A method and system for intelligent detection of the health status of rail vibration damping fasteners

Abstract

The present invention relates to the field of computer technology, and in particular to a method and system for intelligent detection of the health status of rail vibration-damping fasteners. The method comprises: S1: acquiring fastener vibration data and sorting them to obtain a fastener signal sequence, determining a first ratio and a second ratio based on the fastener signal sequence, obtaining an active damage index based on the fastener signal sequence, and determining a first fastener detection area based on the active damage index; S2: obtaining a waveform morphology difference based on the fastener signal sequence, and determining a second fastener detection area based on the waveform morphology difference, and finally determining a suspected fastener defect area based on the first fastener detection area and the second fastener detection area. The present invention is based on bimodal analysis and multi-level region division of vibration signals, combined with Huffman decision tree and coding compression technology, to achieve millisecond-level detection and high-reliability determination of rail fastener damage status.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for intelligently detecting the health status of track vibration-damping fasteners. Background Art

[0002] In recent years, vibration sensors have been widely deployed in the rail transit sector for fastener health monitoring, generating massive amounts of time-series vibration data, such as triaxial acceleration signals and environmental interference waveforms. However, existing data processing technologies face the following core issues in fault diagnosis:

[0003] Insufficient utilization of dynamic time series features: Traditional methods do not differentiate between data characteristics of train-passing and train-absent periods, such as the ratio of instantaneous force to the environmental value during the period. This causes key damage features, such as extreme frequency and abnormal duration, to be mixed with noise, reducing state recognition accuracy.

[0004] Inefficient real-time response: General compression algorithms, such as run-length encoding, uniformly process all data, ignoring the uneven distribution of health states. For example, damage states have low frequency but require priority response. They also fail to dynamically allocate encoding paths based on frequency domain energy characteristics, resulting in delayed damage determination.

[0005] Lack of adaptive multi-source interference: The existing model does not integrate environmental factors, such as abnormal weather index to correct the detection logic. It is unable to dynamically adjust the area division strategy or damage index weight under extreme conditions, resulting in a surge in false detection rate.

[0006] Although research attempts to improve diagnostic efficiency through machine learning, its rule design lacks the coordinated optimization of temporal correlation, such as the force-environment bimodal comparison, and state priority, such as the need for a short coding path to quickly respond to damage status. It is difficult to balance the interpretability and real-time performance of compressed data. Summary of the Invention

[0007] In order to overcome the shortcomings of insufficient efficiency in intelligent processing of time series data, the present invention provides a method and system for intelligent detection of the health status of track vibration-damping fasteners.

[0008] The technical implementation scheme of the present invention is: a method for intelligently detecting the health status of rail vibration damping fasteners, comprising the following steps:

[0009] S1: Acquire and sort fastener vibration data to obtain a fastener signal sequence, determine a first ratio and a second ratio based on the fastener signal sequence, obtain an active damage index based on the fastener signal sequence, and determine a first fastener detection area based on the active damage index;

[0010] S2: Obtaining a waveform morphology difference based on the fastener signal sequence, determining a second fastener detection area based on the waveform morphology difference, and ultimately determining a suspected fastener defect area based on the first fastener detection area and the second fastener detection area;

[0011] S3: Obtaining an abnormal weather index based on the suspected defective area of ​​the fastener, and determining an emergency correction mechanism based on the abnormal weather index; constructing a fastener health status determination tree based on a Huffman tree according to the emergency correction mechanism;

[0012] S4: compressing data on the nodes in the fastener health status determination tree to obtain a data compression result, and determining the health status of the track vibration reduction fastener based on the data compression result.

[0013] Preferably, the acquiring and sorting fastener vibration data to obtain a fastener signal sequence, and determining the first ratio and the second ratio based on the fastener signal sequence includes:

[0014] Obtain fastener vibration data through a three-axis vibration acceleration sensor;

[0015] Sorting the fastener vibration data in time series to obtain an arrangement result, and using the arrangement result as a fastener signal sequence;

[0016] Extracting the instantaneous signal value of the fastener signal sequence at the corresponding moment when the train passes, and taking the ratio of adjacent instantaneous signal values ​​as a first ratio, wherein the first ratio is used to reflect the stress state of the fastener;

[0017] Similarly, the time period mean of the fastener signal sequence corresponding to the time period when the train does not pass is extracted, and the ratio of the adjacent time period means is used as the second ratio, which is used to reflect the environmental interference characteristics.

[0018] Preferably, obtaining an active damage index according to the fastener signal sequence and determining a first fastener detection area according to the active damage index comprises:

[0019] Extracting, based on the fastener signal sequence, the frequency of occurrence of extreme values ​​of the first ratio sequence in a single train passing event and the abnormal duration of the first ratio between adjacent train passing events;

[0020] Performing a weighted summation on the extreme value occurrence frequency and the abnormality duration to obtain an active damage index;

[0021] If the active damage index exceeds a first preset threshold, the corresponding area is divided into a first fastener detection area.

[0022] Preferably, obtaining a waveform morphology difference based on the fastener signal sequence, determining a second fastener detection area based on the waveform morphology difference, and finally determining a fastener suspected defect area based on the first fastener detection area and the second fastener detection area includes:

[0023] extracting, based on the fastener signal sequence, an original waveform sequence of the second ratio when no train passes and a standard healthy template;

[0024] Performing dynamic time warping calculation on the original waveform sequence and the standard healthy template to obtain waveform morphology difference;

[0025] If the waveform difference exceeds a second preset threshold, the corresponding area is divided into a second fastener detection area;

[0026] The intersection of the first fastener detection area and the second fastener detection area is used as the fastener suspected defect area.

[0027] Preferably, obtaining an abnormal weather index based on the suspected defective area of ​​the fastener and determining an emergency correction mechanism based on the abnormal weather index include:

[0028] Extracting precipitation, wind speed, and visibility indicators based on historical meteorological data of the area where the fastener is suspected to be defective;

[0029] Performing weighted normalization calculation on the precipitation, wind speed, and visibility indicators to obtain an abnormal weather index;

[0030] If the abnormal weather index exceeds a third preset threshold, an emergency correction mechanism is activated.

