A method and device for detecting the collaborative state of a track multi-index self-organizing network

Abstract

The present invention discloses a method and device for detecting the collaborative state of a track in a multi-index self-organizing network. The method comprises establishing a multi-source point cloud model after preprocessing the collected data, establishing a three-dimensional point cloud model by using an iterative algorithm, establishing a predefined model after fusion by using a data fusion algorithm, integrating the three-dimensional point cloud model and the predefined model for consistency, extracting key characteristic parameters of the track, and identifying, evaluating and detecting the track in combination with optical image data; generating a track state detection report in combination with historical data and detection and analysis results; the device comprises a rail structure profile detection module, a fastener installation state detection module, a track plate crack separation detection module, a tunnel disease detection module, a posture position measurement module and a host computer, which are arranged on a mobile platform trolley. The present invention realizes simultaneous detection of multiple indicators of the track state, accurately reflects the actual state of the track, makes up for the shortcomings of a single data source in terms of coverage and precision, and improves detection efficiency and accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of track detection, and more specifically, relates to a method and device for detecting the collaborative state of a track multi-index self-organizing network. Background Art

[0002] With the rapid development of rail transit systems such as high-speed railways, subways, urban railways, and trams, the safety performance of tracks, as the infrastructure for train operation, directly affects the safety of train operation and the travel experience of passengers. Therefore, the detection of track status is particularly important. Traditional track status detection methods mainly rely on manual inspections and regular inspections, which have problems such as low efficiency, low accuracy, and susceptibility to human factors. At the same time, with the continuous expansion and complexity of track lines, traditional detection methods have been unable to meet the safety needs of modern rail transit.

[0003] In recent years, track status detection technology has continued to develop and innovate, and detection methods based on high-precision sensors such as laser scanners and line structured light sensors have gradually been applied. These sensors can obtain high-precision three-dimensional point cloud data of track structures, providing a data basis for accurate assessment of track status. However, the detection of a single sensor can often only reflect the status of a certain aspect of the track, while the safety performance of the track involves multiple indicators, such as geometric dimensions, deformation, cracks, wear, etc. Therefore, it is necessary to conduct collaborative detection based on the data of multiple sensors of a specific structure to comprehensively evaluate the safety performance of the track.

[0004] However, the current existing technology is unable to achieve synchronous data collection for multiple heterogeneous targets in orbital target data collection, resulting in low detection accuracy after analyzing the collected data, which cannot accurately reflect the actual status of the orbit and has low detection efficiency. Summary of the invention

[0005] In view of the problem that the existing track target data collection cannot realize the synchronous data collection of multiple heterogeneous targets, the present invention provides a track multi-indicator self-organizing network collaborative state detection method, which establishes a multi-source point cloud model after preprocessing the collected data, and uses an iterative algorithm to establish a three-dimensional point cloud model, and uses a data fusion algorithm to establish a predefined model after fusion, and the three-dimensional point cloud model and the predefined model are consistently integrated to extract the key characteristic parameters of the track, so as to realize all-round detection of the track and realize simultaneous detection of multiple indicators of the track state, accurately reflect the actual state of the track, make up for the shortcomings of a single data source in coverage and accuracy, and improve detection efficiency and accuracy.

[0006] In order to solve the above problems, according to the first aspect of the present invention, a method for detecting the collaborative state of a track multi-index self-organizing network is provided, which is implemented by a track multi-index self-organizing network system state detection device, wherein the detection device includes a rail structure profile detection module, a fastener installation state detection module, a track plate crack separation detection module, a tunnel disease detection module, a posture position measurement module and a host computer arranged on a mobile platform trolley, and includes the following steps:

[0007] S1, the mobile platform trolley drives along the track to be tested, and the host computer collects the track point cloud data, image data and posture data of at least one detection module combined with the posture position measurement module according to the detection requirements;

[0008] S2, the host computer pre-processes the point cloud data and image data;

[0009] S3, the pre-processed point cloud data is fused with the posture data to construct a multi-source point cloud model;

