Laser reflection vibration and visual fusion perception track health monitoring method and system

Abstract

The application discloses a laser reflection vibration and visual fusion perception track health monitoring method and system, and the monitoring method comprises the following steps: when a train passes through a monitoring section, a laser vibration acquisition module starts to work to acquire a slight vibration response signal of the track; a visual event acquisition module performs video monitoring on the monitoring section to acquire visual information when the train passes through the track; data processing is respectively performed on the vibration response signal and the visual information, and the two are synchronously aligned; key features capable of representing a track state are extracted from the above multi-source data; intelligent discrimination is performed on the extracted key features based on a fusion model of a multi-path deep neural network; whether the track state is normal is judged, and damage evaluation is output; and the system comprises the following modules: a laser vibration acquisition module, a visual event acquisition module, a data fusion processing module, a cloud processing platform and an alarm reporting module. The application significantly improves the real-time performance and reliability of railway track health monitoring.

Description

Technical Field

[0001] This invention belongs to the field of railway track structure safety monitoring technology, specifically relating to a method and system for monitoring track health by combining laser reflection vibration and visual perception. Background Technology

[0002] Under the long-term influence of the enormous loads of trains and complex and ever-changing environmental factors, railway tracks gradually deteriorate. Wear, deformation, cracks, and other defects are like hidden "time bombs." If not detected and addressed in time, they may cause serious safety accidents, threatening the lives and property of passengers. Therefore, efficient and accurate health monitoring of tracks is crucial to ensuring the safe operation of rail transit.

[0003] Currently, railway departments typically conduct track condition checks through regular manual inspections and track recording vehicles (TRVs). While TRVs can accurately measure track geometry, their inspection frequency is low, usually only 1-2 times per line per month, resulting in a monthly coverage rate of approximately 6%. Such a long inspection cycle makes it difficult to promptly detect sudden track degradation issues, and potential safety hazards are easily overlooked. Manual inspections, on the other hand, are even less efficient and cannot meet the needs of real-time online monitoring for high-speed railways.

[0004] Some track monitoring solutions attempt to install sensors such as accelerometers and strain gauges beside the track or on the rail to collect vibration and strain data in real time. However, such contact measurement methods have many drawbacks: limited measurement accuracy, large data errors, frequent sensor maintenance, and short sensor lifespan. In addition, the direct attachment of sensors to the rail introduces additional mass effects, interfering with the original track response, which is particularly disadvantageous on high-speed lines, limiting the application of contact sensing solutions.

[0005] In recent years, research has also explored non-contact methods such as fiber optic sensing and laser vibration measurement to monitor the condition of track structures. Among these, laser detection technology, with its advantages of high precision, non-contact operation, and rapid measurement, has been applied in the field of track safety inspection and has become an indispensable key technology in modern track health monitoring. Non-contact laser Doppler vibration measurement does not add any mass burden to the structure being measured, possessing the unique advantage of being completely non-invasive, and can measure extremely minute structural vibrations with high sensitivity. However, existing solutions often fail to integrate train operating parameters with track vibration response and lack mechanisms for automatically triggering monitoring.

[0006] Under the conditions of dense and high-speed train operation on high-speed railways, there is an urgent need for a monitoring method that can automatically collect track vibration responses as trains pass and perform comprehensive analysis in conjunction with train load information. Currently, there is a lack of such an integrated solution that combines visual detection triggering, train parameter acquisition, and laser non-contact vibration measurement. Summary of the Invention

[0007] This invention is proposed to address the problems existing in the prior art, and its purpose is to provide a method and system for monitoring track health by fusing laser reflection vibration with visual perception.

[0008] The technical solution of this invention is: a method for monitoring orbital health by combining laser reflection vibration and visual perception, comprising the following steps:

[0009] A. When the train passes through the monitoring section, the laser vibration acquisition module starts working to acquire the minute vibration response signals of the track;

[0010] B. The visual event acquisition module performs video surveillance on the monitored section to obtain visual information when the train passes over the track;

[0011] C. Perform data processing on the vibration response signal and visual information respectively, and synchronize and align the two;

[0012] D. Extract key features that can characterize the orbital state from the above multi-source data;

[0013] E. A fusion model based on multi-path deep neural networks intelligently identifies the extracted key features;

[0014] F. Determine if the track condition is normal and output a damage assessment.