[0031] Preferably, if the abnormal weather index exceeds a third preset threshold, the emergency correction mechanism is activated, including:

[0032] If the abnormal weather index exceeds a third preset threshold, a linear scaling coefficient formula is used to amplify the active damage index value;

[0033] If the abnormal weather index exceeds a fourth preset threshold, it is determined that an extreme interference condition has been reached, and the second fastener detection area division process for the monitoring period corresponding to the current abnormal weather index is immediately suspended;

[0034] If a maintenance decision needs to be generated, the emergency detection degree value is calculated and output through the emergency detection degree formula.

[0035] Preferably, if a maintenance decision needs to be generated, the emergency detection degree value is calculated and outputted through the emergency detection degree formula, including: the linear scaling coefficient formula is as follows, Among them, K is the scaling factor, α is the weather compensation coefficient, is the abnormal weather index, and the emergency detection degree formula is as follows: in, is the emergency detection level value, is the corrected active damage index, is the waveform morphology difference.

[0036] Preferably, the construction of a fastener health status determination tree based on a Huffman tree according to the emergency correction mechanism includes:

[0037] Extracting frequency domain energy distribution characteristics of the triaxial signal as coding elements based on the difference between the corrected active damage index and the waveform morphology;

[0038] Counting the occurrence frequencies of the frequency domain energy distribution features in the historical health status, and assigning codes according to the occurrence frequencies: high-frequency features are assigned short codes, and low-frequency features are assigned long codes;

[0039] A fastener health status determination tree is constructed with the health status type as the leaf node and the triaxial signal feature combination path as the branch.

[0040] Establish coding rules: the healthy state corresponds to a unique binary coding path, and the branch node path corresponding to the damaged state is prioritized.

[0041] Preferably, compressing the nodes in the fastener health status determination tree to obtain a data compression result, and determining the health status of the track vibration damping fastener according to the data compression result includes:

[0042] Performing prefix coding compression on the nodes of the fastener health status determination tree and merging child node data blocks of the same parent node;

[0043] Run-length encoding is used to compress a sequence of nodes with the same continuous health status;

[0044] After decompression, the decision tree is traversed according to the binary encoding path: when encountering a branch node, the three-axis signal feature combination conditions defined by the node are verified, and when reaching the leaf node, the final health status is output;

[0045] Prioritize short coding paths to achieve rapid response to damage status.

[0046] An intelligent health status detection system for rail vibration damping fasteners, comprising:

[0047] The signal acquisition and preprocessing module is used to obtain fastener vibration data through sensors and sort them to generate a fastener signal sequence; extract the instantaneous signal value when the train passes by to calculate the first ratio, and extract the time period average value when the train does not pass by to calculate the second ratio;

[0048] A damage region identification module is configured to calculate the extreme value frequency and abnormal duration based on the first ratio sequence, weight them to generate an active damage index, and divide the first fastener detection region; calculate the waveform morphology difference based on the second ratio, and divide the second fastener detection region; and take the intersection of the two to generate a suspected fastener defect region;

[0049] The weather correction and decision tree module calculates the abnormal weather index based on meteorological data. When a threshold is exceeded, a correction mechanism is activated: scaling the active damage index or suspending regional division. The frequency domain features of the three-axis signals are extracted based on the corrected index and the difference, and short / long codes are assigned according to the characteristic frequencies. A Huffman decision tree is constructed that prioritizes the short code path for the damage state.

[0050] The health status determination module is used to implement prefix coding compression on the nodes of the fastener health status determination tree; use run-length coding to compress a sequence of nodes with the same continuous health status; traverse the coding path after decompression: check the characteristic conditions of the branch nodes, reach the leaf node and output the health status; give priority to processing the damage status of the short coding path response.

[0051] Beneficial Effects: The present invention divides bimodal data into train-passed and train-absent periods based on the time series characteristics of vibration signals, extracting first and second ratios, respectively. A multi-level region partitioning mechanism is used to generate suspected fastener defect regions, dynamically modifying the damage determination threshold in conjunction with abnormal weather indices. A fastener health status determination tree is constructed based on frequency domain energy distribution characteristics, assigning codes based on feature occurrence frequency: high-frequency features are assigned short codes, and low-frequency features are assigned long codes. Prefix coding and run-length coding are used to compress node data, and after decompression, the short-coded path is prioritized for traversal to verify feature combination conditions. This method achieves millisecond-level response to damage status, reduces the false detection rate caused by environmental interference, improves the real-time performance and storage efficiency of health status determination, and provides highly reliable decision support for track safety operations and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of the intelligent detection method for the health status of rail vibration damping fasteners of the present invention;

[0053] Figure 2 This is a structural diagram of the intelligent detection system for the health status of track vibration-damping fasteners according to the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] Example 1: A method for intelligently detecting the health status of rail vibration damping fasteners, such as Figure 1 As shown, the following steps are included:

[0056] S1: Acquire and sort fastener vibration data to obtain a fastener signal sequence, determine a first ratio and a second ratio based on the fastener signal sequence, obtain an active damage index based on the fastener signal sequence, and determine a first fastener detection area based on the active damage index;

[0057] S2: Obtaining a waveform morphology difference based on the fastener signal sequence, determining a second fastener detection area based on the waveform morphology difference, and ultimately determining a suspected fastener defect area based on the first fastener detection area and the second fastener detection area;

[0058] S3: Obtaining an abnormal weather index based on the suspected defective area of ​​the fastener, and determining an emergency correction mechanism based on the abnormal weather index; constructing a fastener health status determination tree based on a Huffman tree according to the emergency correction mechanism;

[0059] S4: compressing data on the nodes in the fastener health status determination tree to obtain a data compression result, and determining the health status of the track vibration reduction fastener based on the data compression result.

