Track wobble section cause diagnosis method and system based on multi-source data fusion

Abstract

This invention relates to a method and system for diagnosing the causes of track swaying sections based on multi-source data fusion, belonging to the field of rail transit condition monitoring technology. The method includes: acquiring onboard vibration response data and track static geometry data of the target line, and performing spatiotemporal alignment processing; performing statistical analysis on the onboard vibration response data to identify abnormal vibration sections; extracting the track-direction irregularity component and horizontal irregularity component from the track static geometry data corresponding to the abnormal vibration sections, and calculating the phase correlation characteristic value between them; constructing a dynamic model including track stiffness parameters, and performing parameter inversion on the dynamic model with the objective of minimizing the residual between the simulated vibration response output by the dynamic model and the onboard vibration response data, to obtain the equivalent stiffness degradation parameters of the abnormal vibration sections; and determining the swaying cause type of the abnormal vibration sections based on the phase correlation characteristic value and the equivalent stiffness degradation parameter.

Description

Technical Field

[0001] This invention relates to the field of rail transit condition monitoring technology, and in particular to a method and system for diagnosing the causes of track swaying sections based on multi-source data fusion. Background Technology

[0002] With the continuous expansion of urban rail transit networks and the increase in service life, the issue of train running stability has become increasingly prominent. Track sway refers to the phenomenon of low-frequency lateral swaying of the train body during operation, caused by excitation sources such as track geometry deviations, structural support performance degradation, or poor wheel-rail matching. This abnormal vibration not only significantly reduces passenger comfort but, in severe cases, may also induce derailment risks and accelerate fatigue damage to the vehicle's running gear and track components. Therefore, establishing a monitoring system capable of accurately locating swaying sections, quantitatively assessing track conditions, and diagnosing the causes of defects is of great significance for guiding precise track maintenance.

[0003] Currently, track condition evaluation mainly relies on manual rides and track inspection vehicles. Manual rides depend primarily on the subjective judgment of experienced drivers or technicians to assess swaying, lacking quantitative evaluation standards and failing to meet the demands of high-frequency inspections across the entire network. While track inspection vehicles can measure static geometric parameters such as gauge, level, alignment, and elevation using inertial reference systems and calculate the Track Quality Index (TQI), their inspections are typically conducted during non-operational periods or at specific speeds, making it difficult to accurately reflect the dynamic response of trains at actual operating speeds. Furthermore, traditional geometric inspections often analyze the amplitude exceeding limits of a single geometric parameter in isolation, ignoring the phase relationship of different geometric parameters (such as alignment and level irregularities) in spatial distribution. In actual operation, even if no geometric index exceeds limits independently, if the phase relationship between the two is in a specific "malicious coupling" state, it can still induce strong low-frequency lateral swaying through a superposition effect. This hidden composite irregularity is often missed by traditional evaluation systems.

[0004] To overcome the limitations of static detection, dynamic monitoring technology based on onboard acceleration data has gradually become a research hotspot. Existing technologies attempt to collect vibration signals using acceleration sensors installed on operating trains, and then assess track conditions through signal processing or model inversion. For example, CN110728000A discloses a method for inverting track irregularities based on vehicle vibration response. This method establishes a vehicle-track coupled dynamic model and uses the vehicle vibration response to estimate the geometric irregularities of the track. However, such existing technologies still face the challenge of identifying "same phenomenon, different source" in practical applications. Specifically, most existing inversion methods focus on reconstructing "geometric shape and position," lacking effective diagnosis of track "structural performance" (such as fastener stiffness and track bed elasticity). In fact, the stiffness degradation of the track fastener system can also lead to abnormal wheel-rail forces, causing vehicle swaying, and the resulting vibration characteristics are easily confused with geometric irregularities.

[0005] Furthermore, subway systems commonly experience low-frequency lateral swaying of 0.2–3 Hz, corresponding to relatively long wavelengths (tens of meters) and often involving complex train-track coupling mechanisms. Existing monitoring methods lack a mechanism to combine "dynamic response data" with the line's "historical maintenance records" (such as fastener replacement and bolt re-tightening records), failing to construct a complete causal chain from "vibration phenomena" to "structural deterioration." This makes it difficult for maintenance personnel to distinguish whether swaying alarms are primarily caused by geometric deviations or support stiffness degradation, hindering accurate decisions on whether geometric fine-tuning or structural reinforcement is necessary, thus restricting the improvement of preventative maintenance levels in rail transit.

[0006] Furthermore, on the one hand, there are differences in understanding among those skilled in the art; on the other hand, the applicant studied a large number of documents and patents when making this invention, but due to space limitations, not all details and contents were listed in detail. However, this does not mean that the present invention does not possess the features of these prior art. On the contrary, the present invention already possesses all the features of the prior art, and the applicant reserves the right to add relevant prior art to the background art. Summary of the Invention

[0007] In view of the shortcomings of the existing technology, the present invention provides a method and system for diagnosing the causes of track swaying sections based on multi-source data fusion, so as to solve at least some of the above-mentioned technical problems.

[0008] This invention discloses a method for diagnosing the causes of track swaying sections based on multi-source data fusion, comprising the following steps: acquiring on-board vibration response data and track static geometry data of the target line, and performing spatiotemporal alignment processing; performing statistical analysis on the on-board vibration response data to identify vibration anomaly sections; extracting the track-direction irregularity component and horizontal irregularity component from the track static geometry data corresponding to the vibration anomaly sections, and calculating the phase correlation characteristic value between the two; constructing a dynamic model including track stiffness parameters, and performing parameter inversion on the dynamic model with the goal of minimizing the residual between the simulated vibration response output by the dynamic model and the on-board vibration response data, to obtain the equivalent stiffness degradation parameter of the vibration anomaly section; and determining the swaying cause type of the vibration anomaly section based on the phase correlation characteristic value and the equivalent stiffness degradation parameter.

[0009] This invention effectively solves the diagnostic challenge of "same image, different source" problems faced by single data sources by integrating on-board vibration response and track static geometry data, and introducing phase correlation feature calculation and dynamic parameter inversion based on spatiotemporal alignment. While existing technologies (such as CN110728000A) can invert track state using vibration, they are often limited to geometric position reconstruction, making it difficult to distinguish between excitation caused by the superposition of geometric waveform phases and excitation caused by the physical degradation of fastener stiffness. This solution uses phase correlation feature values ​​to quantify the degree of spatial coordination between track alignment and horizontal irregularities, identifying hidden defects where geometric amplitudes are not exceeded but phase coupling is malignant. Simultaneously, by minimizing the residual between the simulated response output by the dynamic model and the measured vibration response as the inversion objective, the hidden structural stiffness degradation parameters can be analyzed based on the apparent vibration signal. This two-dimensional diagnostic logic decouples the causes of abnormal vibration sections, significantly reduces the waste of maintenance resources caused by misjudging defect types, and compensates for the deficiency of traditional TQI evaluation systems in reflecting structural dynamic performance.

[0010] According to a preferred embodiment, when acquiring the vehicle vibration response data and track static geometry data of the target line, the method also includes acquiring the historical maintenance record data of the target line; the dynamic model is a time-varying model of fastener stiffness, which includes maintenance intervention compensation items determined based on the historical maintenance record data.

[0011] This invention innovatively introduces a maintenance intervention compensation term based on historical maintenance records into the dynamic inversion model, overcoming the "ill-conditioned ambiguity" problem commonly found in pure data-driven inversion algorithms. In rail transit operation and maintenance practice, relying solely on vibration data to invert stiffness parameters (as described in CN110728000A) often lacks boundary constraints, leading to significant uncertainty in stiffness value estimation. This solution transforms discrete, management-level maintenance events such as "re-tightening and replacement" into gain factors or boundary conditions in continuous dynamic equations, providing explicit physical constraints for optimization algorithms such as particle swarm optimization, thereby greatly reducing the solution space for parameter optimization. This not only improves the convergence speed of the inversion results but also ensures that the obtained time-varying stiffness curve conforms to objective physical laws, enabling accurate quantitative assessment of the health status of the fastener system based solely on onboard data and records, even without ground sensor monitoring.

