate has not yet shown a stable linear growth and the degradation pattern is still evolving. The deterioration acceleration record adds an undetermined pattern label to the acceleration segment with low goodness of fit and drives the lifetime estimation with a conservative estimate. The higher the r_acc, the more the degradation consumes the remaining elastic reserves at an increasingly faster rate. The deterioration acceleration record containing the secondary acceleration start point outputs Δv and r_acc for each segment. The parameters of the most recent acceleration segment are used as the main basis for the remaining lifetime estimation. When the difference between the parameters of the most recent acceleration segment and the historical acceleration segment is too large, an ultra-historical intensity label is added and the subsequent lifetime prediction is driven by a conservative estimate.

[0050] The remaining lifetime interval is obtained by estimating the time from the acceleration start point to the predicted critical point using the decay acceleration record. The rate difference Δv in the decay acceleration record and the linear slope r_acc of the acceleration segment together determine the predicted trajectory of margin consumption. The upper limit of the linear estimate T_max = E_curr / v_post is obtained by dividing the current average margin E_curr by the post-acceleration rate v_post. E_curr is taken from the average margin of the current batch elastic critical record, reflecting the most optimistic number of remaining batches to maintain the current rate level. The acceleration effect makes the actual number of remaining batches lower than the linear estimate. The lower limit of correction T_min is obtained by solving E_curr = v_post × T + 0.5 × r_acc × T^2 using the uniform acceleration model, where T is the number of remaining batches from the acceleration start point to the predicted critical point. The remaining lifetime interval is represented by [T_min, T_max]. The overhaul cycle for power lines is typically locked in advance on an annual basis. If the upper limit of the remaining lifespan interval is lower than the number of batches corresponding to the next overhaul window, it means that the resilient components are likely to trigger a failure before the planned window. The maintenance department needs to determine whether to apply for a temporary window based on the lower limit of the remaining lifespan interval. The narrower the interval, the clearer the decision-making basis. When the interval is consistently wide, it is recommended to increase the frequency of recent high-precision batch collection to narrow the prediction range. When the morphology of the decay acceleration record is undetermined or the subsequent window is low-precision, the remaining lifespan interval is inherited with low confidence and is treated as a conservative estimate. For batches with super-historical intensity annotations, T_min is already optimistic. The lower limit of the remaining lifespan interval for such batches is tightened to the joint prediction value of v_post and r_acc. The lower limit of the interval for resilient critical records with double warning annotations is also compressed using the tightened decay rate threshold. For decay acceleration records containing secondary acceleration start points, the remaining lifespan interval is calculated based on the parameters of the most recent acceleration segment. The greater the difference between the two segments, the more the lower limit of the remaining lifespan interval is compressed towards the estimated value of the main segment. The prediction update frequency for the remaining lifespan range is automatically increased to per batch during high-frequency operating periods, and triggered according to the batch number threshold during low-frequency periods. The prediction update frequency for batches with accelerated deterioration labels attached to the component health index is also increased to per batch.

[0051] In some embodiments, the step of extracting the remaining trigger duration of each typical fault from the remaining lifetime interval and outputting the fault trigger timing sequence includes: using the remaining lifetime interval to statistically analyze the boundaries of each fault mode interval to obtain the fault lifetime boundary; analyzing the evolution slope characteristics of the interval endpoints of the fault lifetime boundary to determine the degradation trajectory type; identifying the remaining batch confidence range corresponding to the current stage based on the degradation trajectory type to form a trigger confidence interval; and outputting the fault trigger timing sequence by aggregating the trigger batch and the two boundary times according to the trigger confidence interval.