[0015] Furthermore, the laser vibration acquisition module includes a concave reflector mounted on the track, a laser emitter disposed on the incident side of the concave reflector, and a high-speed camera disposed on the reflecting side of the concave reflector.

[0016] Furthermore, in step A, when the train passes through the monitoring section, the laser vibration acquisition module starts working to acquire the minute vibration response signals of the track. The specific process is as follows:

[0017] First, when the train passes through the monitoring section, the track deflects, and the deflection of the concave reflector causes the light spot on the high-speed camera's imaging surface to shift.

[0018] Then, the high-speed camera acquires the light spot trajectory at a high frame rate and obtains the position sequence of the light spot changing over time, thereby obtaining the track vibration displacement time sequence signal, i.e., the vibration response signal.

[0019] Furthermore, the principle of optical displacement amplification can be expressed as:

[0020] ;

[0021] in, δ represents the displacement of the light spot on the camera's imaging plane, δ represents the actual deflection of the track, and M represents the magnification of the optical system.

[0022] Furthermore, in step B, the visual event acquisition module performs video surveillance on the monitored section to obtain visual information about the train passing over the track. The specific process is as follows:

[0023] First, when the train enters the monitoring section, the camera begins recording video or continuously capturing images until the train has completely passed;

[0024] Then, the visual information from the video or continuous snapshots includes the time the train passed, its speed, and the appearance of the track surface;

[0025] Next, visual information is used to determine when the train wheels reach directly above the track measuring point, which is used for precise timing alignment with the vibration response signal;

[0026] Finally, the camera captures visible defects on the track surface.

[0027] Furthermore, step C processes the vibration response signal and visual information separately, and synchronizes and aligns them. The specific process is as follows:

[0028] First, the data fusion processing module performs digital filtering and baseline correction on the vibration response signal to improve the signal-to-noise ratio;

[0029] Then, the timestamps of the camera images are read, and the effective segments traversed by the train are extracted from the vibration displacement timing signal.

[0030] Next, image enhancement and distortion correction are performed on the visual image sequence in the visual information, and key frames are selected or optical flow is calculated to extract valid events;

[0031] Finally, synchronized vibration signal segments and corresponding visual event sequences are obtained, preparing for feature extraction.

[0032] Furthermore, step D extracts key features that characterize the orbital state from the aforementioned multi-source data. The specific process is as follows:

[0033] First, time-frequency analysis was used to obtain the amplitude and frequency characteristics of the measured vibration signal segment;

[0034] Then, time-domain features, namely maximum deflection, root mean square value, and vibration decay time, are extracted to describe the overall shape of the track deflection response.

[0035] Next, a vehicle-track finite element coupled simulation model is established to generate a track response benchmark under healthy conditions;

[0036] Then, the train operation process was simulated by the equivalent moving load method. The load was gradually applied in the simulation to obtain the curve of track deflection as a function of time under healthy conditions.

[0037] Next, the measured response signal and the simulated health baseline response signal are compared point by point to obtain the residual signal;

[0038] Finally, the residual signal is processed to obtain the residual feature set.

[0039] Furthermore, step E uses a multi-path deep neural network fusion model to intelligently identify the extracted key features. The specific process is as follows:

[0040] First, a deep learning network structure containing vibration signal branches and visual signal branches was constructed within the data fusion processing module;

[0041] Then, the vibration signal branch processes the key features of the vibration signal segment in step D;

[0042] Next, the visual signal branch processes the image depth features in the visual information;

[0043] Next, the processed features are input into the fusion layer of the fusion model for fusion, and the classification result of the orbital health status is obtained in the output layer.

[0044] Furthermore, step F determines whether the orbital condition is normal and outputs a damage assessment. The specific process is as follows:

[0045] First, if the fusion model determines that the orbital status is normal, the result is recorded and monitoring continues without the need for an alarm.

[0046] Then, if the fusion model determines that it is abnormal, the alarm reporting module's early warning mechanism is triggered. The alarm reporting module sends signals through audible and visual warnings and sends warning messages to the maintenance center, and generates a maintenance report containing the current diagnostic results.