[0010] S4, rotating and translating the multi-source point cloud model through the transformation matrix, using the iterative registration algorithm to accurately align the overall point cloud data of the rail structure, realizing the overall registration of the track structure, and establishing a three-dimensional point cloud model of the track structure;

[0011] S5. Use data fusion algorithm to fuse and register point cloud data, obtain a full description of the actual state of the track, and establish a predefined model of the track structure state;

[0012] S6, integrating the three-dimensional point cloud model and the predefined model to extract key characteristic parameters of the track;

[0013] S7, based on the extracted key characteristic parameters of the track and combined with image data recognition, the track status assessment and detection analysis are performed;

[0014] S8. Generate a track status inspection report by combining historical data, standard parameters and inspection analysis results.

[0015] Preferably, in S2, the preprocessing includes denoising, filtering, data simplification and data interpolation of the point cloud data using pass-through filtering and statistical filtering methods to eliminate errors and redundant information in the point cloud data, and also includes correcting the image data, eliminating distortion and performing image enhancement.

[0016] Preferably, the straight-through filtering formula is:

[0017] Filtered Point Cloud={p∈P|xmin<x p <x max∧ymin<y p <y max∧z min<z p <z max}

[0018] Where: FilteredPointCloud means the filtered point cloud, P is the original point cloud dataset, p is any point in the point cloud, x p ,y p ,z p are the coordinates of point p on the x, y, and z axes respectively. xmin and xmax represent the minimum and maximum thresholds of point p on the x axis; ymin and ymax represent the minimum and maximum thresholds of point p on the y axis; zmin and zmax represent the minimum and maximum thresholds of point p on the z axis.

[0019] By setting the maximum and minimum thresholds on the coordinate axis to filter out points beyond these ranges, the initial cleaning of the point cloud data can be achieved;

[0020] The statistical filtering formula is:

[0021] Outliers={p∈P||d p -μ|>n×σ}

[0022] Where: Outliers means outliers, P is the original point cloud dataset, p is any point in the point cloud, d p is the distance from point p to its nearest neighbor, μ and σ are the mean and standard deviation of the distances from all points to their nearest neighbors, respectively, and n is a user-defined constant representing the “standard deviation multiple”, which is used to determine the threshold of outliers.

[0023] Preferably, in S3, the pre-processed point cloud data and the posture data are fused using the track mobile measurement system point cloud fusion processing software to construct a multi-source point cloud model.

[0024] Preferably, the step S5 specifically includes the following steps:

[0025] S5-1, using feature point matching or image overlay point cloud method to achieve registration between point cloud and image;

[0026] S5-2. Combine the attitude data and use Kalman filtering, Bayesian network, deep learning and other data fusion algorithms to fuse and align the data to obtain a complete description of the actual state of the track;

[0027] S5-3. Based on the actual structural state of the track, a predefined model of the track structural state is established through a machine learning model.

[0028] Preferably, in S5-1, the registration between the point cloud and the image is achieved by using a feature point matching method. Assume that P c is the set of feature points in the point cloud, P i is a set of feature points in the image. The goal of registration is to find a transformation T such that Pc and P i At the transformed position P c ′=T(P c ) and P i ′ is as close as possible, which is achieved by minimizing the following cost function, the formula is:

[0029]

[0030] Where: N is the number of matching points, ||.|| represents the Euclidean distance.

[0031] Preferably, in S5-2, the Kalman wave data fusion algorithm formula is:

[0032]

[0033] Where: represents the state estimation at the kth time step; A is the state transfer matrix, B is the input matrix, represents the state estimate of the previous time step, i.e., k-1; u k is the input vector, representing the control input or external influence at the kth time step; K k represents the Kalman gain at the kth time step, which determines the influence of the measurement value on the state estimation; yk represents the measurement value at the kth time step; H is the measurement matrix; represents the state predicted based on the state estimate and control input of the previous time step without considering the measurement value of the kth time step;

[0034] In S5-3, a random forest model is used to establish a prediction model for the track structure state. Assuming that there are m features X1, X2, ..., X m and a target variable Y, the training process formula is:

[0035]

[0036] Where: is the predicted orbital structure state, and θ is the model parameter, which is determined by the training process.