[0052] The fault life boundary is obtained by statistically analyzing the boundaries of each fault mode within the remaining lifespan interval. Different fault modes exhibit significantly different sensitivity thresholds to elasticity loss. Mechanical jamming faults can be triggered when elasticity reserves drop to a moderate level, while contact failure faults typically only appear when the margin is critically exhausted. The upper and lower limits of the remaining lifespan interval generate corresponding remaining batch boundaries at the trigger sensitivity thresholds of different fault modes. The fault life boundary is then summed from the boundaries of all fault modes. When a jamming fault occurs in a turnout, the elasticity margin may not yet be nearly fully exhausted, but the elasticity loss during the advance phase is sufficient to trigger jamming. If the lifespan is predicted solely based on the overall margin depletion time, this type of fault cannot be identified in advance. The fault life boundary sets independent trigger thresholds for each fault mode. Jamming faults generate a fault life boundary in the middle of the remaining lifespan interval, ensuring that the warning signal appears earlier than the overall lifespan depletion batch. When the lower limit batch of the remaining lifespan interval is small, the trigger threshold for some fault modes may already be within the current batch. The fault life boundary adds an immediate risk label to such cases, which triggers the highest priority response during the urgent warning set generation phase. The trigger sensitivity thresholds for each fault mode are written into the configuration parameters based on the historical fault batches of this type of component during the initialization phase. Fault modes with insufficient historical sample size are replaced by statistical data of similar components in the industry. The uncertainty of the replacement threshold is reflected by the expansion coefficient in the width of the upper and lower limit range of the fault life boundary. When the remaining life interval is marked with low confidence, the fault life boundary of each mode boundary inherits the mark synchronously and is handled with conservative estimation to prevent low confidence life estimation from driving overly aggressive warnings.

[0053] The degradation trajectory type is determined by analyzing the evolution slope characteristics of the endpoints of the fault lifespan boundary. As the batch progresses, both ends of the fault lifespan boundary contract towards the current batch at different rates. The contraction slope of the lower endpoint reflects the compression rate of the most pessimistic remaining lifespan, while the contraction slope of the upper endpoint reflects the compression rate of the most optimistic remaining lifespan. The relative relationship between the slopes at both ends reveals whether the remaining lifespan interval is narrowing or widening between batches. During the Spring Festival transportation period, the frequency of turnout reversals on intercity high-speed railways far exceeds the annual average for several weeks. The upper endpoint of the fault lifespan boundary contracts rapidly during the peak period, with a contraction slope significantly higher than the lower endpoint, indicating a continuous narrowing of the interval. The degradation trajectory type is identified as accelerated narrowing during the peak period batches. After the peak period ends and the operating density returns to normal, the slopes at both ends tend to converge, and the degradation trajectory type switches to uniform compression. The batches switching the trajectory are marked with changes in operating density to prevent trajectory changes caused by the peak period from being misjudged as structural changes in the degradation mechanism. When both ends of the slope are high and the difference is stable, the degradation trajectory type is identified as uniform compression. When the slope of the lower endpoint is consistently higher than that of the upper endpoint, the interval widens, usually corresponding to increased inter-batch fluctuations in the degradation rate. The degradation trajectory type is identified as diffusion, and the source of uncertainty needs to be reassessed in conjunction with recent batch estimates of r_acc in the decay acceleration record. If a certain failure mode boundary in the failure lifetime boundary remains stationary and does not shrink within consecutive batches, the degradation trajectory type adds a low trigger probability label to such stationary boundaries, and the trigger confidence interval is conservatively treated with a wide interval. The slopes at both ends of batches with added super-historical intensity labels are usually high. The degradation trajectory type tends to be identified as accelerated narrowing in such batches, and is treated with more conservative prediction.