[0047] A system for monitoring orbital health using a laser-reflected vibration and visual fusion sensing method includes the following modules:

[0048] The laser vibration acquisition module is used to acquire track vibration signals.

[0049] The visual event acquisition module acquires track image information;

[0050] The data fusion and processing module constructs a deep learning network structure to identify the health status of the track;

[0051] A cloud processing platform for data storage and processing in the cloud;

[0052] The alarm reporting module provides alarm indications based on health status.

[0053] The beneficial effects of this invention are as follows:

[0054] This invention automatically triggers track vibration monitoring when a train passes by and performs a precise health status assessment based on train operating parameters. The system uses a non-contact measurement method, eliminating the need to deploy numerous sensors on the track, thus reducing installation difficulty and maintenance costs.

[0055] This invention utilizes the principle of optical displacement amplification to significantly magnify minute track displacements, achieving high monitoring accuracy, up to sub-millimeter deflection resolution; it integrates train operating parameters and vibration signal analysis to comprehensively assess the load response characteristics of the track structure; and it leverages cloud-based algorithms to achieve automatic anomaly identification and early warning, greatly improving the intelligence and real-time performance of track health monitoring.

[0056] The deep learning diagnostic model in this invention is trained from massive amounts of simulation data and measured data, and has a strong ability to express complex features. It also has good generalization performance for identifying track damage under different train types, operating speeds and environmental conditions.

[0057] The monitoring method and system provided by this invention have advantages such as easy installation, zero interference with train operation, high monitoring accuracy, and strong algorithm robustness, which can significantly improve the efficiency and reliability of railway track health monitoring and ensure train operation safety. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the overall architecture of the track health monitoring system of the present invention;

[0059] Figure 2 This is a schematic diagram illustrating the working principle of the optical displacement amplification vibration measurement device in this invention;

[0060] Figure 3 This is a schematic flowchart of the track health monitoring method of the present invention;

[0061] Figure 4 This is a schematic diagram of the hierarchical deep learning analysis model structure used in this invention;

[0062] Figure 5 This is a schematic diagram of the track damage alarm threshold determination based on residual signals in this invention;

[0063] The components include: 1. Track; 11. Laser vibration acquisition module; 12. Visual event acquisition module; 13. Data fusion processing module; 14. Cloud processing platform; 15. Alarm reporting module; 111. Laser emitter; 112. Concave reflector; 113. High-speed camera. Detailed Implementation

[0064] The present invention will now be described in detail with reference to the accompanying drawings and embodiments:

[0065] Figure 1This is a schematic diagram of the overall architecture of the track health monitoring system of the present invention, showing the relationship and data flow between components such as the monitoring cameras along the line, the non-contact optical displacement amplification monitoring device, and the cloud analysis platform.

[0066] A system for monitoring orbital health using a laser-reflected vibration and visual fusion sensing method includes the following modules:

[0067] The laser vibration acquisition module 11 is used to acquire vibration signals from track 1.

[0068] The visual event acquisition module 12 acquires image information of track 1;

[0069] The data fusion processing module 13 constructs a deep learning network structure to identify the health status of track 1;

[0070] The cloud processing platform 14 performs cloud-based data storage and processing;

[0071] Alarm reporting module 15 provides alarm indications based on health status.

[0072] Specifically, in conjunction with the laser vibration acquisition module 11 is a monitoring trigger module. The monitoring trigger module uses existing surveillance cameras or other sensing devices along the railway line to monitor the track area in real time. When a train is detected entering the monitoring area, a trigger signal is generated to initiate the subsequent monitoring process.

[0073] Specifically, the track health monitoring system also includes a train information retrieval module. After receiving a trigger signal, it connects with the database interface of the railway control center to retrieve the operating parameters of the currently passing trains, including train number, train type, operating speed, number of passengers, load, and other information.

[0074] Figure 2 This is a schematic diagram illustrating the working principle of the optical displacement amplification vibration measurement device in this invention. The diagram shows the process by which a laser beam forms a light spot on an industrial camera sensor after being reflected by a concave mirror: when the track experiences a slight deflection, the displacement of the light spot within the camera's imaging plane is amplified and manifested as a pixel offset.