[0012] According to a preferred embodiment, the phase correlation feature value includes the inverse phase difference; calculating the phase correlation feature value between the two includes: performing time-frequency signal decomposition on the track irregularity component and the horizontal irregularity component respectively, extracting feature signals within a preset wavelength range; calculating the instantaneous phase of the feature signals respectively using Hilbert transform, and calculating the absolute value of the difference between the two instantaneous phases to obtain the inverse phase difference.

[0013] This invention extracts the instantaneous phase of characteristic signals using Hilbert transform and calculates the inverse phase difference, constructing a "phase microscope" for track composite irregularities. This scheme utilizes this mathematical tool to process two independent signals—track alignment and horizontal irregularities—revealing their phase coupling mechanism in the spatial domain. By decomposing time-frequency signals to filter out high-frequency noise interference, it focuses on signal analysis in specific bands and leverages the Hilbert transform's analytical capability for the instantaneous phase of non-stationary signals to capture transient phase-locking phenomena that are difficult to detect using conventional statistical methods (such as cross-correlation functions). This processing method can accurately identify the superposition effect of lateral forces caused by "phase reversal" between track alignment and horizontal irregularities, providing a mechanistic criterion for explaining the perplexing phenomenon of "normal geometric indicators but severe on-site train shaking."

[0014] According to a preferred embodiment, based on the train design operating speed and the lateral resonance frequency of the car body on the target line, a preset wavelength range is dynamically determined when the reverse phase difference falls within a preset phase-locking interval. When it is determined that there is a compound non-cooperative collaborative excitation, in which The preset angle threshold is used. Preferably, the preset wavelength range can be set to 25 meters to 35 meters.

[0015] This invention further locks the analysis wavelength within the range of 25 meters to 35 meters, and sets... The phase-locked range enables precise targeting of low-frequency lateral swaying causes. According to vehicle dynamics principles, the wavelength range of 25 to 35 meters is typically the sensitive spatial band for low-frequency lateral swaying of the track, easily triggering strong lateral resonance responses in the vehicle body; while the phase difference is close to... (i.e., 180 degrees) means that the centrifugal force component generated by track irregularities is in the same direction as the gravitational force component generated by horizontal irregularities, forming a combined force enhancement effect of "1+1>2". This scheme, by clearly defining this physical threshold, transforms fuzzy qualitative analysis into precise quantitative judgment, effectively eliminating the interference of random phase differences, ensuring a high degree of confidence in the identification of "strongly coupled" composite irregularities, and thus guiding the engineering department to carry out precise maintenance operations for specific bands.

[0016] According to a preferred embodiment, statistical analysis is performed on the vehicle vibration response data to identify abnormal vibration sections. Specifically, this includes: using an adaptive sliding window algorithm to calculate the peak acceleration density per unit distance; determining continuous mileage where the peak acceleration density exceeds a preset alarm threshold as an abnormal vibration section; wherein the length of the adaptive sliding window changes positively with the train's operating speed.

[0017] This invention employs an adaptive sliding window algorithm that positively correlates with train speed to calculate peak acceleration density, resolving the issue of inconsistent spatial resolution in vibration feature extraction under variable speed conditions. Traditional fixed-length windows cover varying actual track lengths as trains accelerate or decelerate, leading to biased assessments of track defects. This solution establishes a dynamic mapping between window length and speed, ensuring a constant track length for each analysis window across different speed levels (e.g., entering / exiting stations or speed-limited sections), thus guaranteeing the comparability of peak density indices across the entire line. Furthermore, using peak density instead of a single maximum value as the criterion effectively distinguishes between accidental wheel-rail impacts and continuous track defects, improving the robustness of abnormal section identification and avoiding false alarms caused by isolated singularities.

[0018] According to a preferred embodiment, parameter inversion is performed on the dynamic model to obtain the equivalent stiffness degradation parameters of the vibration abnormal section, including: constructing a time-varying model of fastener stiffness with initial stiffness, stiffness attenuation coefficient and nonlinear exponent as undetermined parameters; iteratively adjusting the undetermined parameters using an optimization algorithm to minimize the residual between the simulated vibration response output by the dynamic model and the vehicle vibration response data; and using the stiffness attenuation coefficient corresponding to the minimized residual as the equivalent stiffness degradation parameter.

[0019] This invention achieves mathematical reconstruction of the fastener stiffness degradation trajectory by constructing a parameterized model including initial stiffness, attenuation coefficient, and nonlinear exponent, and minimizing the residual using an optimization algorithm. Unlike static evaluations that only focus on the current state, this approach introduces an attenuation coefficient and a nonlinear exponent, enabling the model to describe the dynamic evolution of stiffness over time (or load). By iteratively adjusting these undetermined parameters, the simulated vibration response output by the model approximates the measured data. Essentially, this is a process of parameter inversion and state identification of the physical model using observational data. This method can extract microscopic fastener node stiffness information from the macroscopic vehicle body vibration response, quantifying the aging degree or loosening trend of the fastener material.

[0020] According to a preferred embodiment, determining the type of swaying in abnormal vibration sections includes calculating a comprehensive swaying index. The following steps are taken: First, a normalized component determined based on the vibration amplitude of the abnormal vibration section; second, a normalized component determined based on the phase correlation feature value; and third, a normalized component determined based on the equivalent stiffness degradation parameter. A preset feature fusion algorithm is then used to perform a dimensionless comprehensive evaluation of the first, second, and third normalized components to obtain the comprehensive vehicle sway index. Specifically, the overall swaying index The calculation formula is as follows: ,in, This is the first normalized component determined based on the vibration amplitude of the abnormal vibration section. This is the second normalized component determined based on phase correlation eigenvalues. The third normalized component is determined based on the equivalent stiffness degradation parameter. , , These are the corresponding weighting coefficients.

[0021] This invention constructs a comprehensive sway index by introducing weighting coefficients. This approach achieves dimensionless fusion and evaluation of heterogeneous monitoring data. Vibration amplitude (dynamic indicator), phase correlation value (geometric indicator), and stiffness degradation parameters (structural indicator) involved in track maintenance belong to completely different physical dimensions, making direct comparison meaningless. This solution maps multi-dimensional features into a unified evaluation space through normalization and weighted summation. This comprehensive index not only reflects the current severity of track swaying but also implicitly reveals the causes of defects, providing maintenance personnel with an intuitive and comprehensive score of track health status. This avoids the potential bias of single-indicator evaluations and helps to grasp the operational quality of the track from a holistic perspective.

[0022] According to a preferred embodiment, the types of swaying causes include composite irregularity-dominated type, support stiffness deterioration type, and coupled deterioration type; determining the type of swaying cause in abnormal vibration sections specifically includes: in the comprehensive swaying index If the threshold is exceeded, compare and The percentage of contribution; if Greater than 0 and The largest proportion and If the first preset condition is met, it is determined to be a compound irregularity-dominant type; if The largest proportion and If the second preset condition is met, it is determined to be a support stiffness degradation type; if If the values ​​are greater than 0 and the difference between their proportions is less than a preset value, and both satisfy their respective preset conditions, it is determined to be a deteriorating coupling type; if =0 and If the second preset condition is not met, it is determined to be a single vibration exceeding the limit type.