[0054] Based on the degradation trajectory type, the remaining batch confidence range corresponding to the current stage is identified to form a trigger confidence interval. The remaining batch confidence range is determined according to the degradation trajectory type by category. For uniform compression trajectories, the confidence range is directly formed by the current upper and lower limits of the fault life boundary; for accelerating narrowing trajectories, the confidence range is mainly based on the current lower limit, with the upper limit contracting downwards to the set compression ratio of the interval width; for diffusion trajectories, due to increased uncertainty, the trigger confidence interval is further expanded by a set number of batches on top of the upper limit to prevent the actual triggered batches from exceeding the right side of the predicted interval when degradation rate fluctuations increase. When the operation and maintenance dispatch center receives a trigger confidence interval update notification, the interval width directly determines the urgency of the maintenance request. A wide interval allows for overall scheduling within the next planned window, while a narrow interval may require a temporary window application on the same day or the next day. When the degradation trajectory type remains accelerating narrowing, an emergency warning upgrade is automatically triggered when the lower limit between batches in the trigger confidence interval approaches the nearest window time. The matching degree between the interval window and the window time directly determines whether maintenance resources can be delivered in a timely manner. For degraded trajectory types containing low-trigger-probability markers, the trigger confidence interval is expanded with a wide range instead of being compressed to the lower limit, avoiding false alarms caused by artificially high prediction accuracy for low-probability patterns. When a degraded trajectory type changes between consecutive batches, the trigger confidence interval is re-estimated using the new trajectory type parameters of the batch after the change. The historical records of the trigger confidence interval before the change are retained, and the changed batch is marked with a trajectory change label. The difference in the contraction rate of the lower limit of the trigger confidence interval between batches before and after the change quantifies the actual impact of the trajectory type change on the urgency of life prediction. The larger the difference, the stronger the impact of the change event on maintenance scheduling, requiring immediate notification to the operation and maintenance scheduling center to update the resource allocation plan.

[0055] The fault trigger sequence is output based on the midpoint trigger batch and the two boundary times within the trigger confidence interval. The midpoint trigger batch is determined by the midpoint of the upper and lower limits of the trigger confidence interval. The left boundary batch represents the most pessimistic trigger time, and the right boundary batch represents the most optimistic trigger time. These three points together constitute the trigger time description of each fault mode in the fault trigger sequence. The midpoint batch is the main reference node for maintenance scheduling, and the two boundary batches represent the range constraints of lead time and delay time. Annual track maintenance windows are limited. When the midpoint trigger batches of multiple fault modes in the fault trigger sequence fall within the same batch segment before the same track maintenance window, the maintenance team needs to handle multiple types of faults simultaneously within a single track maintenance window. The fault trigger sequence arranges each mode according to the midpoint batch sequence. The shorter the interval between trigger batches, the more concentrated the risk time window for multiple types of faults. Maintenance time planning needs to be verified in conjunction with the track maintenance window duration quota when the fault trigger sequence is output. If the quota is exceeded, a track maintenance window extension application or fault priority reordering is triggered. When the median trigger batch shifts significantly forward between consecutive batches, a rapidly shifting marker is added to the trigger confidence interval. This rapidly shifting marker triggers an upgrade of the warning level for this mode during the tiered warning phase. The fault trigger sequence outputs updated three time points for rapidly shifting batches with a version label, allowing maintenance personnel to quantify the deviation between historical and latest predictions and assess the impact of sudden changes in degradation rate on existing maintenance plans. For batches with trajectory switching markers added to the trigger confidence interval, the median trigger batch in the fault trigger sequence outputs both before and after the switching. The parallel output of these two types of predictions allows maintenance personnel to intuitively grasp the magnitude of the impact of changes in reversal frequency on the trigger time. When the median batch in the fault trigger sequence exhibits large fluctuations within consecutive batches, it indicates that the corresponding parameters of the decay acceleration record still need further stabilization. The fault trigger sequence adds a predicted fluctuation marker when outputting this type of pattern.

[0056] Step S150: Based on the fault triggering timing, identify the fault nodes with insufficient triggering time of the nearest sunroof to form an urgent warning set. Combine the urgent warning set with the component health index and use the prediction warning model to infer the severity of each fault and the maintenance response time limit to output graded warning instructions.

[0057] In some embodiments, the step of identifying fault nodes with insufficient triggering time for the nearest skylight based on the fault triggering time sequence to form an urgent warning set includes: establishing a fault interval prediction table by calculating the predicted triggering interval for each fault type based on the fault triggering time sequence; determining the skylight gap by comparing the skylight gap with the nearest skylight time distance using the fault interval prediction table; identifying fault types with insufficient time margin based on the skylight gap to form urgent items before the skylight; and constructing an urgent warning set by aggregating the trigger probability of urgent items and the time margin difference based on the urgent items before the skylight.