[0060] Acquiring and sorting fastener vibration data to obtain a fastener signal sequence, and determining a first ratio and a second ratio based on the fastener signal sequence, including:

[0061] Obtain fastener vibration data through a three-axis vibration acceleration sensor;

[0062] Sorting the fastener vibration data in time series to obtain an arrangement result, and using the arrangement result as a fastener signal sequence;

[0063] Extracting the instantaneous signal value of the fastener signal sequence at the corresponding moment when the train passes, and taking the ratio of adjacent instantaneous signal values ​​as a first ratio, wherein the first ratio is used to reflect the stress state of the fastener;

[0064] Similarly, the time period mean of the fastener signal sequence corresponding to the time period when the train does not pass is extracted, and the ratio of the adjacent time period means is used as the second ratio, which is used to reflect the environmental interference characteristics.

[0065] It should be noted that the fastener vibration data is collected in real time by a three-axis vibration acceleration sensor, transmitted wirelessly to the gateway via Bluetooth or Zigbee, and uploaded to the cloud platform; the fastener vibration data is then arranged in chronological order to generate a fastener signal sequence.

[0066] The core restraint functions of rail fasteners in railway track systems include: (1) positioning constraints: fixing the spatial position of the rail; (2) pressure equalization constraints: distributing dynamic wheel-rail loads; (3) anti-displacement constraints: suppressing lateral and longitudinal deformation of the track; and (4) vibration reduction constraints: attenuating vibration energy transfer. Failure of the healthy state will directly weaken the above restraint functions, causing the track geometry to deteriorate.

[0067] Traditional rail fastener monitoring does not distinguish between the data characteristics of the time periods when trains have passed or not passed, resulting in damage characteristics, such as abnormal instantaneous stress, and environmental noise, such as wind vibration interference, aliasing, and reduced detection accuracy.

[0068] Data acquisition: Fastener vibration data is collected in real time using a triaxial vibration accelerometer and arranged in chronological order to generate a fastener signal sequence, which is an array of vibration values ​​in the time dimension. Dual-modal ratio extraction: First ratio: Extract the instantaneous signal value when the train passes, that is, the millisecond-level vibration peak, and calculate the ratio of adjacent instantaneous values. The definition formula is: ,in, is the instantaneous ratio of the dynamic stress state, is the instantaneous value of vibration at the current moment, is the instantaneous vibration value at the previous moment. This ratio reflects the dynamic stress state of the fastener. For example, loose bolts may cause a sudden change in the ratio. Second ratio: extract the mean value of the time period when the train is not passing, such as the 10-minute window mean, and calculate the ratio of the mean values ​​of adjacent time periods. The definition formula is: ,in, is the ratio of the environmental interference characteristic time period, is the vibration mean value of the next period, is the mean vibration value of the previous period. This ratio quantifies the characteristics of environmental interference, such as baseline fluctuations caused by wind vibration.

[0069] This step separates the force and environment data to avoid noise masking the real damage signal, e.g. The sudden increase indicates that the bolt has failed. Stability eliminates weather interference. Dual-modal comparison improves feature differentiation and provides pure input for subsequent damage index calculations.

[0070] Example scenario: A subway track fastener monitoring point with a sampling frequency of 1kHz. Steps: Data acquisition and sorting: A triaxial sensor collects 10 minutes of vibration data, including two train passes. This data is sorted by timestamp to generate a sequence: [t0: 0.02g, t1: 0.05g, ..., t100: 1.8g (train pass), t101: 2.0g, ...], in units of gravitational acceleration (g). First ratio calculation: Extract the instantaneous value at the train pass, for example, t100 = 1.8g, t101 = 2.0g, and calculate the ratio: =2.0 / 1.8≈1.11. > ,For example =1.0, this abnormal ratio indicates that the bolt preload has decreased. Second ratio calculation: extract the mean value of the no-train period, for example, the mean value of [t0-t99] = 0.03g, the mean value of [t200-t299] = 0.07g, calculate the ratio: =0.07 / 0.03≈2.33, if > ,For example =1.5, which indicates that strong wind interference requires subsequent correction. Effect: Through dual-mode ratio separation, bolt loosening can be accurately identified, i.e. anomalies and quantify the wind noise impact, i.e. Exceeding the standard phenomenon to avoid misjudgment.

[0071] Obtaining an active damage index according to the fastener signal sequence, and determining a first fastener detection area according to the active damage index, including:

[0072] Extracting, based on the fastener signal sequence, the frequency of occurrence of extreme values ​​of the first ratio sequence in a single train passing event and the abnormal duration of the first ratio between adjacent train passing events;

[0073] Performing a weighted summation on the extreme value occurrence frequency and the abnormality duration to obtain an active damage index;

[0074] If the active damage index exceeds a first preset threshold, the corresponding area is divided into a first fastener detection area.

[0075] It should be noted that the existing technology only analyzes the vibration peak of a single train passing by, ignoring the relationship between repetitive impact and cross-cycle damage evolution, resulting in the missed detection of progressive faults such as bolt cracks.

[0076] Feature extraction: Extreme value occurrence frequency: Detect extreme values ​​of the first ratio sequence of a single train passing event, such as local maximum / minimum values, and count the number of extreme value points detected, which reflects the accumulation of instantaneous impacts; Abnormal duration: The duration of time that the first ratio continuously exceeds the threshold between adjacent train passing events, which reflects the continuous deterioration of damage. Damage quantification: Weighted summation formula: ,in, : Active damage index, dimensionless, : normalized extreme frequency, dimensionless; : Normalized abnormal duration, dimensionless; α, β: weight coefficients, satisfying α+β=1, adjusted according to the damage type. Before weighted summation, the extreme value frequency and abnormal duration are normalized to make them in the same magnitude range. >First preset threshold , marked as the first fastener detection area. Threshold setting: Determined by the statistical distribution of historical health data, such as the 95% quantile.

[0077] This step improves the detection rate of progressive faults by integrating transient and persistent damage features. Weights are dynamically allocated, for example, α = 0.7 enhances the impact of high-frequency shocks.