[0037] Preferably, the step S7 specifically includes the following steps:

[0038] S7-1. According to the extracted characteristic parameters such as track gauge, rail wear, fastener tightening force, ballast bed condition and sleeper damage, the normal range of each characteristic parameter is set according to the standard specification. If the characteristic parameter value exceeds the normal range, it is marked as abnormal and the next step is carried out. If it does not exceed the normal range, the monitoring is continued;

[0039] S7-2. Through data visualization, trend segmentation or correlation analysis, the root causes of data anomalies can be deeply analyzed.

[0040] According to a second aspect of the present invention, a long-distance continuous shape and position detection device for a track is provided, which is applied to the long-distance continuous shape and position detection method of a track, and further includes a power supply module, the mobile platform trolley includes a frame and a limiting wheel, the rail structure profile detection module, the fastener installation status detection module, the track plate crack separation detection module, the tunnel disease detection module, the posture position measurement module and the host computer are all arranged on the frame, the limiting wheels are at least two and symmetrically arranged at the bottom of the frame, the limiting wheels can move on the track to be tested, and the rail structure profile detection module, the fastener installation status detection module, the track plate crack separation detection module, the tunnel disease detection module, and the posture position measurement module are respectively connected to the host computer;

[0041] The rail structure profile detection module, the fastener installation status detection module and the track plate crack detection module are at least two and symmetrically arranged on the frame;

[0042] The posture position measurement module is used to collect posture data of the mobile platform car;

[0043] The host computer is used to control the operation of each module and the running speed of the mobile platform car;

[0044] The power module is used to supply power to the entire device.

[0045] Preferably, the rail structure profile detection module is a three-dimensional laser scanner;

[0046] The fastener installation status detection module is a line structured light sensor;

[0047] The track plate crack detection module is a laser displacement sensor;

[0048] The tunnel disease detection module is a high-resolution camera;

[0049] The attitude position measurement module includes an inertial navigation combination, a GNSS position receiver and a photoelectric encoder;

[0050] The host computer includes a sensor control unit, a data preprocessing unit and a data analysis and recording unit;

[0051] The power supply modules are two drive motors symmetrically arranged on the frame.

[0052] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0053] 1. A track multi-indicator self-organizing network collaborative state detection device of the present invention, by setting multiple detection modules and installing the multiple modules on a mobile platform trolley, the multiple detection modules work together to realize simultaneous detection of multiple indicators of the track state, comprehensively evaluate the actual working state of the track, avoid the limitations of single indicator detection, comprehensively reflect the safety performance of the track, and significantly improve the detection accuracy and efficiency.

[0054] 2. A method for detecting the collaborative state of a track in a multi-index self-organizing network of the present invention establishes a multi-source point cloud model after preprocessing the collected data, and establishes a three-dimensional point cloud model using an iterative algorithm, and establishes a predefined model after fusing using a data fusion algorithm, and integrates the three-dimensional point cloud model and the predefined model for consistency, extracts key characteristic parameters of the track, and combines them with image data to achieve all-round detection of the track, and simultaneously detects multiple indicators of the track state, accurately reflects the actual state of the track, makes up for the shortcomings of a single data source in terms of coverage and accuracy, and improves detection efficiency and accuracy.

[0055] 3. The present invention combines the track multi-index self-organizing network collaborative status detection equipment and method, which can collect multi-source track data in real time and process and analyze it, timely discover track deformation, unevenness and other problems, and evaluate the track characteristic parameters by combining historical data, preset thresholds and standard specification parameters, generate corresponding detection reports and early warning information, improve detection accuracy and efficiency, and reduce track operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A top view of a track multi-index self-organizing network collaborative state detection device according to an embodiment of the present invention;

[0057] Figure 2 This is a main view of a track multi-indicator self-organizing network collaborative state detection device according to an embodiment of the present invention;

[0058] Figure 3 This is a schematic diagram of multi-sensor data collection according to an embodiment of the present invention;

[0059] Figure 4 This is a flow chart of a method for detecting the collaborative state of a track multi-index self-organizing network according to an embodiment of the present invention;

[0060] Figure 5 The flowchart of establishing a predefined model of track structure state in an embodiment of the present invention.