[0058] A fault interval prediction table is established to calculate the predicted trigger interval for each fault type based on the fault triggering sequence. The predicted trigger interval I_f = B_mid,f - B_curr, where B_mid,f is the median trigger batch number of fault mode f, and B_curr is the current batch number. The smaller I_f is, the closer the fault mode is to the current batch, and the higher the urgency of triggering. For fault modes with added predicted fluctuation annotations in the fault triggering sequence, the median trigger batch varies significantly across batches, and the estimation error of I_f for a single batch is relatively high. For this type of mode, the fault interval prediction table uses the average of the median trigger batches of the most recent batches instead of the median value of a single batch. The mean smoothing makes the fault interval prediction table more robust in the predicted fluctuating batch segment. For fault triggering sequences where the left boundary batch is earlier than the median batch, the most pessimistic trigger time is earlier than the median batch. The fault interval prediction table synchronously records the shortest trigger interval I_f,min calculated based on the left boundary. The larger the difference between I_f and I_f,min, the higher the uncertainty of the trigger time of this mode. Maintenance scheduling uses I_f,min as a conservative lower limit to arrange response preparation. When adjustments to the route operation schedule cause sudden changes in reversing frequency, the actual calendar time corresponding to the same batch difference undergoes a systematic change. The shortened actual time interval for each batch after the adjustment makes the time margin tighter. The fault interval prediction table adds frequency change annotations to the batches affected by the operation schedule adjustment. The actual duration of the corresponding fault trigger interval is recalculated based on the adjusted batch time interval to prevent the use of the conversion factor before the adjustment from causing the fault interval prediction table to overestimate the remaining time. In the fault trigger sequence, the I_f of the batches marked with rapid advancement is calculated based on the latest median batch and a rapid approximation annotation is added. The rapid approximation annotation triggers a more stringent threshold judgment in the subsequent assessment of the gap in the maintenance window. The I_f of each mode in the fault interval prediction table is arranged in ascending order, with the mode with the shortest trigger interval listed first. When multiple modes' I_f are concentrated in similar batch segments, a batch concentration annotation is added.

[0059] The amount of the gap in maintenance coverage is determined by comparing the most recent maintenance window time interval with the fault interval prediction table. The most recent maintenance window time interval W_near is represented by the number of batches corresponding to the start time of the most recent planned maintenance window. The amount of the gap in maintenance coverage G_f is obtained by subtracting the predicted trigger interval I_f of each mode in the fault interval prediction table from W_near. This quantifies the advance gap of each fault prediction trigger relative to the most recent maintenance window. G_f is the time when the fault is expected to trigger before the maintenance window opens. The larger G_f is, the earlier the fault is before the maintenance window time, and the more likely a temporary maintenance window intervention is needed. Maintenance window plans are released after the monthly maintenance meeting but may be adjusted. For lines with a high frequency of historical maintenance window plan adjustments, W_near has greater uncertainty. For such lines, the amount of the gap in maintenance coverage G_f is represented by the planned maintenance window time plus or minus the historical adjustment range. The maintenance department determines whether to trigger a temporary maintenance window application based on the G_f range rather than a single point value. When the lower limit of the range is negative and the upper limit is positive, a critical range label is added to indicate that the fault may still trigger before the maintenance window even if the maintenance window plan is delayed. For patterns with predicted fluctuations in the fault interval prediction table, if the estimated error of I_f is too large, the window gap quantity will be marked as having low gap confidence for such patterns, with G_f,max = W_near - I_f,min as the most unfavorable estimate. When G_f,max is positive, the most pessimistic triggering situation has already fallen before the window, and the window gap quantity will determine the pattern as a high-urgent gap state, driving subsequent urgent project identification to trigger with a lower threshold. For patterns with rapid approximation markings in the fault interval prediction table, the window gap quantity will be evaluated with a stricter time margin threshold in the current batch to prevent the time margin determination of batches with rapidly narrowing trigger intervals from using the same standard as historical normal batches, which would lead to a lag in urgent identification. The window gap quantity is automatically refreshed with each batch update. After the window is opened, W_near is reset to the next planned window time interval. The reset trigger window gap quantity refreshes all patterns G_f with the new window time interval. Patterns whose G_f signs cross zero before and after the reset are removed from the urgent candidate. The window gap quantity adds a window reset marking to the removed patterns, and the operation and maintenance scheduling center re-verifies the current urgent ranking in the window switching batch.