[0078] Example description, scenario: A subway fastener continuously monitors three trains passing through, each with an interval of 10 minutes. Steps: 1. Feature extraction: First train: The first ratio sequence is subjected to extreme value detection, local maximum / minimum values ​​are detected, and the number of extreme value points exceeding the threshold is counted. =3; The interval between the first and second trains: the duration of the first ratio sequence exceeding the threshold is = 120 seconds, the result is the second ratio <1.2 Eliminate environmental interference and confirm damage; 2. Normalize based on historical statistical boundaries: , : =3 / 5=0.6; =120 / 300=0.4; 3. Damage calculation, assuming weight coefficient = 0.6, β = 0.4: =0.6*0.6+0.4*0.4=0.36+0.16=0.52; 4. Region determination, normalized threshold =0.5: =0.52> =0.5, so it is included in the first fastener detection area; Effect: high frequency impact, for example =3 indicates bolt fatigue, and continuous abnormality, such as =120s indicates gasket aging, triggers early warnings in a coordinated manner, and locates high-risk areas.

[0079] The first ratio: the ratio of adjacent instantaneous vibration peaks when a train passes, which characterizes the dynamic stress state of the fastener; the first ratio sequence: a collection of all continuously calculated "first ratios" in a single train passing event arranged in time sequence. The first preset threshold: the critical value for damage judgment determined based on the statistical distribution of historical health data. A single train passing event refers to the complete vibration signal collection period from the time when the train wheels begin to contact the sleeper to which the current fastener belongs to, to the time when they completely detach from the sleeper. An adjacent train passing event refers to the time interval between two consecutive trains passing the sleeper where the same fastener is located, that is, the continuous monitoring period from the time when the previous train detaches from the sleeper to the time when the next train contacts the sleeper.

[0080] Obtaining a waveform morphology difference according to the fastener signal sequence, determining a second fastener detection area according to the waveform morphology difference, and finally determining a fastener suspected defect area according to the first fastener detection area and the second fastener detection area, including:

[0081] extracting, based on the fastener signal sequence, an original waveform sequence of the second ratio when no train passes and a standard healthy template;

[0082] Performing dynamic time warping calculation on the original waveform sequence and the standard healthy template to obtain waveform morphology difference;

[0083] If the waveform difference exceeds a second preset threshold, the corresponding area is divided into a second fastener detection area;

[0084] The intersection of the first fastener detection area and the second fastener detection area is used as the fastener suspected defect area.

[0085] It should be noted that existing methods directly analyze the original environmental vibration waveform and ignore the time axis expansion and contraction deformation. For example, the gradual change of wind speed causes the waveform to stretch, which can easily misjudge environmental interference as damage.

[0086] Data extraction: Original waveform sequence: A continuous second ratio data sequence during periods when trains are not passing, reflecting the form of environmental interference; Standard health template: A benchmark waveform of the second ratio under historical normal conditions, such as a 24-hour average sequence in calm weather. Difference calculation: Dynamic time warping algorithm is used to align the time axis and calculate the minimum path distance: ,in: : waveform morphology difference; : original waveform sequence points; : Standard template sequence point. The waveform morphology difference is normalized: ,in is the original DTW distance, is the statistical mean of the DTW distance of the healthy template. Region determination: If >Second preset threshold , divided into the second fastener detection area, that is, the environmental interference significant area; take the intersection of the first fastener detection area, that is, the stress abnormal area, and the second area to generate the fastener suspected defect area to exclude pure environmental interference. Threshold setting: =2.0* , is the standard deviation of the DTW distance of the healthy template.

[0087] This step uses the DTW algorithm to eliminate the influence of time offset and accurately quantify morphological differences; the dual-region intersection mechanism suppresses false positives caused by environmental noise.

[0088] Example description, scenario: monitoring data of a track fastener in strong winds. Steps: Data extraction: Original waveform sequence: Ratio sequence such as 1.05, 0.98, 1.20, ...; Standard healthy template: windless weather benchmark sequence [1.02, 0.98, 1.01, 0.99, ...]. Difference calculation: calculated after DTW alignment =8.6, healthy template =2.1 calculated by the threshold formula =4.2. Area determination: =8.6>4.2, so it is included in the second fastener inspection area; if the location also belongs to the first fastener inspection area, such as a loose bolt, it is marked as a suspected fastener defect area. Effect: Strong wind causes the overall waveform to shift as shown The image quality is high, but it does not cover up the real damage. It is located through intersection to avoid false detection.

[0089] Obtaining an abnormal weather index based on the suspected defective area of ​​the fastener, and determining an emergency correction mechanism based on the abnormal weather index, including:

[0090] Extracting precipitation, wind speed, and visibility indicators based on historical meteorological data of the area where the fastener is suspected to be defective;

[0091] Performing weighted normalization calculation on the precipitation, wind speed, and visibility indicators to obtain an abnormal weather index;

[0092] If the abnormal weather index exceeds a third preset threshold, an emergency correction mechanism is activated.

[0093] It should be noted that the existing technology does not integrate meteorological factors to correct damage judgment. Extreme weather, such as heavy rain and strong winds, will amplify vibration signals, causing healthy fasteners to be misjudged as damaged.

[0094] Data acquisition: Real-time weather station data from the suspected fastener defect area is collected to extract: precipitation (in mm / h), wind speed (in m / s), visibility (in km); Index calculation: Normalized weighted formula: ,in: : Abnormal weather index, value range [0,1]; P, V, D: real-time precipitation, wind speed, visibility; , , : Historical maximum threshold; , , : weight coefficient, and + + = 1. Correction mechanism: If >Third preset threshold , start the emergency correction mechanism. Threshold setting: =0.7, determined by the 90th percentile of the distribution of meteorological impact levels in historical false alarm cases.

[0095] This step dynamically suppresses misjudgments caused by environmental noise by quantifying the intensity of meteorological interference.