[0061] 1. Mobile platform trolley; 2. Track to be tested; 3. Frame; 4. Limiting wheel; 5. Rail structure profile detection module; 6. Fastener installation status detection module; 7. Track plate crack detection module; 8. Tunnel disease detection module; 9. Posture position measurement module; 10. Host computer; 11. Power supply module. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0063] like Figures 1 to 3 As shown, an embodiment of the present invention provides a track multi-index self-organizing network collaborative state detection device, which mainly includes a mobile platform trolley 1 and a rail structure profile detection module 5, a fastener installation state detection module 6, a track plate crack detection module 7, a tunnel disease detection module 8, a posture position measurement module 9 and a host computer 10 arranged on the mobile platform trolley 1. Each detection module is connected to the host computer 10 through the same standard interface. The mobile platform trolley 1 includes a frame 3 and a limiting wheel 4. The limiting wheel 4 has four and is symmetrically arranged at the bottom of the frame 3. The limiting wheel 4 can move on the track 2 to be tested, thereby driving the mobile platform 1 trolley to move. The sliding speed of the mobile platform trolley 1 on the track 2 to be tested can reach 20km / h.

[0064] The rail structure profile detection module 5, the fastener installation status detection module 6 and the track plate crack detection module 7 are each provided with two and symmetrically arranged on the frame 3, for collecting point cloud data of the track 2 to be tested and the surrounding environment. The frame 3 is also provided with two symmetrically arranged power supply modules.

[0065] The rail structure profile detection module 5 is a three-dimensional laser scanner, the fastener installation status detection module 6 is a linear structured light sensor, the track plate crack detection module 7 is a laser displacement sensor, and the tunnel disease detection module 8 is a high-resolution camera for collecting image data of the track 2 to be tested and the surrounding environment. The attitude position measurement module 9 includes an inertial navigation combination, a GNSS position receiver and a photoelectric encoder, which is used to collect attitude data of the track 2 to be tested and the surrounding environment;

[0066] The host computer 10 includes a sensor control unit, a data preprocessing unit and a data analysis and recording unit, which are used to control the operation of each module and the running speed of the mobile platform car 1, so as to realize data collection and storage, processing and analysis, real-time display and briefing generation. The power module 10 is a driving motor, which is used to provide power and power for the entire device.

[0067] like Figure 4 As shown, a method for detecting the collaborative state of a track multi-indicator self-organizing network is implemented by using a track multi-indicator self-organizing network collaborative state detection device, and includes the following steps:

[0068] S1. Place the mobile platform trolley 1 on the track 2 to be tested, turn on the power, perform sensor self-test through the host computer 10 to ensure that each sensor and detection module can work normally, start the mobile platform trolley 1, and the host computer 10 controls multiple detection modules to start synchronously. The mobile platform trolley 1 travels along the track 2 to be tested. The host computer 10 collects track point cloud data, image data and posture data of at least one detection module combined with the posture position measurement module according to the detection requirements.

[0069] S2. The host computer 10 preprocesses the point cloud data and image data: The preprocessing includes denoising, filtering, data simplification and data interpolation of the point cloud data using pass-through filtering and statistical filtering methods to eliminate errors and redundant information in the point cloud data. It also includes correcting the image data, eliminating distortion and performing image enhancement to improve contrast clarity and improve the quality of two-dimensional and three-dimensional data.