[0060] Based on the insufficient time leeway for identifying fault types with insufficient time leeway during track maintenance window gaps, urgent projects are created before track maintenance windows. Nodes corresponding to fault modes where the gap amount G_f exceeds the set time leeway threshold G_th are added to the candidate list of urgent projects before track maintenance windows. During the initialization phase, G_th is converted into the number of batches based on the shortest preparation time for track maintenance window applications on that line and written into the configuration parameters. For high-density urban rail lines, nighttime track maintenance windows are typically only a few hours long, and the number of faults that can be handled in a single window is strictly limited. When multiple fault modes simultaneously satisfy G_f > G_th, the actual number of projects that can be handled is lower than the total number of candidates. For such lines, urgent projects before track maintenance windows are constrained by track maintenance window time quotas. Among the candidate modes that meet the conditions, priority is given in descending order of G_f. Candidate projects exceeding the time quota are postponed to secondary track maintenance windows. Postponed projects are marked as secondary track maintenance windows in the urgent projects before track maintenance windows and retain their candidate status. After the primary project is completed, the secondary project is automatically upgraded to a primary project for reassessment. For patterns where the confidence level of gaps in the skylight gap quantity is low, urgent projects before the skylight window will use G_f,max instead of G_f to determine the threshold, driving urgency identification with the worst-case scenario to prevent high-risk patterns from being missed due to underestimated time margin. A pattern where G_f in the skylight gap quantity continuously increases across consecutive batches indicates that the lead time between the fault trigger time and the skylight window is expanding. Urgent projects before the skylight window will add dynamic upgrade labels to this pattern, indicating that the urgency is still intensifying and the response time limit needs to be further tightened. When the number of urgent projects before the skylight window increases monotonically across consecutive batches, it usually corresponds to multiple fault modes simultaneously approaching the trigger time. Batches exceeding the set upper limit will be marked with a centralized warning, prompting the operation and maintenance dispatch center to coordinate additional skylight resources in advance. A sudden drop in the number of projects after the skylight window opens is a normal switching effect caused by window reset and is not included in the overall fault risk mitigation assessment.

[0061] An urgent warning set is constructed based on the trigger probability and time margin difference of urgent projects before the maintenance window. For each urgent project before the maintenance window, the trigger probability and time margin difference are collected. The trigger probability P_f is estimated based on the batch coverage ratio from the left boundary of the trigger confidence interval of each mode in the fault triggering sequence to the nearest maintenance window time. The time margin difference D_f = G_f - G_th quantifies the degree to which the current time margin exceeds the threshold. The comprehensive urgency score U_f = P_f × D_f is calculated. The urgent warning set is arranged in descending order of U_f, with the project with the highest U_f receiving the highest priority for maintenance window resource allocation and maintenance response. For urgent projects before the maintenance window with secondary maintenance window labels, P_f is re-estimated based on the secondary maintenance window time, and U_f is usually lower than that of primary urgent projects. These projects are only included in the disposal plan when maintenance window resources are sufficient. After a primary urgent project is completed and removed from the urgent warning set, the secondary projects are upgraded to primary candidates according to their actual urgency in the next batch update. For urgent projects with low reliability in the event window, P_f uses a conservative estimate of the trigger probability instead of a precise calculation. The urgent warning set adds a conservative probability label to such patterns. When the operation and maintenance dispatch center verifies the ranking of the urgent warning set, it prioritizes manual review of the actual urgency of the conservatively labeled projects to prevent the conservative estimate from lowering the comprehensive score of high-risk projects and ranking them lower. When the frequency of line reversals increases several times due to large-scale events, the number of urgent projects before the event window expands rapidly in several batches, and the comprehensive score in the urgent warning set is generally high. After receiving the urgent warning set, the operation and maintenance dispatch center immediately initiates the emergency event window application process and coordinates multiple shifts to handle the situation simultaneously. The urgent warning set locks the final project list and freezes the U_f ranking several batches before the nearest event window opening time. When a component health index drops sharply before the event window, the locked state is forcibly released, and the latest assessment is forcibly updated. The updated batch is marked with an emergency unlock label. After receiving the emergency unlock label, the operation and maintenance dispatch center immediately reviews the resource allocation plan and applies for an extension of the event window if necessary.