[0096] Example description, scenario: A suspected defective area of ​​a rail fastener encounters heavy rain. Steps: Meteorological data extraction: precipitation P = 40mm / h, =100; wind speed V=15m / s, =30; visibility D=1km, =10; exponential calculation, let =0.4, =0.3, =0.3: =0.16+0.15+0.27=0.58; Mechanism trigger: If =0.7. Since 0.58 < 0.7, no emergency correction is initiated and the meteorological interference is deemed acceptable. Effect: The heavy rain does not exceed the threshold, avoiding the initiation of redundant maintenance on normal fasteners.

[0097] Abnormal weather index: integrates the normalized weighted values ​​of precipitation, wind speed and visibility to quantify the intensity of environmental interference; emergency correction mechanism: when the abnormal weather index exceeds the threshold, the emergency strategy of the damage judgment logic is dynamically adjusted.

[0098] If the abnormal weather index exceeds the third preset threshold, an emergency correction mechanism is activated, including:

[0099] If the abnormal weather index exceeds a third preset threshold, a linear scaling coefficient formula is used to amplify the active damage index value;

[0100] If the abnormal weather index exceeds a fourth preset threshold, it is determined that an extreme interference condition has been reached, and the second fastener detection area division process for the monitoring period corresponding to the current abnormal weather index is immediately suspended;

[0101] If a maintenance decision needs to be generated, the emergency detection degree value is calculated and output through the emergency detection degree formula.

[0102] It should be noted that existing systems still perform inspections mechanically in extreme weather conditions, where heavy rain / strong winds can distort vibration signals, leading to distorted damage indices. For example, healthy fasteners may be misjudged as high-risk, or areas of environmental interference may be mislabeled.

[0103] The scaling compensation of the hierarchical correction mechanism is activated when any of the following conditions are met: (1) > , that is, the abnormal weather index exceeds the threshold, (2) < , that is, the environmental interference ratio is lower than the lower threshold, such as =0.67, Is the second ratio The lower limit threshold of environmental interference is used to trigger the damage index compensation mechanism when the environmental vibration is abnormally low. The calculation rule of the scaling factor K is: < , that is, regardless of If the threshold is exceeded or not exceeded, , otherwise if > ,but , in other cases K=1, no scaling. The process is suspended, > :like =0.9 Based on the 99th percentile setting of historical extreme weather cases, the second fastener inspection area division is suspended to avoid environmental interference and mislabeling. Maintenance decision: Emergency inspection degree formula: ; The numerator strengthens the damage characteristics, and the denominator reduces the weight of weather interference.

[0104] This step balances reliability and efficiency through a hierarchical response mechanism. =0.7, =0.9; The corrected damage and environmental interference are integrated to output the credible maintenance priority.

[0105] Example description, scenario: a fastener in typhoon weather =0.85, =0.7, =0.9. Steps: Scaling compensation, set α=0.2: =5.85; Process decision: =0.85< =0.9, so the area division is not suspended; maintenance calculation: set the waveform difference =6.0, substituting into the formula we get = =234, this value is high priority. Effect: The active damage index is reasonably amplified during typhoon days, but does not reach the extreme threshold, and the output is normal. Guide maintenance.

[0106] Linear scaling coefficient formula: a compensation algorithm that dynamically amplifies the active damage index through the weather index; fourth preset threshold: a critical value for suspending the detection process based on historical extreme weather event statistics.

[0107] If a maintenance decision needs to be generated, the emergency detection degree value is calculated and output through the emergency detection degree formula, including: the linear scaling coefficient formula is as follows, Among them, K is the scaling factor, α is the weather compensation coefficient, is the abnormal weather index, and the emergency detection degree formula is as follows: in, is the emergency detection level value, is the corrected active damage index, is the waveform morphology difference.

[0108] It's important to note that existing rail fastener inspection systems can easily misinterpret environmental disturbances as damage due to vibration signal distortion during unusual weather conditions, such as heavy rain or strong winds. This can lead to healthy fasteners being labeled as high-risk, or the true extent of damage being underestimated. A core flaw lies in the lack of a dynamic correlation between meteorological disturbances and damage signatures.

[0109] Data acquisition: Abnormal weather index :Get the precipitation in the fastener area in real time through the weather station, unit is mm / h, wind speed, unit is m / s, visibility, unit is km, and calculate it by normalization weighting. , quantify the intensity of environmental interference. Corrected Active Damage Index Active Damage Index Dynamically adjusted by the linear scaling formula, the formula is Waveform morphology difference :Compare the morphological differences between the environmental vibration waveform and the healthy template through the dynamic time warping algorithm. Formula definition: Linear scaling coefficient formula: ,when < (like =0.67), and synchronously start damage index scaling, where K is the scaling factor and α is the weather compensation factor. The default value of α is 0.2, which is used to compensate for the underestimation of damage caused by signal attenuation. Emergency detection degree formula: ,when ≥0.99, that is, extreme weather interference, in order to avoid the denominator Close to zero causes calculation overflow, use As the emergency detection level value. The coefficient 100 is used to amplify the damage feature weight to ensure that the damage status can still be identified first in extreme weather conditions. ; Molecularly enhanced damage characteristics, i.e. and The product, the denominator suppresses the weather interference weight, , output maintenance priority. Threshold setting: The third preset threshold, e.g. =0.7: Based on the 90th percentile of the degree of meteorological impact in historical false alarm cases, triggering scaling compensation. The fourth preset threshold, for example =0.9: Based on the 99th percentile setting of extreme weather events, such as typhoons, the detection process is triggered to be suspended. Graded response: > Time Zoom , > Suspend the environmental area division at this time to balance the detection efficiency and reliability. Trusted decision: Integrate corrected damage and environmental interference to avoid misjudgment of a single indicator.