[0023] This invention establishes a logically rigorous decision tree for classifying the causes of vehicle swaying based on the contribution ratio of each normalized component, achieving automated and accurate classification of fault types. By comparing the weights of geometric coupling components and structural stiffness components, it can clearly determine whether the current vehicle swaying is dominated by external geometric excitation, internal structural support failure, or a vicious coupling of the two. This classification mechanism is directly related to subsequent maintenance strategies: composite irregularities require geometric fine-tuning, support stiffness degradation requires fastener replacement or re-tightening, and deteriorating coupling requires comprehensive remedial measures. This technology directly solves the pain point of existing technologies that "only know the vehicle is swaying, but don't know the cause," significantly improving the pertinence and effectiveness of maintenance decisions.

[0024] According to a preferred embodiment, the method further includes: calculating a vibration response gain factor based on equivalent stiffness degradation parameters. This is used to characterize the amplification factor of stiffness degradation on vibration amplitude; a geometric degradation cumulative function is established. It is used to characterize the cumulative destructive effect of vibration excitation on track geometry; combined with the vibration response gain factor and the geometric degradation cumulative function, a diagnostic report containing maintenance level recommendations is generated.

[0025] This invention establishes a correlation mechanism between the vibration response gain factor and the geometric degradation cumulative function, thereby achieving predictive diagnostic capabilities and revealing the closed-loop evolution of track defects, from stiffness failure and vibration amplification to geometric deterioration. The vibration response gain factor quantitatively characterizes the amplification effect of insufficient stiffness on the dynamic response, while the geometric degradation cumulative function describes how this amplified dynamic effect, in turn, accelerates the destruction of track geometry. By integrating these two aspects of analysis, the generated diagnostic report can not only identify current defects but also provide early warnings of potential geometric deterioration trends. This guides maintenance departments to intervene in maintenance at an early stage, before geometric parameters exceed limits but before structural stiffness has significantly decreased, breaking the vicious cycle of defects and truly achieving full life-cycle track health management.

[0026] The present invention also discloses a track swaying section cause diagnosis system based on multi-source data fusion, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, it implements the steps of the aforementioned method. Attached Figure Description

[0027] Figure 1 This is a hardware connection diagram of a causal diagnosis system according to a preferred embodiment of the present invention; Figure 2 This is a flowchart of the steps of a preferred embodiment of the cause diagnosis method provided by the present invention.

[0028] List of reference numerals 100: Computer equipment; 200: Dynamic detection data acquisition equipment; 300: Static geometric data acquisition equipment; 400: Vehicle-mounted acceleration sensor; 500: Global positioning system module; 600: Track geometry detection equipment; 700: Railway maintenance management system. Detailed Implementation

[0029] The following is a detailed explanation with reference to the accompanying drawings.

[0030] Example 1 This invention proposes a method and system for diagnosing the causes of track swaying sections based on multi-source data fusion. Its core lies in constructing a three-dimensional diagnostic process integrating perception, analysis, and attribution, aiming to solve the swaying phenomenon in urban rail transit caused by poor track geometry or deteriorated structural support performance, particularly addressing the challenge of diagnosing heterogeneous sources of the same phenomenon in low-frequency lateral swaying. At the perception level, this solution achieves deep fusion of multi-source heterogeneous data by simultaneously acquiring onboard triaxial acceleration, GPS location information, track geometry detection data, and historical maintenance data including fastener replacement or bolt re-tightening records, providing comprehensive data support for subsequent diagnosis. At the analysis level, the system first uses a sliding window method dynamically adjusted with operating speed to accurately locate swaying sections by calculating the peak acceleration density per unit distance. Then, wavelet packet decomposition and Hilbert transform are used to extract the dominant wavelength components of the track alignment and horizontal irregularities. By calculating the inverse phase difference between the two, the enhanced coupling effect is identified, thereby revealing the mechanism by which composite irregularities affect vehicle dynamics. At the causal level, this invention introduces a time-varying model of fastener stiffness, combining vehicle-track coupled dynamics simulation and on-site vibration response data. It utilizes a particle swarm optimization algorithm to invert the stiffness degradation rate of fastener nodes, thereby establishing a closed-loop feedback quantification system from vibration excitation, geometric instability, structural damage to vibration aggravation. Through the synergistic effect of the above three-dimensional process, this solution can accurately distinguish different causes of vehicle sway, such as geometric irregularity-dominated, support stiffness degradation, and coupling deterioration types, achieving a leap from phenomenon perception to mechanism diagnosis, and providing a scientific decision-making basis for preventive maintenance of rail transit.

[0031] This embodiment discloses the hardware operating environment and overall architecture of a track swaying section cause diagnosis system based on multi-source data fusion. For example... Figure 1 As shown, the system includes a computer device 100. At the hardware level, the computer device 100 includes a processor, a memory, a communication interface, and a display control unit interconnected via a system bus. The memory stores computer programs and various detection data, and the processor executes the computer programs in the memory to implement the steps of the track sway section cause diagnosis method described in subsequent embodiments. Depending on the actual application requirements, the computer device 100 can be configured as a ground data center server, a cloud processing platform, or an embedded industrial control terminal integrated into a train-mounted detection platform and trackside monitoring network.

[0032] Computer equipment 100 establishes communication connections with dynamic detection data acquisition equipment 200, static geometric data acquisition equipment 300, and the track maintenance management system 700 (which serves as an external database) via communication interfaces. Regarding dynamic detection data acquisition, the communication interface of dynamic detection data acquisition equipment 200 is connected to an onboard accelerometer 400 and a global positioning system module 500. The onboard accelerometer 400 is installed at key locations on the train body or bogies, preferably at the bottom of the third car in the train formation. The onboard accelerometer 400 is a three-axis microelectromechanical system accelerometer with a sampling frequency set to 200Hz or higher, used to acquire lateral vibration signals of the train body. The global positioning system module 500 is used to synchronously record the train's spatial location information to achieve temporal and spatial alignment of vibration data with geographic coordinates.

[0033] In terms of static geometric data acquisition, the communication interface of the static geometric data acquisition device 300 is connected to the track geometry detection device 600. The track geometry detection device 600 includes a large track inspection vehicle or an on-board inertial navigation system, used to provide track geometry position data with a spatial resolution of 0.25m. The track geometry position data includes track orientation deviation, horizontal deviation, track gauge deviation, and elevation deviation.

[0034] The track maintenance management system 700 is used to provide historical maintenance record data for the line. The historical maintenance record data includes fastener replacement records, bolt re-tightening time, and records of partial missing parts. The historical maintenance record data is used as the boundary conditions for constructing a time-varying model of fastener stiffness.

[0035] The processor is configured to perform weighted fusion processing on the aforementioned multi-source data, executing computational steps including peak acceleration density calculation, composite irregularity phase analysis, and fastener stiffness inversion. The display control unit is configured to generate a visual interactive interface, preferably overlaying diagnostic results and suggested maintenance levels onto a geographic information system map.

[0036] Example 2 This embodiment is a further improvement on embodiment 1, and repeated content will not be described again.

[0037] This embodiment details a method for diagnosing the causes of track swaying sections based on multi-source data fusion. Preferably, as follows... Figure 2 As shown, the method may include the following steps: Acquire vehicle vibration response data and track static geometry data of the target line, and perform spatiotemporal alignment processing; Statistical analysis was performed on the vehicle vibration response data to identify abnormal vibration sections; Extract the track-direction irregularity component and the horizontal irregularity component from the track static geometry data corresponding to the vibration anomaly section, and calculate the phase correlation characteristic value between the two. A dynamic model including track stiffness parameters is constructed. With the goal of minimizing the residual between the simulated vibration response output by the dynamic model and the on-board vibration response data, the parameters of the dynamic model are inverted to obtain the equivalent stiffness degradation parameters of the vibration anomaly section. Based on the phase correlation eigenvalues ​​and the equivalent stiffness degradation parameters, the type of swaying caused by the abnormal vibration section is determined.