[0070] The formula for the direct filter is:

[0071] Filtered Point Cloud={p∈P|xmin<x p <x max∧y min<y p <y max∧z min<z p <zmax}

[0072] Where: FilteredPointCloud means the filtered point cloud, P is the original point cloud dataset, p is any point in the point cloud, x p ,y p ,z p are the coordinates of point p on the x, y, and z axes respectively. xmin and xmax represent the minimum and maximum thresholds of point p on the x axis; ymin and ymax represent the minimum and maximum thresholds of point p on the y axis; zmin and zmax represent the minimum and maximum thresholds of point p on the z axis.

[0073] By setting the maximum and minimum thresholds on the coordinate axis to filter out points beyond these ranges, preliminary cleaning of the point cloud data can be achieved.

[0074] The statistical filtering formula is:

[0075] Outliers = {p||P||d p -μ|>n×σ}

[0076] Where: Outliers means outliers, P is the original point cloud dataset, p is any point in the point cloud, d pis the distance from point p to its nearest neighbor, μ and σ are the mean and standard deviation of the distances from all points to their nearest neighbors, respectively, and n is a user-defined constant, usually called the “standard deviation multiple”, which is used to determine the threshold of outliers.

[0077] S3. Use the point cloud fusion processing software of the track mobile measurement system to fuse the pre-processed point cloud data with the posture data to construct a multi-source point cloud model.

[0078] S4. Rotate and translate the multi-source point cloud model through the transformation matrix, use the iterative registration algorithm to accurately align the overall point cloud data of the rail structure, realize the overall registration of the track structure, and establish a three-dimensional point cloud model of the track structure. Specifically: The iterative nearest algorithm is to rotate and translate through the registration between points. The measurement standard is based on the least squares method, and the distance between points must be within a certain threshold range. Now let the point clouds to be registered be P and Q, and select any point p in the point cloud P i , find the distance p in the point cloud Q i The point q with the shortest Euclidean distance i , the corresponding points of the two are used to obtain the transformation matrix. After multiple iterations, the optimal transformation matrix is ​​obtained to make the point clouds P and Q coincide. It is also necessary to stop the iteration when the f(R, T) value is minimum, otherwise continue the iteration. The specific formula is:

[0079]

[0080] Where: f(R,T) is the optimal transformation matrix, k is a constant variable, R is the rotation transformation matrix, and T is the translation transformation matrix.

[0081] S5. Use data fusion algorithm to fuse and register point cloud data, obtain a full description of the actual state of the track, and establish a predefined model of the track structure state, such as Figure 5 As shown, the specific steps include:

[0082] S5-1, using feature point matching or image overlay point cloud method to achieve registration between point cloud and image;

[0083] The registration between point cloud and image is realized by using feature point matching method. Feature point matching is accomplished by identifying and matching the significant feature points in two sets of data sets. Feature points are corner points, edge points in point cloud or corner points, SIFT feature points in image. Assume P c is the set of feature points in the point cloud, P i is a set of feature points in the image. The goal of registration is to find a transformation T such that P c and P i At the transformed position P c ′=T(P c ) and P i′ is as close as possible, which is achieved by minimizing the following cost function, the formula is:

[0084]

[0085] Where: N is the number of matching points, ||.|| represents the Euclidean distance, and transformation T refers to a mathematical operation that can change the position of a point in space. This transformation includes operations such as translation, rotation, scaling, shearing, or a combination of them. In three-dimensional space, transformation T is often represented in the form of a matrix, especially a transformation matrix under homogeneous coordinates, so that translation and linear transformations (such as rotation and scaling) can be handled uniformly. In point cloud registration, T represents a rigid body transformation, that is, a transformation that includes rotation and translation, which can be represented by a homogeneous transformation matrix. The specific parameters of the transformation matrix T need to be determined according to the registration goal and the optimization process, such as by minimizing a certain distance metric between the point cloud and the image feature points.