[0075] Specifically, the laser vibration acquisition module 11 is a non-contact measurement device, consisting of a laser emitter 111, a concave reflector 112, and a high-speed camera 113. When a train passes, the laser vibration acquisition module 11 projects a laser beam onto the concave reflector 112 mounted on the surface of track 1. The reflected beam is converged by the concave reflector 112 and then slightly diverges before being projected onto the high-speed camera 113. The minute vibrations of track 1 caused by the train load cause slight changes in the position of the reflected laser spot. The high-speed camera 113 captures and records these changes in spot displacement, thereby acquiring the track's vibration response signal.

[0076] Specifically, the data fusion processing module 13 can be integrated into the cloud processing platform 14. The aforementioned triggering events, train parameters, and vibration response signals are uploaded to the cloud processing platform 14 in real time. The fusion model of the track health assessment algorithm is run to perform fusion analysis on the collected data, and to perform spectral analysis and modal feature extraction on the track vibration signals. Combined with train speed and load parameters, the health status of the track structure is assessed. The cloud processing platform 14 includes a finite element simulation model and a deep learning diagnostic algorithm to predict the baseline response under track health conditions and identify track anomaly characteristics.

[0077] Specifically, when the analysis results of the cloud processing platform 14 indicate potential structural problems or anomalies in the track, a track health status report and alarm information are generated and sent to maintenance management personnel or the control center via the network. The alarm information generated by the alarm report module 15 includes the location, nature, and severity of the detected anomaly; the system can also periodically generate track health monitoring reports, record historical data and assessment results, and provide a basis for maintenance decisions.

[0078] Specifically, the laser vibration acquisition module 11 is housed in a protective housing. The laser emitter 111, high-speed camera 113, and local data transmission unit are mounted inside the housing via a mounting bracket. The housing has a window on the side facing the concave reflector 112 for laser emission and reflected light to enter.

[0079] Specifically, the track health monitoring system uses solar photovoltaic power, with photovoltaic panels installed on the top of the mounting housing to achieve independent power supply.

[0080] Specifically, the high-speed camera 113 is positioned directly above the laser emitter 111 and facing the direction of the reflected light spot to ensure that the imaging field of view covers the range of light spot movement.

[0081] Specifically, the mounting housing is fixedly installed in a safe position next to the railway sleeper using a bracket, which ensures the stability of the optical path between the laser and the reflector without affecting the train's operating clearance.

[0082] Specifically, the distance between the mounting housing and the concave reflector 112 is determined according to the on-site working conditions, and is preferably 5 to 30m.

[0083] Specifically, the track status monitored by the cloud processing platform 14 includes the track's vibration amplitude, vibration frequency, and potential structural anomaly indicators.

[0084] Figure 3 This is a flowchart illustrating the track health monitoring method of the present invention, showing the complete process from the moment a train passes by, through optical vibration signal acquisition, data processing, simulation comparison, deep learning analysis, to the output of an alarm.

[0085] A laser-reflected vibration and visual fusion sensing method for track health monitoring includes the following steps:

[0086] A. When the train passes through the monitoring section, the laser vibration acquisition module 11 starts to work and acquires the small vibration response signal of the track;

[0087] B. The visual event acquisition module 12 performs video monitoring on the monitored section to acquire visual information when the train passes over the track;

[0088] C. Perform data processing on the vibration response signal and visual information respectively, and synchronize and align the two;

[0089] D. Extract key features that can characterize the orbital state from the above multi-source data;

[0090] E. A fusion model based on multi-path deep neural networks intelligently identifies the extracted key features;

[0091] F. Determine if the track condition is normal and output a damage assessment.

[0092] The laser vibration acquisition module 11 includes a concave reflector 112 installed on the track 1, a laser emitter 111 disposed on the incident side of the concave reflector 112, and a high-speed camera 113 disposed on the reflecting side of the concave reflector 112.

[0093] Step A: When the train passes through the monitoring section, the laser vibration acquisition module 11 starts working to acquire the minute vibration response signals of the track. The specific process is as follows:

[0094] First, when the train passes through the monitoring section, track 1 deflects, and the deflection of the concave reflector 112 causes the light spot on the imaging surface of the high-speed camera 113 to shift.