[0062] The combined urgent warning set and component health index are used by a predictive warning model to infer the severity of each fault and the maintenance response time limit, outputting tiered warning commands. The predictive warning model uses the features of the urgent warning set and the component health index as joint inputs, and is trained under supervision by historical fault escalation events. It outputs the probability distribution of the severity of each fault node and the optimal maintenance response time limit, and simultaneously outputs the dominant feature annotations driving the current judgment for verification by maintenance personnel. The model is periodically updated with newly confirmed fault events, allowing the severity judgment boundary to adaptively drift with the line's operational characteristics. For turnouts with an overall low component health index, even if there are only low-severity fault nodes in the urgent warning set, the severity inferred by the model still needs to be increased. When multiple fault modes simultaneously approach the critical threshold, the model identifies a single fault as a surface signal of overall degradation rather than an isolated event. Tiered warning commands are divided into three levels: A, B, and C, based on fault severity. The larger the time margin difference, the higher the probability of the model outputting severity. Nodes with a component health index below the model threshold have their severity increased by one level under the same time margin conditions. When the health index of multiple turnout components declines simultaneously during peak operation periods, the predictive early warning model outputs concentrated upgrade risk scores for densely packed B-level projects and prompts the operation and maintenance dispatch center to apply for additional maintenance windows in advance. During high-frequency operation periods, the maintenance response time limit is automatically tightened based on the adjusted batch time interval; multi-mode concurrent labeled nodes are overlaid with collaborative handling labels in the tiered early warning instructions, and spare parts quota suggestions are output simultaneously. Nodes with accelerated deterioration labels on component health indices are forcibly upgraded one level based on model inference. When this occurs concurrently with a rapid shift in the fault triggering sequence, maintenance intervention must be initiated no later than the next day's maintenance window application window.

[0063] To implement the above-described method embodiment, a comprehensive monitoring and AI prediction early warning method for all parameters of a turnout is proposed to achieve the corresponding functions and technical effects. See also... Figure 2 , Figure 2 This paper presents a structural block diagram of a turnout full-parameter integrated monitoring and AI prediction early warning system 200 according to an embodiment of this application, including: Data acquisition module 201 is used to acquire switch machine current data and contact detection data. Based on the switch machine current data, it collects the inflection point time and duration of the current slope of each conversion stage to generate segment division records. Based on the contact detection data, it tracks the rebound amount after contact is in place and determines the contact elasticity margin. The segment analysis module 202 is used to identify the segment current slope batch drift direction and rate based on the segment division record to form a segment decay sequence, and to obtain the elastic critical record by using the closely attached elastic margin statistical margin batch continuously decreasing node. Health assessment module 203 is used to compare the segment degradation differences across temperature zones to generate a temperature decoupling index, identify the degradation amount of thermally stable segments based on the temperature decoupling index to determine the wear accumulation level, and calculate the score of each fault mode driven segment based on the wear accumulation level to obtain the component health index. The lifetime prediction module 204 is used to establish a decay acceleration record based on the slope change of the margin decay rate according to the elastic critical record, calculate the remaining lifetime interval from the acceleration start point to the predicted critical time using the decay acceleration record, and extract the remaining trigger time of each typical fault from the remaining lifetime interval to output the fault trigger timing sequence. The early warning output module 205 is used to identify fault nodes whose trigger duration is insufficient for the nearest window based on the fault triggering sequence to form an urgent early warning set. The urgent early warning set and the component health index are combined and the prediction early warning model is used to infer the severity of each fault and the maintenance response time limit to output a graded early warning command.