[0086] S5-2. Combined with attitude data, the data fusion algorithms such as Kalman filter, Bayesian network, and deep learning are used to fuse and align the data to obtain a complete description of the actual state of the track; the Kalman wave data fusion algorithm formula is:

[0087]

[0088] Where: represents the state estimation at the kth time step; A is the state transfer matrix, B is the input matrix, represents the state estimate of the previous time step, i.e., k-1; u k is the input vector, representing the control input or external influence at the kth time step; K k represents the Kalman gain at the kth time step, which determines the influence of the measurement value on the state estimation; yk represents the measurement value at the kth time step; H is the measurement matrix; It represents the state predicted based on the state estimate and control input of the previous time step without considering the measurement value of the kth time step.

[0089] S5-3. According to the actual structural state of the track, a predefined model of the track structural state is established through a machine learning model: a random forest model is used to establish a prediction model of the track structural state. Assuming that there are m features X1, X2, ..., X m and a target variable Y, the training process formula is:

[0090]

[0091] Where: is the predicted orbital structure state, and θ is the model parameter, which is determined by the training process.

[0092] S6. Integrate the 3D point cloud model and the predefined model for consistency and extract key characteristic parameters of the track: Compare the differences between the point cloud data in the 3D point cloud model and the predefined model to check whether the two are consistent in shape, size, etc. For the consistent parts, the point cloud data can be fused with the predefined model to create a more complete and accurate 3D model. Use computer vision and machine learning technology to automatically identify and extract key features of the track, such as sleepers, rail joints, switches, etc. from the fused 3D model, and perform quantitative analysis on the identified features.

[0093] S7. Based on the extracted key characteristic parameters of the track and combined with image data recognition, the track status assessment and detection analysis are performed, which specifically includes the following steps:

[0094] S7-1, extract key track characteristic parameters from the two-dimensional image data after preprocessing in S2, as well as characteristic parameters such as track gauge, rail wear, fastener tightening force, ballast bed condition and sleeper damage extracted from the three-dimensional model after fusion in S6, set the normal range of each characteristic parameter according to the standard specification, if the characteristic parameter value exceeds the normal range, mark it as abnormal and proceed to the next step, if it does not exceed the normal range, continue monitoring;

[0095] S7-2. Through data visualization, trend analysis or correlation analysis, deeply analyze the root causes of data anomalies.

[0096] S8. Generate a track status inspection report by combining historical data, standard parameters and inspection analysis results. For example, if gauge anomalies are frequently detected in a certain area, it may be due to unstable trackbed or severe rail wear in that area. The diagnosis process may require combining historical maintenance records, environmental data, train operation logs and other information.

[0097] When conducting track shape and position detection, the collected track shape and position point cloud data can be divided into different categories, such as straight segments, curve segments, switch areas, etc., and compared with standard track shape and position or historical data, the difference can be calculated, and combined with the type of track disease characteristics, a qualitative evaluation of the overall track condition can be conducted to determine whether there is obvious deformation, wear or defects.

[0098] The threshold or standard range is set in combination with the parameter variation range, and compared with the track shape and position parameters to judge the track shape and position status: after the qualitative analysis is completed, statistical methods are used to perform descriptive statistics and inferential statistics on the data, such as mean, standard deviation, regression analysis, variance analysis, etc.; for track smoothness indicators such as height unevenness and directional unevenness, geometric dimensional indicators such as gauge, superelevation, and track direction used to evaluate the geometric dimensional accuracy of the track, and wear and defect indicators used to evaluate the degree of wear of the track and possible safety hazards, indicators are calculated, evaluated and compared according to standard values ​​and empirical values ​​to determine whether the track shape and position meet the requirements, and finally the measurement and calculation results are visualized using tables, images, etc., to facilitate intuitive understanding and analysis of track shape and position information.

[0099] The standard deviation calculation formula is:

[0100]

[0101] Where: x i represents the i-th value in the data set, μ represents the mean of the data set, which is calculated as the sum of all values ​​divided by the total number of values, and n represents the total number of values ​​in the data set.