[0095] Then, the high-speed camera 113 acquires the light spot trajectory at a high frame rate and obtains the position sequence of the light spot changing over time, thereby obtaining the track vibration displacement time sequence signal, i.e., the vibration response signal.

[0096] The principle of optical displacement amplification can be expressed as:

[0097] in, δ represents the displacement of the light spot on the camera's imaging plane, δ represents the actual deflection of the track, and M represents the magnification of the optical system.

[0098] Specifically, step A captures sub-millimeter-level micro-vibrations of the track under train load in a non-contact manner, providing basic data for subsequent analysis.

[0099] Step B: The visual event acquisition module 12 performs video monitoring on the monitored section to acquire visual information when the train passes over the track. The specific process is as follows:

[0100] First, when the train enters the monitoring section, the camera begins recording video or continuously capturing images until the train has completely passed;

[0101] Then, the visual information from the video or continuous snapshots includes the time the train passed, its speed, and the appearance of the track surface;

[0102] Next, visual information is used to determine when the train wheels arrive directly above track 1 measuring point, which is used for precise timing alignment with the vibration response signal;

[0103] Finally, the camera captures visible defects on the track surface.

[0104] Specifically, visible defects in step B include loose fasteners and foreign object intrusion.

[0105] Specifically, in step B, the camera can also obtain information about the train's operating status, such as whether the wheels are abnormally bumpy.

[0106] Specifically, the aforementioned visible defects and train operating status, as auxiliary features, help explain the causes of vibration signal changes and improve the reliability of track health status assessment.

[0107] Step C involves data processing of the vibration response signal and visual information, and synchronizing and aligning them. The specific process is as follows:

[0108] First, the data fusion processing module 13 performs digital filtering and baseline correction on the vibration response signal to improve the signal-to-noise ratio;

[0109] Then, the timestamps of the camera images are read, and the effective segments traversed by the train are extracted from the vibration displacement timing signal.

[0110] Next, image enhancement and distortion correction are performed on the visual image sequence in the visual information, and key frames are selected or optical flow is calculated to extract valid events;

[0111] Finally, synchronized vibration signal segments and corresponding visual event sequences are obtained, preparing for feature extraction.

[0112] Specifically, when necessary, the vibration response signal needs to be interpolated or resampled to match its frame rate with the camera frame rate, thereby ensuring that the two types of data are aligned on the time axis.

[0113] Step D extracts key features that characterize the orbital state from the above multi-source data. The specific process is as follows:

[0114] First, time-frequency analysis was used to obtain the amplitude and frequency characteristics of the measured vibration signal segment;

[0115] Then, time-domain features, namely maximum deflection, root mean square value, and vibration decay time, are extracted to describe the overall shape of the track deflection response.

[0116] Next, a vehicle-track finite element coupled simulation model is established to generate a track response benchmark under healthy conditions;

[0117] Then, the train operation process was simulated by the equivalent moving load method. The load was gradually applied in the simulation to obtain the curve of track deflection as a function of time under healthy conditions.

[0118] Next, the measured response signal and the simulated health baseline response signal are compared point by point to obtain the residual signal;

[0119] Finally, the residual signal is processed to obtain the residual feature set.

[0120] Specifically, in step D, time-frequency analysis is used to obtain the amplitude and frequency characteristics of the measured vibration signal segment. The specific process is as follows:

[0121] First, a Fast Fourier Transform (FFT) is performed on the vibration time series to obtain the main frequency components and energy distribution in the spectrum;

[0122] Then, the spectrum of a healthy track is usually dominated by low-frequency main modes, with concentrated energy distribution and clear attenuation patterns; when the track bed is loose or cracks appear in the track, additional vibration modes may be introduced, causing secondary peaks or high-frequency components to appear in the spectrum.

[0123] Specifically, in step D, the vehicle-track finite element coupled simulation model consists of a vehicle dynamics model and a track structure model: the vehicle part includes the car body, wheelsets and suspension system, and uses rigid body elements, mass elements and spring-damping elements to characterize the dynamic characteristics; the track part is modeled using beam elements and elastic support elements to simulate the supporting role of sleepers and track bed.