[0038] Furthermore, the method may specifically include the following steps: S1. Collect multi-source data such as vehicle acceleration, track geometry and maintenance records, and perform spatiotemporal alignment preprocessing; S2. Calculate peak acceleration density based on adaptive sliding window to initially screen potential swaying sections; S3. Use wavelet packet decomposition and Hilbert transform to extract the inverse phase difference features of the orbital and horizontal irregularities; S4. Construct a time-varying model of fastener stiffness and invert the stiffness degradation parameters of the fastener system based on the measured vibration response. S5. Establish a two-way feedback mechanism for vibration and structural deterioration, and output a cause classification diagnosis report and maintenance suggestions.

[0039] For step S1, the processor first acquires multi-source heterogeneous data through the communication interface, specifically including onboard acceleration data, track geometry detection data, and historical track maintenance records. The onboard acceleration data is collected by a triaxial accelerometer installed on the car body or bogie at a sampling frequency of 200Hz or higher, used to characterize the vehicle's lateral vibration response. The track geometry detection data comes from a track inspection vehicle or onboard inertial navigation system, with a spatial resolution set to 0.25m, and includes track alignment, level, gauge, and elevation geometry parameters. During data processing, the processor uses the timestamps and odometers of the Global Positioning System (GPS) to spatially match the time-domain acceleration signals with the spatial-domain track geometry data, achieving synchronous alignment of the multi-source data and laying the data foundation for subsequent analysis.

[0040] Specifically, the processor first establishes a communication connection with the multi-source heterogeneous data acquisition terminal through a communication interface to synchronously acquire basic detection information, including vehicle dynamic response data, track static geometry data, and track facility operation and maintenance data. Regarding the acquisition of vehicle dynamic response data, the system utilizes an inertial measurement unit (IMU) installed at a specific location on the train for real-time monitoring. Preferably, the IMU employs a three-axis microelectromechanical system (MEMS) accelerometer, which is configured at the bottom of the third car in the train formation or installed on a key component of the train bogie, designed to directly pick up lateral vibration signals of the car body or running gear during operation. To ensure effective capture of transient impact responses and meet the accuracy requirements of subsequent frequency domain analysis, the sampling frequency of the accelerometer is set to no less than 200Hz. Simultaneously, the data acquisition terminal also includes a Global Positioning System (GPS) module, which operates synchronously with the accelerometer to record the timestamp information and geographic latitude and longitude coordinates of the train's trajectory in real time, thereby assigning precise spatial location labels to the time-domain vibration signals.

[0041] In acquiring track static geometry data, the processor reads data files from the track inspection vehicle or onboard inertial navigation measurement system via a dedicated interface. The track geometry data encompasses key geometric parameters reflecting track smoothness, specifically including track alignment deviation, horizontal deviation, gauge deviation, and elevation deviation. Considering the relatively long wavelengths of low-frequency swaying in subway lines, the spatial sampling resolution of the track geometry inspection data is set to 0.25m. This high-precision spatial resolution ensures accurate reconstruction of composite irregularities in the 25-35 meter frequency band during subsequent analysis. Furthermore, this type of data is typically updated according to a preset period (e.g., monthly) to reflect the periodic trends in track geometry.

[0042] Regarding the acquisition of facility operation and maintenance data, the processor accesses the 700 database of the engineering management system to retrieve maintenance records of the target line within a preset historical period (e.g., the past five years). These maintenance records primarily include replacement records of the fastener system, records of the re-tightening of connecting bolts, and inspection records of missing fasteners. This data is not merely for archiving; it serves as a key boundary condition for subsequently constructing a time-varying model of fastener stiffness. The "time of the most recent replacement or tightening" will be used to quantify the restorative effect of maintenance intervention on structural stiffness.

[0043] After acquiring the aforementioned multi-source data, the processor performs rigorous data preprocessing and spatiotemporal alignment operations. Since acceleration data is acquired based on time series data, while track geometry data and maintenance records are based on spatial mileage distribution, there is a dimensional difference between the two. Therefore, the processor establishes a mapping relationship using timestamps and mileage marker data recorded by the Global Positioning System (GPS), employing linear interpolation or spline interpolation algorithms to accurately map the time-domain acceleration signal to the track spatial coordinate system, achieving synchronous alignment of the time and spatial dimensions. For the spatially highly discrete maintenance record data, the processor uses a spatial mapping algorithm based on the radius of influence to transform maintenance events (such as fastener replacement and bolt re-tightening) at isolated mileage points into continuous spatial distribution functions. Specifically, centered on the mileage point where the maintenance event occurs, a Gaussian kernel function with a preset width (preferably matching a sensitive wavelength of 25m~35m) is used for smoothing, ensuring that discrete maintenance gains can be mapped to the corresponding mileage segment, providing physically continuous boundary conditions for subsequent dynamic inversion within the sliding window. Furthermore, in order to eliminate environmental noise and focus on the characteristic frequency band of the swaying motion, the processor is also configured to perform frequency domain filtering on the original acceleration signal, in particular to retain the low-frequency lateral vibration components in the 0.2Hz to 3Hz frequency band, thereby providing a high-quality data foundation for subsequent steps to identify abnormal swaying motion caused by long-wave irregularities and structural stiffness degradation.

[0044] For step S2, the processor uses a sliding window algorithm to calculate the peak acceleration density per unit distance to identify potential swaying sections. The length of the sliding window is dynamically adjusted according to the designed operating speed of the line: when the train's operating speed... When the speed is less than or equal to 60 km / h, the sliding window length is set to 50 m; when the operating speed When the speed is greater than 60km / h and less than or equal to 80km / h, the sliding window length is set to 75m; when the operating speed When the speed is greater than 80 km / h, the sliding window length is set to 100 m. Within the defined window length, the processor counts the number of standard peak values ​​that exceed three times the root mean square (RMS) value, defines this number as the acceleration peak density, and uses this density to filter out abnormal sections.

[0045] In step S2, the processor performs initial screening and positioning operations for the swaying section based on an adaptive sliding window.

[0046] Specifically, after the processor acquires the onboard acceleration data that has undergone spatiotemporal alignment preprocessing, it first performs frequency domain feature extraction. Given that swaying in urban rail transit environments typically manifests as low-frequency oscillations with longer wavelengths, rather than high-frequency impacts generated by wheel-rail interaction, the processor is configured to use a bandpass filter to filter the raw lateral acceleration signal. The passband frequency range of the bandpass filter is set to 0.2 Hz to 3 Hz, aiming to filter out high-frequency interference components caused by short-wave irregularities in the track or vehicle mechanical noise, thereby accurately extracting the low-frequency lateral vibration energy that characterizes the lateral stability of the vehicle body and passenger comfort. This step effectively solves the problems of low signal-to-noise ratio and high false alarm rate caused by full-band analysis in existing technologies, ensuring that subsequent analysis focuses on low-frequency swaying characteristics that pose a substantial hazard.

[0047] Based on the extracted low-frequency lateral vibration signal, the processor introduces a dynamic sliding window algorithm to perform segmented scanning and statistical analysis of the entire line data. The coverage length of the sliding window is not a fixed value, but is adaptively adjusted according to the real-time operating speed of the train in different sections to match the spatial wavelength characteristics of the vehicle vibration response under different speed conditions, ensuring that the detection results have normalized comparability at different operating speeds. The processor reads the GPS speed information or train network speed signal corresponding to the current acceleration data and determines the length of the current sliding window according to a preset speed classification strategy. Specifically, when the train speed is less than or equal to 60 km / h, the processor sets the sliding window length to 50 meters; when the train speed is greater than 60 km / h but less than or equal to 80 km / h, the sliding window length is set to 75 meters; and when the train speed is greater than 80 km / h, the sliding window length is set to 100 meters. This variable window length strategy fully considers the speed differences of the train when entering and leaving stations, turning, and running straight, ensuring that each window contains a sufficient number of complete vibration cycles, avoiding feature truncation due to excessively short windows or the averaging and masking of local abnormal features due to excessively long windows.