[0110] Example description, scenario: A rail fastener is monitored in typhoon weather =0.85, precipitation P=45mm / h, wind speed V=18m / s, visibility D=0.8km, original active damage index =5.0, waveform difference =6.0. Step: Scaling compensation > =0.7: =1.17, calculated =1.17*5.0=5.85; process decision < =0.9: Do not suspend the second fastener inspection area division. Inspection priority calculation: = =234; Effect: Active damage index in typhoon weather is adjusted from 5.0 to 5.85 to reflect the actual damage level; output high emergency value, =234, guiding priority repair to avoid missing real faults due to environmental noise.

[0111] According to the emergency correction mechanism, a fastener health status determination tree is constructed based on the Huffman tree, including:

[0112] Extracting frequency domain energy distribution characteristics of the triaxial signal as coding elements based on the difference between the corrected active damage index and the waveform morphology;

[0113] Counting the occurrence frequencies of the frequency domain energy distribution features in the historical health status, and assigning codes according to the occurrence frequencies: high-frequency features are assigned short codes, and low-frequency features are assigned long codes;

[0114] A fastener health status determination tree is constructed with the health status type as the leaf node and the triaxial signal feature combination path as the branch.

[0115] Establish coding rules: the healthy state corresponds to a unique binary coding path, and the branch node path corresponding to the damaged state is prioritized.

[0116] It's important to note that existing rail fastener health monitoring systems use a fixed decision tree, assigning the same processing path to high-frequency health states and low-frequency damage states. This results in delayed damage response. For example, a bolt fracture may take 200ms, but the system takes 500ms to traverse a long path. Furthermore, traditional methods fail to leverage the state distribution characteristics to optimize decision logic.

[0117] Data acquisition and processing: Frequency domain energy distribution characteristics of triaxial signals: Corrected active damage index Difference from waveform After FFT conversion to frequency domain, the X / Y / Z three-axis energy ratio vector is extracted and normalized to meet =1, such as [0.3, 0.3, 0.4], It is the normalized proportion of each axial vibration energy in the total energy after the three-axis signal is converted by FFT. i=1, 2, 3 are X, Y, and Z axes respectively, which serve as coding elements. Feature frequency statistics: Analyze the probability of occurrence of each feature combination in the healthy / damaged state in the historical data. For example, the occurrence rate of [0.3, 0.3, 0.4] in the healthy state is 87%. Huffman tree construction rules: Short code allocation: high-frequency features with an occurrence probability > 70% are assigned short binary codes, such as "0"; long code allocation: low-frequency features, such as damage features [0.8, 0.1, 0.1] with an occurrence rate of 5%, are assigned long codes, such as "1110"; tree structure design: leaf nodes: health status type, such as "normal" and "loose bolts"; branch nodes: three-axis signal feature combination conditions, such as >0.6; Priority path: Place the short coding path corresponding to the damage state on the left branch of the decision tree to achieve priority judgment and millisecond-level priority judgment.

[0118] This step uses frequency domain feature compression: using energy distribution to characterize damage patterns and improve feature differentiation; dynamic coding optimization: long encoding of high-frequency healthy states to reduce storage, and short encoding of low-frequency damaged states to accelerate response.

[0119] Example description, scenario: Historical data of a subway fastener monitoring point shows: Health status characteristics =[0.3,0.3,0.4], the probability of occurrence is 85%; bolt loosening damage characteristics =[0.7,0.2,0.1], the probability of occurrence is 8%; gasket aging damage characteristics =[0.1,0.1,0.8], the probability of occurrence is 5%, , , for Specific instantiation features. Huffman tree construction and judgment: coding allocation, in ascending order of frequency: 85%, assigned the shortest code "0"; 8%, assigned the medium-long code "10"; 5%, the longest code is assigned to "11". Tree structure: Branch node 1: Check <0.5? If not, corresponding If the status is "normal", the output is "normal" and the code is "0". If so, go to branch node 2: check >0.6? If yes, corresponding status, then output "bolt loose", code "10"; if not, the corresponding Status, then output "gasket aging", code "11". Detection process: real-time features =[0.1,0.2,0.7] satisfies >0.5, so "gasket aging" is directly output, with only one step of judgment. For the gasket aging damage type, the correlation between the vibration frequency domain characteristics of the gasket aging damage type and the rail corrugation phenomenon can be further analyzed. When the corrugation-related features are identified, the fastener parameter customization design can be triggered synchronously: Corrugation feature extraction: Based on the frequency domain energy distribution of the three-axis signal, the energy proportion of the corrugation characteristic frequency band 200-500Hz is calculated ; Customized design conditions: If > , =0.7, and the active damage index > , then the design parameters of the anti-wave wear fastener are generated: Stiffness improvement: , β=0.2, is the base stiffness; resonance control: adjust the mass to make the resonant frequency <150Hz, avoid the corrugation frequency band; : The energy ratio of the corrugation characteristic frequency band is calculated using the existing frequency domain characteristic data; : The judgment threshold value calibrated based on historical corrugation data; : Customized fastener stiffness value, unit: N / mm; : Current fastener base stiffness; β: Stiffness compensation coefficient; : resonant frequency of the fastener-sleeper system; m: equivalent mass of the system, adjusted by adding a mass block Control target.

[0120] Performing data compression on nodes in the fastener health status determination tree to obtain a data compression result, and determining the health status of the track vibration damping fastener based on the data compression result, including:

[0121] Performing prefix coding compression on the nodes of the fastener health status determination tree and merging child node data blocks of the same parent node;

[0122] Run-length encoding is used to compress a sequence of nodes with the same continuous health status;

[0123] After decompression, the decision tree is traversed according to the binary encoding path: when encountering a branch node, the three-axis signal feature combination conditions defined by the node are verified, and when reaching the leaf node, the final health status is output;

[0124] Prioritize short coding paths to achieve rapid response to damage status.

[0125] It's important to note that existing rail fastener inspection systems store all decision tree node data, resulting in high storage overhead for massive monitoring points. For example, a single kilometer of track consumes an average of 1GB of data per day. Furthermore, after decompression, all nodes must be traversed sequentially, resulting in high damage response latency, exceeding 500ms for example. A core flaw lies in the lack of optimized compression and retrieval logic based on state distribution characteristics.