[0102] The invention discloses a method and device for detecting the state of a track in a multi-index self-organizing network. The method and device establish a multi-source point cloud model after preprocessing the collected data, establish a three-dimensional point cloud model by using an iterative algorithm, establish a predefined model after fusion by using a data fusion algorithm, integrate the three-dimensional point cloud model and the predefined model for consistency, extract key characteristic parameters of the track, and perform state evaluation and detection analysis on the track; generate a track state detection report in combination with historical data and detection and analysis results; the device comprises a rail structure profile detection module, a fastener installation state detection module, a track plate crack separation detection module, a tunnel disease detection module, a posture position measurement module and a host computer, which are arranged on a mobile platform trolley, realize simultaneous detection of multiple indicators of the track state, accurately reflect the actual state of the track, make up for the shortcomings of a single data source in terms of coverage and precision, and improve detection efficiency and accuracy.

[0103] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting the collaborative state of a track multi-index self-organizing network, characterized in that: The detection is implemented by using a track multi-index self-organizing network system state detection device, the detection device comprising a rail structure profile detection module (5) arranged on a mobile platform trolley (1), a fastener installation state detection module (6), a track plate crack separation detection module (7), a tunnel disease detection module (8), a posture position measurement module (9) and a host computer (10), and comprising the following steps: S1, the mobile platform trolley (1) travels along the track (2) to be tested, and the host computer (10) collects track rail point cloud data, image data and posture data of at least one detection module combined with the posture position measurement module (9) according to the detection requirements; S2, the host computer (10) pre-processes the point cloud data and the image data; S3, the pre-processed point cloud data is fused with the posture data to construct a multi-source point cloud model; S4, rotating and translating the multi-source point cloud model through the transformation matrix, using the iterative registration algorithm to accurately align the overall point cloud data of the rail structure, realizing the overall registration of the track structure, and establishing a three-dimensional point cloud model of the track structure; S5. Using the data fusion algorithm to fuse and register the point cloud data, a full description of the actual state of the track is obtained, and a predefined model of the track structure state is established, which specifically includes the following steps: S5-1, use feature point matching or image overlay point cloud method to achieve registration between point cloud and image; S5-2, combined with attitude data, the Kalman filter data fusion algorithm, Bayesian network algorithm, and deep learning data fusion algorithm are used to fuse and align the data to obtain a complete description of the actual state of the track; S5-3. According to the actual structural state of the track, a predefined model of the structural state of the track is established through a machine learning model; S6, integrating the three-dimensional point cloud model and the predefined model to extract key characteristic parameters of the track; S7. Based on the extracted key characteristic parameters of the track and combined with image data recognition, the track status assessment and detection analysis are performed, which specifically includes the following steps: S7-1. According to the extracted characteristic parameters of track gauge, rail wear, fastener tightening force, ballast bed condition and sleeper damage, the normal range of each characteristic parameter is set according to the standard specification. If the characteristic parameter value exceeds the normal range, it is marked as abnormal and the next step is carried out. If it does not exceed the normal range, the monitoring is continued; S7-2. Through data visualization, trend analysis or correlation analysis methods, deeply analyze the root causes of data anomalies; S8. Generate a track status inspection report by combining historical data, standard parameters and inspection analysis results.

2. A method for detecting the collaborative state of a track multi-index self-organizing network according to claim 1, characterized in that: In S2, the preprocessing includes denoising, filtering, data simplification and data interpolation of the point cloud data using pass-through filtering and statistical filtering methods to eliminate errors and redundant information in the point cloud data, and also includes correcting the image data, eliminating distortion and performing image enhancement.