[0048] Within a defined sliding window, the processor performs statistical calculations of the peak acceleration density to quantify the intensity of vibration in that segment. The processor first calculates the root mean square (RMS) value of the lateral acceleration signal within the current window, which characterizes the fundamental vibration energy level of that segment. Then, the processor sets a dynamic judgment threshold, which is three times the RMS value. The processor iterates through the acceleration waveform data within the current window, counting the number of standard peaks whose amplitude exceeds the dynamic judgment threshold. This statistical number is defined as the peak acceleration density, and its physical significance lies in eliminating occasional random impact interference and identifying abnormal vibration sequences with persistent and periodic characteristics, thereby distinguishing between sudden interference and substantial vehicle swaying.

[0049] Finally, the processor compares the calculated peak acceleration density with a preset alarm threshold to identify potential swaying sections and determine their start and end mileage. Based on operational and maintenance experience of actual lines and historical data analysis, the alarm threshold is preferably set at 1.8 peaks per 10 meters. When the peak acceleration density of a certain section consistently exceeds this alarm threshold, the processor marks that continuous mileage as a suspected swaying section and generates a swaying heatmap. Taking the detection of an actual operating line as an example, the processor uses a 75-meter sliding window corresponding to an average speed of 70 km / h for analysis. Through the above algorithm, it successfully identified the section from mileage marker K12+300 to K12+600 as an abnormally high-value area, where the peak density consistently exceeds the standard of 1.8 peaks per 10 meters. The system then records the starting and ending mileage markers of this section, limiting subsequent complex causal analysis to a specific spatial range. This processing method significantly reduces the computational load of data processing and greatly improves the detection rate for hidden swaying sections.

[0050] In step S3, the processor analyzes the selected swaying sections using inverse phase-locked loop technology. The processor performs wavelet packet decomposition on the directional and horizontal irregularity sequences of this section, extracting feature components within the target wavelength range. Subsequently, the processor uses Hilbert transform to calculate the instantaneous phase angles of these feature components, denoted as the directional irregularity phases. and horizontal unsmooth phase Based on this, the processor uses the formula Calculate the inverse phase difference between the two. When the calculated reverse phase difference Falling into the range At that time, the processor determines that there is an inverse reinforcement effect between the directional and horizontal irregularities in the section, that is, there is a combined irregularity co-excitation.

[0051] In step S3, the processor can perform refined extraction and analysis of composite irregularity features based on reverse phase locking technology on the suspected swaying sections screened and locked by the previous steps. This step aims to reveal the deep frequency domain and phase correlation between track geometry deviation and lateral sway of the vehicle body.

[0052] Specifically, the processor first retrieves the corresponding static geometric position data segments from the track geometry detection dataset pre-stored in the database, based on the start and end mileage markers of the suspected swaying section. The static geometric position data mainly includes the track-direction unevenness sequence and the horizontal unevenness sequence. These two types of geometric parameters are identified as the main external excitation sources causing the lateral swaying and roll coupling motion of the vehicle body.

[0053] To overcome the time-domain information loss defect of traditional Fourier transform when processing non-stationary and nonlinear track irregularity signals, the processor is configured to use wavelet packet decomposition algorithm to perform multi-scale time-frequency analysis on the track-direction irregularity sequence and the horizontal irregularity sequence in the aforementioned spatial domain. In terms of specific algorithm parameter settings, to balance computational efficiency and frequency resolution of feature extraction, the processor preferably uses the db4 wavelet basis function and sets the decomposition level to 5. Through this specific wavelet packet decomposition operation, the processor can effectively decompose the original irregularity signal into a series of sub-signals with different frequency bands and accurately reconstruct the target component containing specific dominant wavelength features. Based on the analysis of low-frequency sway characteristics of urban rail transit (such as subways) in this embodiment of the invention, the processor specifically identifies and extracts frequency band components with wavelengths between 25 meters and 35 meters as feature components. The selection of this wavelength band is based on the fact that when the train is running at a normal speed (e.g., 70 km / h), the frequency of the irregularity excitation in this wavelength range falls exactly between 0.2 Hz and 3 Hz. This frequency range has been proven to be a key sensitive frequency band that induces low-frequency lateral resonance in the vehicle body and causes motion sickness in passengers.

[0054] After acquiring the directional and horizontal irregularity characteristic components within the target wavelength range, the processor further uses Hilbert transform to analyze these two real-valued signals to obtain their instantaneous phase properties. Specifically, the processor performs a Hilbert transform on the directional irregularity characteristic components, constructs an analytic signal, and calculates its instantaneous phase angle, defining it as the directional irregularity phase, denoted as . Similarly, the processor performs a Hilbert transform on the horizontal irregularity feature component, calculates its instantaneous phase angle, and defines it as the horizontal irregularity phase, denoted as . The technical purpose of this step is to transform the physical morphological characteristics of orbital geometric irregularities into mathematically comparable phase angles, thereby providing a quantitative basis for analyzing the synergistic relationship between the spatial distribution of orbital and horizontal irregularities.

[0055] Based on this, the processor executes the logic for calculating and determining the inverse phase difference. The processor follows the formula... Real-time calculation of the absolute value of the phase difference between the alignment irregularity and the level irregularity at the same mileage point This phase difference reflects the synchronicity or antagonism of the two irregularity excitations in spatial distribution. To identify the existence of an "anti-phase enhancement mode" that degrades vehicle dynamics, the processor sets a critical judgment interval based on anti-phase. Based on experimental analysis and simulation verification of vehicle-track coupled dynamics, when the phase difference between the track-oriented irregularity and the horizontal irregularity is close to... At a radius of 180 degrees, the dynamic effects of the two elements will superimpose rather than cancel each other out, resulting in the most unfavorable combination condition. Based on different degrees of coupling effect, the processor has established a graded judgment standard: when the reverse phase difference... When a reverse compounding trend is detected, a primary alarm is triggered; to accurately identify strong coupling characteristics, the processor will determine the threshold. The preferred setting is 15 degrees, which constitutes the determination interval. (The corresponding angle range is 165 degrees to 195 degrees), which serves as the core basis for determining the "inverted enhancement mode".

[0056] When the calculated reverse phase difference When the vehicle falls within the aforementioned judgment range, for example, in this embodiment, the calculated phase difference is approximately 168 degrees. The processor determines that there is a strong reverse phase-locked relationship between the directional irregularities and the horizontal irregularities in this section. This relationship indicates that, at a specific spatial location, the directional deviation and horizontal tilt deviation of the track form a combined irregularity synergistic excitation effect, resulting in a significant amplification of the lateral disturbance force experienced by the vehicle when passing through this section. The processor then marks this section as a "composite irregularity-dominated" or "strongly coupled" risk area and directly correlates the calculated phase difference value and wavelength characteristics to the subsequent cause diagnosis report. This clearly excludes the traditional judgment logic of exceeding the limit of a single geometric parameter, achieving accurate tracing of the complex vehicle swaying mechanism.

[0057] For step S4, the processor evaluates the support performance of the track structure by constructing a time-varying model of fastener stiffness. The expression for the time-varying model of fastener stiffness is as follows: In this formula, Represents the initial stiffness. Represents the attenuation coefficient. Represents a nonlinear exponent. Represents the maintenance intervention gain factor. This is a time indication function for the most recent replacement or tightening. The processor uses a particle swarm optimization (PSO) algorithm, with the objective function of minimizing the residual between the simulated vibration response output by the dynamic model and the vibration response data collected in the field, to iteratively optimize the model parameters and solve for the optimal parameter set. This allows for the quantification of the stiffness degradation rate of fastener nodes.