[0126] Data compression: Prefix coding compression: Merge child node data blocks of the same parent node in the Huffman tree, for example, "bolt loosening" and "gasket aging" under the parent node "abnormal" are merged into a single data block to eliminate duplicate path descriptions. Run-Length Encoding (RLE): Compress a sequence of nodes with the same continuous health status, for example, "normal-normal-normal" is recorded as "normal × 3", reducing the storage of duplicate states. Decompression and judgment: Traverse the tree according to the binary encoding path: Branch nodes check the three-axis feature conditions, for example >0.6, the leaf node outputs the status; short code priority: the damage state corresponds to the short code, for example, the short code "01" path is processed first, and the healthy state long code path is skipped.

[0127] The continuous sequence of nodes with the same health state refers to a continuous group of nodes in the Huffman tree that output the same health state for adjacent fastener units in space or time. Prefix coding compression utilizes the Huffman tree hierarchy to merge and store child nodes of the same parent node. The fastener health state determination tree is a Huffman tree-structured decision tree constructed based on the frequency domain characteristics of the triaxial signal. This prioritizes low-frequency damage states by assigning short coding paths.

[0128] This step uses two-level compression: prefix coding optimizes tree structure storage and RLE compression of state sequences, improving storage efficiency by 60%; priority retrieval: short code paths are given priority response, speeding up damage determination to milliseconds.

[0129] For example, consider a scenario where the health status sequence of 10 fasteners in a track section includes 7 normal fasteners, 1 loose bolt, and 2 gaskets with aging. The Huffman encoding for these fasteners is: Normal = "001", Loose Bolt = "01", and Aged Gasket = "1". The compression and decision process involves prefix coding compression, which merges sub-blocks with the same parent node, such as the loose / aging nodes under the "abnormal" parent node. RLE compression compresses consecutive normal fastener sequences into "001 × 7". Decompression and decision processing prioritizes short codes. A code of "1" indicates gasket aging, triggering the immediate output of two fastener damages. The code "01" indicates a loose bolt, triggering the output of one damage. Finally, decompression of "001 × 7" triggers the output of a normal state. The result: Storage capacity is reduced from the original 30 nodes to 12 data units, achieving a 60% compression ratio. The damage status response time is reduced from 300ms to 100ms, with short code paths prioritized.

[0130] Triaxial signal: refers to the raw acceleration data sequence collected by a triaxial vibration accelerometer, reflecting the vibration intensity of the rail fastener in the three spatial dimensions of X / Y / Z. Triaxial signal feature combination: refers to the feature vector formed by normalizing the frequency domain energy distribution of the triaxial signal through FFT transformation, which represents the proportion of vibration energy in each axis. Triaxial signal feature combination condition: refers to the logical judgment rule set in the branch node of the fastener health status determination tree, which is used to verify whether the frequency domain feature vector meets the specific damage mode threshold.

[0131] Example 2: Based on Example 1, an intelligent health status detection system for track vibration damping fasteners, such as Figure 2 Shown, including:

[0132] The signal acquisition and preprocessing module is used to obtain fastener vibration data through sensors and sort them to generate a fastener signal sequence; extract the instantaneous signal value when the train passes by to calculate the first ratio, and extract the time period average value when the train does not pass by to calculate the second ratio;

[0133] A damage region identification module is configured to calculate the extreme value frequency and abnormal duration based on the first ratio sequence, weight them to generate an active damage index, and divide the first fastener detection region; calculate the waveform morphology difference based on the second ratio, and divide the second fastener detection region; and take the intersection of the two to generate a suspected fastener defect region;

[0134] The weather correction and decision tree module calculates the abnormal weather index based on meteorological data. When a threshold is exceeded, a correction mechanism is activated: scaling the active damage index or suspending regional division. The frequency domain features of the three-axis signals are extracted based on the corrected index and the difference, and short / long codes are assigned according to the characteristic frequencies. A Huffman decision tree is constructed that prioritizes the short code path for the damage state.

[0135] The health status determination module is used to implement prefix coding compression on the nodes of the fastener health status determination tree; use run-length coding to compress a sequence of nodes with the same continuous health status; traverse the coding path after decompression: check the characteristic conditions of the branch nodes, reach the leaf node and output the health status; give priority to processing the damage status of the short coding path response.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent detection method for the health status of rail vibration damping fasteners, characterized in that: The following steps are involved: S1: Acquire and sort fastener vibration data to obtain a fastener signal sequence, determine a first ratio and a second ratio based on the fastener signal sequence, obtain an active damage index based on the fastener signal sequence, and determine a first fastener detection area based on the active damage index; S2: Obtaining a waveform morphology difference based on the fastener signal sequence, determining a second fastener detection area based on the waveform morphology difference, and ultimately determining a suspected fastener defect area based on the first fastener detection area and the second fastener detection area; S3: Obtaining an abnormal weather index based on the suspected defective area of ​​the fastener, and determining an emergency correction mechanism based on the abnormal weather index; constructing a fastener health status determination tree based on a Huffman tree according to the emergency correction mechanism; S4: compressing data on the nodes in the fastener health status determination tree to obtain a data compression result, and determining the health status of the track vibration reduction fastener based on the data compression result.

2. The intelligent detection method for the health status of rail vibration damping fasteners according to claim 1 is characterized in that: The acquiring and sorting fastener vibration data to obtain a fastener signal sequence, and determining a first ratio and a second ratio based on the fastener signal sequence, includes: Obtain fastener vibration data through a three-axis vibration acceleration sensor; Sorting the fastener vibration data in time series to obtain an arrangement result, and using the arrangement result as a fastener signal sequence; Extracting the instantaneous signal value of the fastener signal sequence at the corresponding moment when the train passes, and taking the ratio of adjacent instantaneous signal values ​​as a first ratio, wherein the first ratio is used to reflect the stress state of the fastener; Similarly, the time period mean of the fastener signal sequence corresponding to the time period when the train does not pass is extracted, and the ratio of the adjacent time period means is used as the second ratio, which is used to reflect the environmental interference characteristics.