[0124] Specifically, in step D, the measured response signal and the simulated health baseline response signal are compared point by point to obtain the residual signal. The specific process is as follows:

[0125] The train operation process was simulated using the equivalent moving load method. The load was gradually applied in the simulation to obtain the track deflection versus time curve under healthy conditions. The measured signal Health baseline signal generated by simulation By comparing point by point, the residual signal is obtained:

[0126] .

[0127] Specifically, in step D, the residual signal is processed to obtain the residual feature set. The specific process is as follows:

[0128] After the obtained residuals are processed by denoising and normalization, the corresponding time-domain and frequency-domain features are extracted to form a residual feature set.

[0129] The large-scale health response database generated by simulation provides reliable support for subsequent baseline-residual diagnostic analysis.

[0130] Figure 4 This is a schematic diagram of the hierarchical deep learning analysis model structure used in this invention. It includes a first-layer deep neural network (DNN) model for health benchmark prediction and a second-layer convolutional neural network model (using ResNet and FPN structures) for residual feature damage identification, as well as the data transfer relationship between the two. The diagram also shows a simulation database as the data source for model input, connected to both the DNN and CNN models to support the model training and analysis process.

[0131] Step E uses a multi-path deep neural network fusion model to intelligently identify the extracted key features. The specific process is as follows:

[0132] First, a deep learning network structure containing vibration signal branches and visual signal branches was constructed inside the data fusion processing module 13;

[0133] Then, the vibration signal branch processes the key features of the vibration signal segment in step D;

[0134] Next, the visual signal branch processes the image depth features in the visual information;

[0135] Next, the processed features are input into the fusion layer of the fusion model for fusion, and the classification result of the orbital health status is obtained in the output layer.

[0136] Specifically, the data fusion processing module 13 internally constructs a deep learning network structure that includes a vibration signal branch and a visual signal branch. The vibration branch receives vibration feature parameters obtained in step D, such as the main vibration frequency and amplitude, and further processes them through several fully connected layers; the visual branch receives image depth features and extracts multi-scale information through convolutional layers of a Residual Network (ResNet) and a Feature Pyramid Network (FPN). Subsequently, the two feature paths are merged in the fusion layer, connected to multiple layers of fully connected neurons for comprehensive analysis, and output the classification result of the track health status.

[0137] Specifically, in this invention, the fusion model is trained as a binary classifier to determine whether the track condition is "normal" or "abnormal". The classification criteria include both the physical characteristics of the vibration signal, such as whether the main vibration frequency has shifted or whether the vibration energy has increased abnormally, and visual inspection results, such as whether track components are loose or whether train operation is abnormal, thereby improving diagnostic accuracy.

[0138] Specifically, when the track bed support stiffness decreases, the vibration signal attenuation slows down and the high-frequency components increase. Simultaneously, the visual system may detect subtle track subsidence. The model interprets this combined information as an anomaly. By using deep learning to identify complex patterns, this system can detect minute track anomalies, significantly reducing false alarms and missed alarms.

[0139] Figure 5 This diagram illustrates the determination of track damage alarm thresholds based on residual signals. It shows the changes in residual signal characteristics and deep learning model outputs under different levels of track damage, as well as the process by which the system triggers an alarm when the residual or damage index exceeds a preset threshold.

[0140] Step F determines whether the orbital status is normal and outputs a damage assessment. The specific process is as follows:

[0141] First, if the fusion model determines that the orbital status is normal, the result is recorded and monitoring continues without the need for an alarm.

[0142] Then, if the fusion model determines that it is abnormal, the alarm reporting module 15 is triggered to issue an early warning mechanism. The alarm reporting module 15 issues signals by means of audible and visual warnings and sending warning information to the maintenance center, and generates a maintenance report containing the current diagnostic results.

[0143] Meanwhile, to further refine damage localization and assessment, a vehicle-track finite element coupled simulation model was established in the cloud. Vehicle-track joint dynamics simulation was performed using ANSYS software to obtain the track response benchmark under healthy conditions.