[0058] In step S4, the processor can perform a fastener stiffness assessment operation based on structural state inversion, aiming to establish a quantitative correlation between vehicle vibration response and track microstructure deterioration.

[0059] After completing the geometric feature extraction in the preceding steps, the processor further constructs or invokes a pre-defined time-varying model of the fastener system stiffness to analyze the causes of track swaying from a structural mechanics perspective. Given that the stiffness characteristics of the track fastener system exhibit a nonlinear degradation pattern over service time and are significantly affected by manual maintenance activities, the processor-constructed time-varying model of fastener stiffness adopts a composite function expression including natural aging terms and maintenance compensation terms, specifically characterized as follows: In this mathematical model, the parameters This represents the nominal stiffness value of the fastening system in its initial installation or ideal state. For typical subway integral track bed fasteners, this initial stiffness value is usually set to 60 kN / mm; variable Represents the actual operating time of the rail line; parameters Defined as the stiffness attenuation coefficient, it is used to characterize the material fatigue creep rate of fastener rubber pads or elastic clips under cyclic loading; parameters Defined as a nonlinear exponent, it describes the acceleration or deceleration trend of stiffness decay over time; parameters Defined as a maintenance intervention gain factor, used to quantify the contribution of tightening bolts or replacing parts to stiffness recovery; variables This is a time-dependent indicator function, the value of which is determined based on the historical maintenance records obtained in step S1, corresponding to the time point of the most recent fastener replacement or bolt re-tightening operation. To decouple discrete maintenance data from the continuous sliding window, the indicator function... In the current analysis window The comprehensive value within is determined by the following formula: ,in The window contains the mileage of discrete maintenance points. This corresponds to the weighting contribution factor. Through this mapping logic, the processor can transform isolated maintenance records into continuous parameters participating in the inversion of the dynamic equations, eliminating the logical gap between discrete events and continuous calculation windows, and ensuring that the inversion optimization algorithm can accurately capture the recovery characteristics of structural stiffness after maintenance intervention.

[0060] To solve for the unknown parameters in the above model and obtain the current actual fastener stiffness distribution, the processor is configured to run an inversion algorithm based on vehicle-track coupled dynamics. Specifically, the processor first uses multibody dynamics simulation software (such as VAMPIRE or UM software) to establish a refined vehicle-track spatial coupling model, and loads the actual track geometric irregularity data (i.e., track-direction and horizontal irregularities) extracted in step S3 as the external excitation input of the system into the model. Subsequently, the processor introduces a particle swarm optimization (PSO) algorithm for parameter optimization. During the optimization process, the processor uses the lateral vibration acceleration response data of the vehicle body collected on-site as the observation value, and uses minimizing the residual between the simulated vibration response output by the dynamic model and the observed value as the objective function to optimize the set of undetermined parameters in the model. As optimization variables, the processor iteratively adjusts the above parameter combinations to drive the dynamic simulation model to output simulated vibration response and calculates the residual between the simulated vibration response and the field-measured vibration response. When the residual converges to a preset minimum range, the corresponding parameter set is the optimal solution, thus realizing the reverse derivation of microscopic structural state parameters from macroscopic vibration data.

[0061] Based on the aforementioned inversion mechanism, the processor can accurately calculate the current equivalent stiffness value and degradation rate of each fastener node within the target section. As a specific application example of this embodiment, when analyzing a subway line section with an operating life of 12 years, the processor uses the above calculations to inversely solve for the average stiffness attenuation coefficient of that section. Approximately 0.08 / year, nonlinear exponent The value is approximately 1.3, indicating an accelerated degradation trend in the stiffness of the fasteners in this section. Furthermore, analysis of maintenance records revealed that the most recent comprehensive re-tightening operation in this section occurred three years ago, and there were instances of localized fastener loss rates reaching 12%, which corroborates the low stiffness results obtained from the inversion. Through this step, the system successfully mapped the simple physical vibration phenomenon into specific structural mechanical indicators, providing a decisive quantitative basis for subsequently determining whether the swaying was caused by "support stiffness degradation," thus effectively solving the technical challenge of establishing an evolutionary path from vibration response to structural deterioration in existing technologies.

[0062] For step S5, the processor establishes a two-way feedback dynamic model based on the above analysis results. This model contains a two-stage causal chain: the first stage characterizes the low-frequency lateral vibration of the car body (0.2~3Hz) transmitted to the track structure via the bogie, inducing accumulated lateral displacement of the sleepers, leading to increased track alignment and horizontal geometric deviations; the second stage characterizes the nonlinear decrease in the stiffness of the fastener system over service time, reducing the track's lateral bending stiffness, and thus amplifying the wavelength and amplitude of the original composite irregularities. Based on the combined geometric analysis and stiffness inversion results, the system generates a diagnostic report including causal classification, clarifying whether the swaying is caused by composite irregularities, weakened support stiffness, or deteriorating coupling, and outputs targeted maintenance recommendations, such as prioritizing fastener re-tightening or track fine-tuning.

[0063] In step S5, the processor can perform a closed-loop diagnostic and maintenance decision generation operation based on a bidirectional feedback dynamics model.

[0064] Specifically, based on the composite irregularity phase characteristics obtained in step S3 and the fastener stiffness distribution parameters obtained in step S4, the processor constructs or invokes a preset bidirectional feedback dynamic model. This model aims to quantify the nonlinear coupling relationship between the vehicle's dynamic response and the track structure's state decay, and its internal computational logic includes two continuous and mutually influential physical evolution processes.

[0065] First, the processor performs the first stage of "vibration-induced degradation" effect analysis, which assesses the cumulative destructive effect of car body vibration on track geometry. The processor uses the 0.2 Hz to 3 Hz low-frequency lateral vibration signal extracted in steps S1 and S2 as the system's input excitation source, and calculates the lateral excitation force distribution of this frequency band vibration energy transmitted to the track structure via the bogie using a multibody dynamics transfer function. Based on this, the processor introduces a geometric degradation accumulation function (represented in this embodiment as a state transition function S(t)) to characterize the evolution of track geometry. Specifically, its calculation formula is defined as... ,in, This refers to the dynamic increment of track geometry deviation (such as gauge or orientation deviation). To accumulate the number of train axles passing through, This is a degradation rate coefficient matrix calibrated based on historical data. This formula quantifies the cumulative lateral residual displacement at the sleeper-track interface under long-term alternating lateral loads, thus simulating the positive causal chain of "vibration leading to geometric deterioration." In a specific application scenario of this embodiment, the processor calculates using this model that the current vibration excitation level will cause the track gauge to deteriorate at a rate of 0.15 mm per 10,000 train axle passes, thus providing a quantitative basis for predicting the time of geometric parameter exceedance.

[0066] Secondly, the processor performs the second-stage "stiffness gain" effect analysis, which assesses the amplification effect of structural stiffness degradation on vibration response. The processor calls the time-varying model parameters of fastener stiffness determined in step S4 (including attenuation coefficients). With nonlinear exponent The residual lateral bending stiffness of the track is calculated for its current service life. To quantify the amplification effect of structural degradation on vibration, a vibration response gain factor is defined in the model. Its calculation expression is: ,in, For initial stiffness, The equivalent stiffness of the fastener at the current moment is obtained from the inversion in step S4. This is the empirical coupling coefficient (taken as 0.5 in this embodiment, but can be set manually). This characterizes the amplification factor of the vehicle's lateral vibration amplitude due to the reduced constraint capacity of the fastener system under the same geometrically irregular excitation input. Through this analysis, the processor can identify whether a feedback mechanism of "reduced structural stiffness leading to vibration amplification" exists. In the above embodiment scenario, the processor calculated that due to the significant attenuation of fastener stiffness, the current vibration response gain factor reached 1.7 times. This indicates that even if the amplitude of track geometric irregularities does not increase significantly, structural loosening alone is sufficient to cause the vehicle body sway amplitude to nearly double, thus revealing the underlying physical mechanism of vibration amplification.