[0064] The aforementioned turnout full-parameter integrated monitoring and AI prediction early warning system 200 can implement the turnout full-parameter integrated monitoring and AI prediction early warning method of the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0065] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

Claims

1. A method for comprehensive monitoring and AI-based prediction and early warning of all parameters of a turnout, characterized in that, include: Collect switch machine current data and contact detection data. Based on the switch machine current data, collect the inflection point time and duration of the current slope of each conversion stage to generate segment division records. Based on the contact detection data, track the rebound amount after contact is in place to determine the contact elasticity margin. Based on the segment division record, the segment current slope batch drift direction and rate are identified to form a segment decay sequence. The elastic critical record is obtained by using the closely attached elastic margin statistical margin batch continuously decreasing node. A temperature decoupling index is generated by comparing the segment degradation differences across temperature zones in the segment degradation sequence. The wear accumulation level is determined by identifying the degradation amount of thermally stable segments based on the temperature decoupling index. The component health index is obtained by calculating the score of each fault mode-driven segment based on the wear accumulation level. Based on the elastic critical record, analyze the slope change of the margin decay rate to establish a decay acceleration record. Use the decay acceleration record to calculate the remaining lifetime interval from the acceleration start point to the predicted critical time. Extract the remaining trigger time of each typical fault from the remaining lifetime interval and output the fault trigger timing sequence. Based on the fault triggering timing, fault nodes with insufficient triggering time to the nearest sunroof are identified to form an urgent warning set. The severity of each fault and the maintenance response time limit are inferred by the predictive warning model by combining the urgent warning set and the component health index, and graded warning instructions are output.

2. The method according to claim 1, characterized in that, The step of determining the adhesive elasticity margin by tracking the rebound amount after the adhesive is in place based on the adhesive detection data includes: The steady-state displacement record is generated by statistically analyzing the steady-state values ​​of the displacement sensor after the close-fitting detection data is in place. The difference between the instantaneous value and the steady-state value of the recorded steady-state displacement is used to obtain the single rebound amount. Based on the single rebound amount, the gradually increasing range of the batch average rebound amount is identified to form a rebound growth trend; The close-fitting elasticity margin is determined by aggregating the average difference in growth rates under various load conditions based on the described rebound growth trend.

3. The method according to claim 1, characterized in that, The process of obtaining the elastic critical record by utilizing the continuously decreasing nodes of the close-fitting elastic margin statistical margin batch includes: Establish a difference sign record from the positive and negative sign sequence of the margin difference between consecutive batches of the close-fitting elastic margin; The number of consecutive batches with negative signs exceeding the threshold is used to determine the number of batches with a sustained decline in the difference sign record; Based on the number of batches continuously decreasing, the first batch exceeding the threshold is identified, forming a critical initial batch; Based on the critical starting batch, the mean margin of the batch and the standard deviation within the batch are used to obtain the elastic critical record.

4. The method according to claim 1, characterized in that, The step of generating a temperature decoupling index by comparing the segmental degradation differences across temperature zones in the segmental degradation sequence includes: The average segment degradation amount of the segment degradation sequence in the low temperature batch and the high temperature batch is extracted to obtain the average degradation amount in the temperature zone. The temperature difference outlier segments are screened using the temperature zone degradation mean to determine the segment temperature difference response; Based on the segment temperature difference response, a set of segments with response values ​​exceeding the average value of all segments is identified, and a set of thermosensitive segments is constructed. The temperature decoupling index is generated by collecting the proportion of non-thermal segment degradation to the total degradation based on the set of thermally sensitive segments.

5. The method according to claim 1, characterized in that, The process of establishing a decay acceleration record based on the slope change of the elastic critical record analysis margin decay rate includes: The time series of batch margin difference values ​​extracted from the elastic critical record constitutes the margin change time series. The rate trend sequence is determined by a sudden jump correction window mean analysis of the margin change time series. Based on the rate trend sequence, the initial batch that transitions from pseudo-steady deceleration to accelerated growth is identified, forming the starting point of the slope acceleration. Based on the slope acceleration starting point, the difference in the mean rate before and after the acceleration starting point is used to establish a depletion acceleration record.