[0144] Specifically, the vehicle model consists of rigid body elements for the vehicle body, mass elements for the wheels, and spring-damping elements for the suspension system. The vehicle body is modeled using rigid body elements (MPC184) and assigned corresponding mass and moment of inertia, while the front and rear wheels are modeled using MASS21 mass elements. The vertical and lateral stiffness and damping of the suspension system are defined using COMBIN14 spring-damping elements. The vehicle body and wheels are rigidly coupled through rigid body elements and connecting elements to form a complete vehicle-rail dynamics model.

[0145] Specifically, the track model is discretized using finite element methods based on the actual structure, and the supports are simulated using fixed-support constraints to simulate the actual load boundary conditions.

[0146] Based on the above, the system obtains the track response under healthy conditions by inputting the train operation parameters into the simulation model, and then compares the measured response with the simulation benchmark to calculate the residual sequence for damage identification.

[0147] Furthermore, the threshold can be determined by statistical analysis of a large amount of measured and simulated residual data under the orbital health state, and the upper limit of the distribution of residual peak value, residual energy or frequency offset can be selected as the threshold.

[0148] Specifically, when the monitored residual characteristics exceed the threshold, the system determines that there may be an anomaly in the track. The anomaly can be categorized into three levels—minor, moderate, and severe—based on the degree of exceedance, to trigger different levels of alarms or reporting measures. Minor anomalies prompt maintenance personnel to perform trend tracking, moderate anomalies trigger an early warning, and severe anomalies trigger an immediate alarm and generate a detailed diagnostic report.

[0149] Through the implementation of the above steps, this invention enables all-weather online monitoring and intelligent diagnosis of track health status. Each time a train passes, the system automatically collects vibration and visual data for comprehensive analysis: on the one hand, it uses laser reflection vibration sensors to capture the subtle structural responses of the track; on the other hand, it utilizes visual events to provide information on train load conditions and track appearance, combining this with a deep learning model to achieve a comprehensive perception of the track's condition.

[0150] Experiments show that the system of this invention has high sensitivity to early signs such as slight uneven settlement of track structures and loose fasteners, providing timely warnings before they are detected by traditional manual inspections. Especially in highly interference environments such as electrified railways, the system can still operate stably and is unaffected by electromagnetic noise. In summary, the track health monitoring method and system provided by this invention significantly improve the intelligence and proactive prevention capabilities of track maintenance, which is of great significance for ensuring the safe operation of rail transit.

[0151] It should be noted that the above embodiments are intended to better illustrate the technical solution of the present invention, but the present invention is not limited to the specific content described above. Without departing from the principles of the present invention, those skilled in the art can make various equivalent substitutions and modifications to the implementation methods, and these changes should all be considered within the scope of protection of the present invention. The scope of protection of the present invention is defined by the appended claims.

Claims

1. A method for monitoring track health by fusing laser reflection vibration with visual event perception, characterized in that: Includes the following steps: A. When the train passes through the monitoring section, the laser vibration acquisition module (11) starts to work and acquires the small vibration response signal of the track; B. The visual event acquisition module (12) performs video monitoring on the monitored section and acquires visual information when the train passes through the track. C. Perform data processing on the vibration response signal and visual information respectively, and synchronize and align the two; D. Extract key features that can characterize the orbital state from multi-source data; E. A fusion model based on multi-path deep neural networks intelligently identifies the extracted key features; F. Determine if the track condition is normal and output a damage assessment; Step D extracts key features that characterize the orbital state from the above multi-source data. The specific process is as follows: First, time-frequency analysis was used to obtain the amplitude and frequency characteristics of the measured vibration signal segment; Then, time-domain features, namely maximum deflection, root mean square value, and vibration decay time, are extracted to describe the overall shape of the track deflection response. Next, a vehicle-track finite element coupled simulation model is established to generate a track response benchmark under healthy conditions; Then, the train operation process was simulated by the equivalent moving load method. The load was gradually applied in the simulation to obtain the curve of track deflection as a function of time under healthy conditions. Next, the measured response signal and the simulated health baseline response signal are compared point by point to obtain the residual signal; Finally, the residual signal is processed to obtain the residual feature set.