[0067] Based on the quantitative analysis results from the two stages mentioned above, the processor intelligently classifies and determines the causes of vehicle swaying and generates a structured diagnostic report.

[0068] To achieve quantitative decision-making from multi-source heterogeneous data, the processor uses an improved weighted fusion model to calculate the comprehensive sway index S, as shown in the following formula: .in, , , These are the weighting coefficients, and In this embodiment, considering the directness of the vehicle body vibration response, it is preferable to set... ; To address the combined effects of orbital geometry, the following settings are made To address the long-term effects of structural degradation, the following measures are proposed: . The normalized peak acceleration density obtained in step S2 is calculated using the following formula: In the formula, This represents the number of standard peak values ​​exceeding three times the root mean square value within the current sliding window. The preset alarm threshold is (1.8 alarms / 10m in this embodiment, which can be manually set). When When the value is 1, it is set to ensure that the component is on the same dimensional scale as other normalization parameters; The composite irregularity index extracted in step S3 is calculated based on the reverse phase difference. The normalized value is specifically defined as follows: (when Valid only if in use; otherwise set to 0). The fastener stiffness degradation coefficient obtained in step S4 is defined as follows: ,in This is the equivalent stiffness value obtained from the current inversion.

[0069] Based on the calculation results of the comprehensive sway index S, this embodiment adopts the following hierarchical logic for cause diagnosis to solve the problem of one-sided judgment by a single indicator: The first step, overall index determination: real-time calculation of the comprehensive swaying index S, when... When the comprehensive alarm threshold is reached, it is determined that there is abnormal vehicle shaking in the section, and the cause classification procedure is triggered.

[0070] The second step is to trace and classify the contribution: (1) If (i.e., reverse phase difference) And weighted components It accounts for the largest proportion of the S value, and If the weighted components are not specified, it is determined to be "composite non-roughness-dominated type"; (2) if the weighted components are not specified, it is determined to be "composite non-roughness-dominated type"; It accounts for the largest proportion of the S value, and If, then it is determined to be "support stiffness deterioration type"; (3) if Furthermore, the difference in the proportions of the two components mentioned above is less than 10%, and both exceed 50% of their respective weights. This is combined with the vibration response gain factor calculated in step S5. If it is within the judgment interval, it is determined to be "coupling deterioration type"; (4) if the judgment interval is within the interval and This indicates that neither geometric coupling effects nor structural stiffness degradation are the primary causes, thus classifying it as a "single vibration exceeding limit type," corresponding to swaying caused by sporadic wheel-rail impacts or localized point-like irregularities. This revised logic introduces... The non-zero determination and fallback type eliminate logical gaps when variables suddenly become 0, ensuring the completeness of the decision tree across the entire parameter domain.

[0071] This logic organically integrates the phase-locking characteristics and stiffness inversion characteristics described in this invention, avoiding the one-sidedness of judging by a single index.

[0072] Ultimately, the processor outputs differentiated maintenance recommendations matching the diagnostic results through a visual interactive interface (such as a Geographic Information System, GIS). For "complex irregularities-dominated" types, it recommends prioritizing track fine-tuning to improve geometry; for "support stiffness deterioration" types, it recommends prioritizing fastener tightening, component replacement, or track bed repair; and for "coupled deterioration" types, it outputs high-priority comprehensive remediation instructions. This mechanism-based closed-loop diagnostic approach achieves a leap from simple phenomenon description to deep-seated problem tracing, effectively avoiding ineffective maintenance due to misjudgment and significantly improving the operation and maintenance efficiency and driving safety of rail transit infrastructure.

[0073] Example 3 This embodiment is a further improvement on the foregoing embodiment, and repeated content will not be described again.

[0074] This embodiment provides a track swaying section cause diagnosis system based on multi-source data fusion. This system is a device-based manifestation of the diagnosis method described in Embodiment 2. Each functional module runs in the hardware environment described in Embodiment 1 through computer program instruction code. The diagnostic system mainly includes a swaying identification module, a composite irregularity analysis module, a structural state inversion module, a cause diagnosis engine, and a visual interactive interface in its logical architecture.

[0075] The vehicle swaying identification module is configured to perform the initial screening step for swaying sections in the above method embodiments. Specifically, the module receives preprocessed vehicle acceleration data and uses an adaptive sliding window algorithm to calculate the peak acceleration density per unit distance. Based on the calculation results, the vehicle swaying identification module identifies potential swaying sections and generates a swaying heatmap reflecting the vibration distribution of the line, while simultaneously outputting the starting and ending mileage markers of the suspected swaying sections.

[0076] The composite irregularity analysis module is configured to perform the composite irregularity feature extraction step in the above method embodiments. For suspected sections screened by the sway detection module, this module extracts the track-direction irregularity sequence and the horizontal irregularity sequence from the track geometry detection data, and performs wavelet packet decomposition to extract feature components within the target wavelength range. Subsequently, the module uses inverse phase-locked loop technology to calculate the inverse phase difference between the feature components and determines whether a composite irregularity co-excitation effect exists based on a preset phase determination interval.

[0077] The structural state inversion module is configured to execute the structural state inversion steps in the above method embodiments. This module internally constructs a time-varying model of fastener stiffness, which includes key parameters such as initial stiffness, attenuation coefficient, nonlinear exponent, and maintenance intervention gain factor. The structural state inversion module uses a particle swarm optimization algorithm to iteratively optimize the model parameters, with the objective function of minimizing the residual between the simulated vibration response output by the dynamic model and the vibration response data collected on-site. This inversion process yields the stiffness degradation rate of each fastener node and generates a fastener stiffness distribution cloud map to quantify the support performance of the track structure.

[0078] The causal diagnosis engine is configured to execute the closed-loop feedback diagnosis and output steps in the above method embodiments. This module integrates a bidirectional feedback dynamic model to quantify the cumulative impact of vibration excitation on track geometric stability and the amplification factor of vibration amplitude due to structural stiffness degradation. Combining the calculation results from the composite irregularity analysis module and the structural state inversion module, the causal diagnosis engine classifies and determines the causes of swaying, generating a structured diagnostic report that includes conclusions such as composite irregularity-dominated type, support stiffness degradation type, or coupled deterioration type.

[0079] The visual interactive interface communicates with the cause diagnosis engine to intuitively display the system's analysis results. This interface is configured to support Geographic Information System (GIS) functions, enabling the overlay of vehicle sway heat maps, fastener stiffness distribution cloud maps, and specific fault location points onto an electronic map. Furthermore, based on the diagnosed cause type, the visual interactive interface can display targeted maintenance level suggestions, including priority maintenance instructions for fastener re-tightening or track fine-tuning, thereby assisting maintenance personnel in making informed decisions.

[0080] Example 4 This embodiment is a further improvement on the foregoing embodiment, and repeated content will not be described again.

[0081] This embodiment verifies the application effect of the track swaying section cause diagnosis system and method proposed in this invention in a practical engineering project, combined with a specific application scenario. The application scenario uses a subway line 6, which has been in operation for 12 years, as the test object. This line has recently received frequent complaints from passengers about severe "side-to-side swaying" and discomfort when trains travel through specific sections. However, routine inspections conducted by the maintenance department using conventional track inspection vehicles show that while the peak values ​​of various geometric parameters in this area are close to the alarm limits, they are not completely exceeded. This has made it difficult to accurately locate the root cause of the swaying for a long time, classifying it as a typical difficult-to-diagnose track defect section.