6. The method according to claim 1, characterized in that, The urgent early warning set is formed by identifying fault nodes whose trigger duration is insufficient for the nearest skylight based on the fault triggering timing sequence, including: A fault interval prediction table is established based on the predicted trigger intervals of each fault type according to the fault trigger timing. The amount of skylight gap is determined by comparing the time interval between the most recent skylights with the fault interval prediction table. Based on the insufficient clearance of the sunroof gap, the fault type is identified to form urgent projects before the sunroof. An urgent warning set is constructed by aggregating the trigger probability of urgent items and the time margin difference of the urgent items before the skylight.

7. The method according to claim 1, characterized in that, The step of extracting the remaining trigger duration of each typical fault from the remaining lifespan interval and outputting the fault trigger timing sequence includes: The fault lifetime boundary is obtained by statistically analyzing the boundaries of each fault mode interval using the remaining lifetime interval. The degradation trajectory type is determined by the evolution slope characteristics of the endpoints of the fault lifetime boundary analytical interval; Based on the degradation trajectory type, identify the remaining batch confidence range corresponding to the current stage to form a trigger confidence interval; The fault trigger timing sequence is output based on the trigger confidence interval, the trigger batch, and the two boundary times.

8. The method according to claim 5, characterized in that, The step of performing abrupt correction window mean analysis on the margin change time series to determine the rate trend sequence includes: Based on the time-series statistics of the margin change, a window mean sequence is generated from the average batch difference within each fixed window. The window mean change rate is obtained by analyzing the continuous change pattern of the difference between adjacent window means in the window mean sequence. Based on the window mean change rate, the smoothed and suppressed abrupt jump residuals are separated to obtain the abrupt jump residual sequence; The rate trend sequence is determined by correcting the window mean sequence based on the jump residual sequence.

9. The method according to claim 4, characterized in that, The step of using the temperature zone degradation mean to screen outlier segments for temperature difference and determine the segment temperature difference response includes: The absolute value of the segment temperature difference is obtained by calculating the absolute value of the difference between the average high and low temperature degradation values ​​of each segment for the average degradation value of the temperature zone. The distribution characteristics of the difference are obtained by statistically analyzing the mean and standard deviation of the absolute value of the temperature difference across all segments. Based on the difference distribution characteristics, a set of temperature difference outliers is constructed by screening standard deviation outliers. The segment temperature difference response is determined by aggregating the absolute value of the segment temperature difference based on the set of outlier segments with respect to temperature difference.

10. A comprehensive monitoring and AI prediction and early warning system for all parameters of a turnout, characterized in that, include: The data acquisition module is used to collect switch machine current data and contact detection data. Based on the switch machine current data, it collects the inflection point time and duration of the current slope of each conversion stage to generate segment division records. Based on the contact detection data, it tracks the rebound amount after the contact is in place to determine the contact elasticity margin. The segment analysis module is used to identify the segment current slope batch drift direction and rate based on the segment division record to form a segment decay sequence, and to obtain the elastic critical record by using the closely attached elastic margin statistical margin batch continuously decreasing node. The health assessment module is used to compare the segment degradation differences across temperature zones in the segment degradation sequence to generate a temperature decoupling index, identify the degradation amount of thermally stable segments based on the temperature decoupling index to determine the wear accumulation level, and calculate the score of each fault mode driven segment based on the wear accumulation level to obtain the component health index. The lifetime prediction module is used to establish a decay acceleration record based on the slope change of the margin decay rate according to the elastic critical record, and to calculate the remaining lifetime interval from the acceleration start point to the predicted critical time using the decay acceleration record. The remaining trigger time of each typical fault is extracted from the remaining lifetime interval and the fault trigger timing is output. The early warning output module is used to identify fault nodes whose trigger duration is insufficient for the nearest window based on the fault triggering sequence to form an urgent early warning set. The module combines the urgent early warning set with the component health index and uses a predictive early warning model to infer the severity of each fault and the maintenance response time limit, and outputs graded early warning instructions.

Images