3. The intelligent detection method for the health status of rail vibration damping fasteners according to claim 1 is characterized in that: Obtaining an active damage index according to the fastener signal sequence, and determining a first fastener detection area according to the active damage index, includes: Extracting, based on the fastener signal sequence, the frequency of occurrence of extreme values ​​of the first ratio sequence in a single train passing event and the abnormal duration of the first ratio between adjacent train passing events; Performing a weighted summation on the extreme value occurrence frequency and the abnormality duration to obtain an active damage index; If the active damage index exceeds a first preset threshold, the corresponding area is divided into a first fastener detection area.

4. The intelligent health status detection method for rail vibration damping fasteners according to claim 1 is characterized in that: The method of obtaining a waveform morphology difference based on the fastener signal sequence, determining a second fastener detection area based on the waveform morphology difference, and finally determining a fastener suspected defect area based on the first fastener detection area and the second fastener detection area includes: extracting, based on the fastener signal sequence, an original waveform sequence of the second ratio when no train passes and a standard healthy template; Performing dynamic time warping calculation on the original waveform sequence and the standard healthy template to obtain waveform morphology difference; If the waveform difference exceeds a second preset threshold, the corresponding area is divided into a second fastener detection area; The intersection of the first fastener detection area and the second fastener detection area is used as the fastener suspected defect area.

5. The intelligent health status detection method for rail vibration damping fasteners according to claim 1 is characterized in that: Obtaining an abnormal weather index based on the suspected defective area of ​​the fastener, and determining an emergency correction mechanism based on the abnormal weather index, includes: Extracting precipitation, wind speed, and visibility indicators based on historical meteorological data of the area where the fastener is suspected to be defective; Performing weighted normalization calculation on the precipitation, wind speed, and visibility indicators to obtain an abnormal weather index; If the abnormal weather index exceeds a third preset threshold, an emergency correction mechanism is activated.

6. The intelligent detection method for the health status of a track vibration damping fastener according to claim 5, characterized in that: If the abnormal weather index exceeds the third preset threshold, the emergency correction mechanism is activated, including: If the abnormal weather index exceeds a third preset threshold, a linear scaling coefficient formula is used to amplify the active damage index value; If the abnormal weather index exceeds a fourth preset threshold, it is determined that an extreme interference condition has been reached, and the second fastener detection area division process for the monitoring period corresponding to the current abnormal weather index is immediately suspended; If a maintenance decision needs to be generated, the emergency detection degree value is calculated and output through the emergency detection degree formula.

7. The intelligent health status detection method for rail vibration damping fasteners according to claim 6 is characterized in that: If a maintenance decision needs to be generated, the emergency detection degree value is calculated and outputted through the emergency detection degree formula, including: the linear scaling coefficient formula is as follows, Among them, K is the scaling factor, α is the weather compensation coefficient, is the abnormal weather index, and the emergency detection degree formula is as follows: in, is the emergency detection level value, is the corrected active damage index, is the waveform morphology difference.

8. The intelligent health status detection method for rail vibration damping fasteners according to claim 7 is characterized in that: The method of constructing a fastener health status determination tree based on the Huffman tree according to the emergency correction mechanism includes: Extracting frequency domain energy distribution characteristics of the triaxial signal as coding elements based on the difference between the corrected active damage index and the waveform morphology; Counting the occurrence frequencies of the frequency domain energy distribution features in the historical health status, and assigning codes according to the occurrence frequencies: high-frequency features are assigned short codes, and low-frequency features are assigned long codes; A fastener health status determination tree is constructed with the health status type as the leaf node and the triaxial signal feature combination path as the branch. Establish coding rules: the healthy state corresponds to a unique binary coding path, and the branch node path corresponding to the damaged state is prioritized.

9. The intelligent health status detection method for rail vibration damping fasteners according to claim 1 is characterized in that: The step of compressing data on the nodes in the fastener health status determination tree to obtain a data compression result, and determining the health status of the track vibration damping fastener according to the data compression result includes: Performing prefix coding compression on the nodes of the fastener health status determination tree and merging child node data blocks of the same parent node; Run-length encoding is used to compress a sequence of nodes with the same continuous health status; After decompression, the fastener health status determination tree is traversed according to the binary code path: when encountering a branch node, the three-axis signal feature combination conditions defined at the node are verified, and when reaching a leaf node, the final health status is output; Prioritize short coding paths to achieve rapid response to damage status.

10. An intelligent health status detection system for rail vibration damping fasteners, used to implement the intelligent health status detection method for rail vibration damping fasteners according to any one of claims 1 to 9, characterized in that: include: The signal acquisition and preprocessing module is used to obtain fastener vibration data through sensors and sort them to generate a fastener signal sequence; extract the instantaneous signal value when the train passes by to calculate the first ratio, and extract the time period average value when the train does not pass by to calculate the second ratio; A damage region identification module is configured to calculate the extreme value frequency and abnormal duration based on the first ratio sequence, weight them to generate an active damage index, and divide the first fastener detection region; calculate the waveform morphology difference based on the second ratio, and divide the second fastener detection region; and take the intersection of the two to generate a suspected fastener defect region; The weather correction and decision tree module calculates the abnormal weather index based on meteorological data. When a threshold is exceeded, a correction mechanism is activated: scaling the active damage index or suspending regional division. The frequency domain features of the three-axis signals are extracted based on the corrected index and the difference, and short / long codes are assigned according to the characteristic frequencies. A Huffman decision tree is constructed that prioritizes the short code path for the damage state. A health status determination module is configured to implement prefix coding compression on the nodes of the fastener health status determination tree; compress a sequence of consecutive nodes with the same health status using run-length coding; traverse the coding path after decompression, verify the characteristic conditions of the branch nodes, and output the health status when reaching the leaf node; Prioritize short coding paths in response to impairment conditions.

Images