[0082] In the diagnostic process of this embodiment, the system first accesses the comprehensive detection data of the line. By performing a preliminary screening step based on an adaptive sliding window, the system automatically identifies the section from kilometer marker K12+300 to K12+600. Analysis shows that although this section does not exhibit obvious single over-limit characteristics on the traditional geometric waveform diagram, the peak density of its vehicle body lateral acceleration is significantly higher than the average level of the entire line, reaching the alarm threshold of 1.8 peaks per 10 meters, thus marking it as a key analysis target.

[0083] For this locked section, the system immediately initiated a composite irregularity feature extraction module for in-depth analysis. Using wavelet packet decomposition and Hilbert transform techniques, the system extracted the directional and horizontal irregularities within the sensitive wavelength range of 25 to 35 meters for this section, and calculated their inverse phase difference in the spatial domain to be 168 degrees. This value falls within... The determination interval indicates that there is a typical "reverse phase lock" phenomenon in this section, that is, the direction of the track deviation and the roll effect of the horizontal deviation form an anti-phase superposition in dynamics, which leads to a significant amplification of the lateral disturbance force of the vehicle.

[0084] Simultaneously, the system invokes the structural state inversion module to evaluate the fastener support status of this section. Combining maintenance record data from the past five years for this line, the system uses a particle swarm optimization algorithm to optimize the parameters of the time-varying fastener stiffness model. The inversion results show the stiffness attenuation coefficient of the fastener system in this section. The value is as high as 0.08, far exceeding the average level of 0.03 for normal aging tracks. This calculation result is highly consistent with the record that no comprehensive fastener re-tightening operation has been carried out in this section in the past three years, revealing a deep-seated hidden danger of insufficient support stiffness leading to a decline in the track's dynamic geometric holding capacity.

[0085] Based on the above multidimensional analysis results, the cause diagnosis engine ultimately diagnosed the swaying section as "coupled deterioration type," meaning it was jointly induced by the anti-phase superposition effect of complex irregularities and excessive attenuation of fastener stiffness. According to the maintenance recommendations output by the system, the track maintenance department prioritized targeted fastener re-tightening and refined track realignment work for this section. Retest results showed that after the targeted maintenance, the root mean square value (RMS) of the train's lateral acceleration during operation in this section decreased by 52%, returning to an excellent state. Furthermore, no further complaints about swaying were received regarding this section in the following three months, thus strongly demonstrating the effectiveness and accuracy of this invention in resolving complex and hidden track defects.

[0086] It should be noted that the specific embodiments described above are exemplary. Those skilled in the art can devise various solutions inspired by the disclosure of this invention, and these solutions all fall within the scope of this invention and its protection. Those skilled in the art should understand that this specification and its accompanying drawings are illustrative and not intended to limit the scope of the claims. The scope of protection of this invention is defined by the claims and their equivalents. This specification contains multiple inventive concepts; terms such as "preferredly," "according to a preferred embodiment," or "optionally" indicate that the corresponding paragraph discloses an independent concept. The applicant reserves the right to file divisional applications based on each inventive concept.

Claims

1. A method for diagnosing the causes of track swaying sections based on multi-source data fusion, characterized in that, Includes the following steps: Acquire vehicle vibration response data and track static geometry data of the target line, and perform spatiotemporal alignment processing; Statistical analysis was performed on the vehicle vibration response data to identify abnormal vibration sections; Extract the track-direction irregularity component and the horizontal irregularity component from the track static geometry data corresponding to the vibration anomaly section, and calculate the phase correlation characteristic value between the two. A dynamic model including track stiffness parameters is constructed. With the goal of minimizing the residual between the simulated vibration response output by the dynamic model and the on-board vibration response data, the parameters of the dynamic model are inverted to obtain the equivalent stiffness degradation parameters of the vibration anomaly section. Based on the phase correlation eigenvalues ​​and the equivalent stiffness degradation parameters, the type of swaying caused by the abnormal vibration section is determined.

2. The method according to claim 1, characterized in that, When acquiring the vehicle vibration response data and track static geometry data of the target line, the method also includes acquiring the historical maintenance record data of the target line. The dynamic model is a time-varying model of fastener stiffness, which includes maintenance intervention compensation items determined based on the historical maintenance record data.

3. The method according to claim 1 or 2, characterized in that, The phase correlation feature value includes the inverse phase difference; the calculation of the phase correlation feature value between the two includes: The track irregularity component and the horizontal irregularity component are respectively subjected to time-frequency signal decomposition to extract feature signals within a preset wavelength range; The instantaneous phase of the characteristic signal is calculated using Hilbert transform, and the absolute value of the difference between the two instantaneous phases is calculated to obtain the inverse phase difference.

4. The method according to any one of claims 1 to 3, characterized in that, The preset wavelength range is dynamically determined based on the train design operating speed and the lateral resonance frequency of the car body on the target line. When the reverse phase difference falls into the preset phase-locking interval When it is determined that there is a compound non-cooperative collaborative excitation, in which This is a preset angle threshold.

5. The method according to any one of claims 1 to 4, characterized in that, The statistical analysis of the vehicle vibration response data to identify abnormal vibration zones specifically includes: An adaptive sliding window algorithm is used to calculate the peak acceleration density per unit distance. The continuous mileage in which the peak acceleration density exceeds the preset alarm threshold is determined as the abnormal vibration section; The length of the adaptive sliding window changes positively with the train speed.

6. The method according to any one of claims 1 to 5, characterized in that, The parameter inversion of the dynamic model to obtain the equivalent stiffness degradation parameters of the abnormal vibration section includes: A time-varying model of fastener stiffness is constructed with initial stiffness, stiffness attenuation coefficient and nonlinear exponent as undetermined parameters; The parameters to be determined are iteratively adjusted using an optimization algorithm to minimize the residual between the simulated vibration response output by the dynamic model and the vehicle vibration response data. The stiffness attenuation coefficient corresponding to minimizing the residual is used as the equivalent stiffness degradation parameter.

7. The method according to any one of claims 1 to 6, characterized in that, Determining the type of swaying cause in the abnormal vibration section includes calculating the comprehensive swaying index. : A first normalized component determined based on the vibration amplitude of the vibration anomaly section, a second normalized component determined based on the phase correlation characteristic value, and a third normalized component determined based on the equivalent stiffness degradation parameter are obtained respectively. The first normalized component, the second normalized component, and the third normalized component are evaluated using a preset feature fusion algorithm in a dimensionless manner to obtain the comprehensive sway index. .

8. The method according to any one of claims 1 to 7, characterized in that, The types of causes of vehicle swaying include complex irregularity-dominated type, support stiffness deterioration type, coupling deterioration type, and single vibration exceeding limit type. Determining the cause type of vehicle swaying in the abnormal vibration section specifically includes: The comprehensive sway index If the threshold is exceeded, compare and The percentage of contribution; like Greater than 0 and The largest proportion and If the first preset condition is met, it is determined to be a compound irregularity-dominant type; like The largest proportion and If the second preset condition is met, it is determined to be a support stiffness degradation type; like If the values ​​are greater than 0 and the difference between the two percentages is less than the preset value, and both satisfy their respective preset conditions, it is determined to be a deteriorating coupling type. like =0 and If the second preset condition is not met, it is determined to be a single vibration exceeding the limit type.

9. The method according to any one of claims 1 to 8, characterized in that, Also includes: Calculate the vibration response gain factor based on the equivalent stiffness degradation parameter. , used to characterize the amplification factor of vibration amplitude due to stiffness degradation; Establish a geometric degradation cumulative function It is used to characterize the cumulative destructive effect of vibration excitation on track geometry; By combining the vibration response gain factor and the geometric degradation accumulation function, a diagnostic report containing maintenance level recommendations is generated.

10. A diagnostic system for the causes of track swaying sections based on multi-source data fusion, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the steps of the method as claimed in any one of claims 1 to 9.

Images