3. A method for detecting the collaborative state of a track multi-index self-organizing network according to claim 2, characterized in that: The straight-through filtering formula is: Where: Filtered Point Cloud means the filtered point cloud, P is the original point cloud dataset, p is any point in the point cloud, x p ,y p ,z p are the coordinates of point p on the x, y, and z axes respectively, x min, x max represents the minimum and maximum thresholds of point p on the x-axis; y min, y max represents the minimum and maximum thresholds of point p on the y-axis; z min, z max represents the minimum and maximum thresholds of point p on the z-axis; By setting the maximum and minimum thresholds on the coordinate axis to filter out points beyond these ranges, the initial cleaning of the point cloud data can be achieved; The statistical filtering formula is: Where: Outliers means outliers, P is the original point cloud dataset, p is any point in the point cloud, d p is the distance from point p to its nearest neighbor, μ and σ are the mean and standard deviation of the distances from all points to their nearest neighbors, respectively, and n is a user-defined constant representing "standard deviation multiple" used to determine the threshold for outliers.

4. A method for detecting the collaborative state of a track multi-index self-organizing network according to claim 1, characterized in that: In S3, the pre-processed point cloud data and the posture data are fused using the track mobile measurement system point cloud fusion processing software to construct a multi-source point cloud model.

5. A method for detecting the collaborative state of a track multi-index self-organizing network according to claim 1, characterized in that: In S5-1, the registration between the point cloud and the image is achieved by using the feature point matching method. Assume P c is the set of feature points in the point cloud, P i is a set of feature points in the image. The goal of registration is to find a transformation T , so that P c and P i After the transformation P c ′= T ( P c )and P i ′ is as close as possible, which is achieved by minimizing the following cost function, the formula is: Where: N is the number of matching points, represents the Euclidean distance.

6. A method for detecting the collaborative state of a track multi-index self-organizing network according to claim 1, characterized in that: In S5-2, the Kalman filter data fusion algorithm formula is: Where: Indicated in k State estimation at a time step; A is the state transition matrix, B is the input matrix, Indicates the previous time step k −1 state estimate; u k is the input vector, indicating that k Control input or external influence of the time step; K k Indicates k The Kalman gain of the time step determines the influence of the measurement value on the state estimate; y k Indicates k The measured value of the time step; H is the measurement matrix; Indicates that without considering the k In the case of a time step measurement, the state predicted by the control input is estimated based on the state at the previous time step; In S5-3, a random forest model is used to establish a prediction model for the track structure state, assuming that m Features X 1, X 2,..., X m and a target variable Y , the training process formula is: Where: is the predicted orbital structural state, θ are model parameters, determined by the training process.

7. A track multi-index self-organizing network collaborative state detection device, applied to a track multi-index self-organizing network collaborative state detection method according to any one of claims 1 to 6, characterized in that: It also includes a power module (11), the mobile platform trolley (1) includes a frame (3) and a limiting wheel (4), the rail structure profile detection module (5), the fastener installation status detection module (6), the track plate crack detection module (7), the tunnel disease detection module (8), the posture position measurement module (9) and the host computer (10) are all arranged on the frame (3), the limiting wheel (4) has at least two and is symmetrically arranged at the bottom of the frame (3), the limiting wheel (4) can move on the track (2) to be tested, and the rail structure profile detection module (5), the fastener installation status detection module (6), the track plate crack detection module (7), the tunnel disease detection module (8), and the posture position measurement module (9) are respectively connected to the host computer (10); The rail structure profile detection module (5), the fastener installation status detection module (6) and the track plate crack detection module (7) are at least two and are symmetrically arranged on the frame (3); The posture position measurement module (9) is used to collect posture data of the mobile platform vehicle (1); The host computer (10) is used to control the operation of each module and control the operating speed of the mobile platform vehicle (1); The power module (11) is used to supply power to the entire device.

8. A track multi-index self-organizing network collaborative state detection device according to claim 7, characterized in that: The rail structure profile detection module (5) is a three-dimensional laser scanner; The fastener installation status detection module (6) is a line structured light sensor; The track slab crack detection module (7) is a laser displacement sensor; The tunnel disease detection module (8) is a high-resolution camera; The attitude position measurement module (9) comprises an inertial navigation combination, a GNSS position receiver and a photoelectric encoder; The host computer (10) comprises a sensor control unit, a data preprocessing unit and a data analysis and recording unit; The power modules (11) are two drive motors symmetrically arranged on the frame (3).

Images