2. The method for track health monitoring based on the fusion perception of laser reflection vibration and visual events according to claim 1, characterized in that: The laser vibration acquisition module (11) includes a concave reflector (112) installed on the track (1), a laser emitter (111) is provided on the incident side of the concave reflector (112), and a high-speed camera (113) is provided on the reflecting side of the concave reflector (112).

3. The method for track health monitoring based on the fusion perception of laser reflection vibration and visual events according to claim 2, characterized in that: When the train passes through the monitoring section in step A, the laser vibration acquisition module (11) starts working to acquire the small vibration response signal of the track. The specific process is as follows: First, when the train passes through the monitoring section, the track (1) deflects, and the deflection of the concave reflector (112) causes the light spot on the imaging surface of the high-speed camera (113) to shift. Then, the high-speed camera (113) acquires the light spot trajectory at a high frame rate and obtains the position sequence of the light spot changing over time, thereby obtaining the track vibration displacement time sequence signal, i.e. the vibration response signal.

4. The method for track health monitoring based on the fusion perception of laser reflection vibration and visual events according to claim 3, characterized in that: The principle of optical displacement amplification can be expressed as: ； in, δ represents the displacement of the light spot on the camera's imaging plane, δ represents the actual deflection of the track, and M represents the magnification of the optical system.

5. The method for track health monitoring based on the fusion perception of laser reflection vibration and visual events according to claim 1, characterized in that: Step B: The visual event acquisition module (12) performs video monitoring on the monitored section to obtain visual information when the train passes over the track. The specific process is as follows: First, when the train enters the monitoring section, the camera begins recording video or continuously capturing images until the train has completely passed; Then, the visual information from the video or continuous snapshots includes the time the train passed, its speed, and the appearance of the track surface; Then, visual information is used to determine when the train wheels arrive directly above the track (1) measuring point, which is used for precise timing alignment with the vibration response signal; Finally, the camera captures visible defects on the track surface.

6. The method for track health monitoring based on the fusion perception of laser reflection vibration and visual events according to claim 1, characterized in that: Step C involves data processing of the vibration response signal and visual information, and synchronizing and aligning them. The specific process is as follows: First, the data fusion processing module (13) performs digital filtering and baseline correction on the vibration response signal to improve the signal-to-noise ratio; Then, the timestamps of the camera images are read, and the effective segments traversed by the train are extracted from the vibration displacement timing signal. Next, image enhancement and distortion correction are performed on the visual image sequence in the visual information, and key frames are selected or optical flow is calculated to extract valid events; Finally, synchronized vibration signal segments and corresponding visual event sequences are obtained, preparing for feature extraction.

7. The method for track health monitoring based on the fusion perception of laser reflection vibration and visual events according to claim 1, characterized in that: Step E uses a multi-path deep neural network fusion model to intelligently identify the extracted key features. The specific process is as follows: First, a deep learning network structure containing vibration signal branches and visual signal branches was constructed inside the data fusion processing module (13); Then, the vibration signal branch processes the key features of the vibration signal segment in step D; Next, the visual signal branch processes the image depth features in the visual information; Next, the processed features are input into the fusion layer of the fusion model for fusion, and the classification result of the orbital health status is obtained in the output layer.

8. The method for track health monitoring based on the fusion perception of laser reflection vibration and visual events according to claim 1, characterized in that: Step F determines whether the orbital status is normal and outputs a damage assessment. The specific process is as follows: First, if the fusion model determines that the orbital status is normal, the result is recorded and monitoring continues without the need for an alarm. Then, if the fusion model determines that it is abnormal, the alarm reporting module (15) will trigger the early warning mechanism. The alarm reporting module (15) will issue signals by means of sound and light warnings and sending warning information to the maintenance center, and generate a maintenance report containing the current diagnostic results.

9. The system for track health monitoring based on the fusion perception of laser reflection vibration and visual events according to claim 1, characterized in that: Includes the following modules The laser vibration acquisition module (11) is used to acquire the vibration signal of the track (1); The visual event acquisition module (12) acquires image information of the track (1); The data fusion processing module (13) constructs a deep learning network structure to identify the health status of track (1); The cloud processing platform (14) performs cloud-based data storage and processing; The alarm reporting module (15) provides alarm indications based on health status.

Images