Acoustics and vision fused intelligent inspection system and method for railway vehicle

Abstract

The invention provides a railway vehicle defect detection method, which comprises the following steps: constructing an acoustic-visual fusion inspection system, and combining with an inspection robot to detect the defects of multiple parts of a train; the system comprises a system triggering module, a defect positioning module, an acoustic signal acquisition module, a visual data acquisition module, a data transmission module, an acoustic signal processing module, a visual data processing module and a feature fusion module. On the acoustic part, defect positioning and defect diagnosis are realized by collecting acoustic signals, filtering irrelevant information and extracting characteristic frequency; for the visual part, collecting image data, filtering the data, extracting feature points, and registering to carry out defect detection on point cloud data; for a specific part, acoustic and visual feature fusion needs to be carried out, and then defects are diagnosed; the method is suitable for defect detection of the rail train, defect detection of a plurality of key parts of the train can be covered through sound and vision fusion, the detection precision is improved, and the labor burden is relieved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection for rail vehicles, and particularly to methods for detecting defects in components such as bearings, bolts, wheels, and surfaces of rail vehicles. Background Art

[0002] With the rapid development of railway transportation, the safety and reliability of trains have become important concerns. Bearings, bolts, surfaces, wheels, etc. of trains are key components during their operation. Any defect in these components may lead to equipment failures and even serious safety accidents in severe cases. Therefore, it is particularly important to detect defects in these components in a timely and effective manner, especially during the operating state.

[0003] Most traditional train detection methods rely on manual inspections or simple sensor monitoring. These methods have limitations such as insufficient ability to identify subtle defects and low detection efficiency. In recent years, with the continuous progress of sensing technology and data analysis technology, acoustic signals and three-dimensional point cloud data have become new means of defect detection, with high precision and reliability.

[0004] Acoustic signal detection captures the sound waves generated by a train during operation through sensors. By analyzing characteristics such as the frequency spectrum and amplitude of the sound waves, abnormal vibrations or faults in components such as bearings and bolts can be detected, and potential problems can be pre-warned. The advantage of acoustic signal detection is that it can monitor in real time and does not rely on the train being out of service.

[0005] Three-dimensional point cloud data technology is based on devices such as lidar or binocular cameras. By capturing the three-dimensional shapes of the train's surface and components, an accurate point cloud model is constructed. By processing and analyzing the point cloud data, defects such as cracks, wear, and deformation on the surfaces of components such as bearings, wheels, and bolts can be identified. Compared with traditional two-dimensional images or manual inspections, three-dimensional point cloud data can provide more comprehensive and detailed spatial information to help locate the specific position and nature of the defects.

[0006] Combining acoustic signals and three-dimensional point cloud data can not only improve the accuracy of defect detection but also achieve multi-dimensional and multi-angle monitoring, thus significantly enhancing the safety and maintenance efficiency of train operation. The application of this technology is expected to promote the intelligent detection and predictive maintenance of railway equipment, providing strong guarantees for the safe operation of trains. Summary of the Invention

[0007] To solve the above problems, the present invention proposes an intelligent inspection system and method for rail vehicles that integrates acoustics and vision, which can comprehensively and effectively detect defects in multiple components of a train by being mounted on an inspection robot.

[0008] An intelligent inspection system and method for rail vehicles integrating acoustics and vision. The method includes the following steps:

[0009] S0. Construct a detection system integrating sound and vision: The system includes: a defect location module that locates defect points of rail vehicles using sound and vision data; an acoustic signal acquisition module that acquires acoustic signal data of various components of rail vehicles; a visual data acquisition module that acquires image data of the wheels, surfaces, and some bolts of rail vehicles; a system trigger module that triggers the visual data acquisition module and the acoustic signal acquisition module to start working by detecting the approach of a train; a data transmission module that transmits data to an acoustic signal processing module and a visual signal processing module; an acoustic signal processing module that processes the acquired acoustic signals, filters out irrelevant information, extracts characteristic frequencies, determines the components where defects occur, and realizes defect classification; a visual signal processing module that processes the acquired image data, filters the image data, extracts feature points, generates a point cloud data model, and determines the locations where defects occur after registration for defect detection.

[0010] S1. Obtain rail vehicle detection data. The detection data includes: system trigger module sensor data, acoustic signal data, and visual image data:

[0011] S2. Process the detected acoustic signals, including the following steps:

[0012] When the system trigger module detects the approach of a train, start collecting data. Perform autocorrelation noise reduction on the acquired acoustic signal data to reduce the influence of random noise in the original signal;

[0013] Perform mean removal on the signal after autocorrelation noise reduction to eliminate the DC component in the signal;

[0014] Perform Hilbert transform to calculate the envelope signal;

[0015] Perform fast Fourier transform on the envelope signal and take the amplitude spectrum as its envelope spectrum;

[0016] S3. Determine the components where defects occur by comparing the envelope spectrum with the fault characteristic frequency ranges of different components;

[0017] If the defect occurs in the wheels or bolts, transmit the extracted characteristic information to the feature fusion module for decision fusion with visual feature information through a two-stream neural network for defect diagnosis; if the defect occurs in components such as bearings and other components where visual images cannot be acquired, determine the fault classification threshold through the calculated theoretical characteristic frequency;

[0018] S4. Process the detected visual data, including the following steps:

[0019] When the system trigger module detects the approach of a train, data collection begins; the collected image data is subjected to Gaussian filtering to remove image noise and reduce the impact of noise on feature point extraction and matching;

[0020] Feature points are extracted from the denoised data, and the SIFT algorithm is used to complete the extraction and description of feature points; based on SIFT feature point matching, the FLANN algorithm is combined for rough registration to preliminarily align two images;

[0021] For the coarsely registered point cloud data, the ICP algorithm is used for fine registration to further improve the registration accuracy and generate a point cloud model;

[0022] S5. According to the point cloud model generated in step S4, determine the component where the defect occurs;

[0023] If the defect occurs in the wheel or bolt, the extracted feature information is transmitted to the feature fusion module, and decision fusion is performed with the acoustic feature information through a two-stream neural network for defect diagnosis; if the defect occurs in components such as the train surface where acoustic signals cannot be collected, local geometric analysis is performed on the data in the point cloud model to detect changes in curvature features or surface normal vectors to determine the type of defect.

[0024] S6. The inspection robot uploads the detected defect information to the cloud server, and the location information of the fault is displayed on the visualization platform through the digital twin system.

[0025] Furthermore, the sampling frequency of the acoustic signal should be greater than 25 kHz, and the sampling time should be more than 5 s.

[0026] Furthermore, the camera for collecting images should be a binocular industrial camera, equipped with an anti-vibration lens, with a pixel requirement of not less than 3 million and a frame rate of not less than 100 fps, and the train running speed should not exceed 20 m / s.

[0027] Furthermore, a high-precision clock is embedded in the system acquisition module to ensure the matching of the timestamps of the two acquisition modules and guarantee data alignment.

[0028] Furthermore, the system trigger module uses optoelectronic sensors installed on both sides of the track. When a train approaches, it blocks the signal reception to determine whether the train enters the site.

[0029] Furthermore, the acoustic sensors are installed on the inspection robot in the form of a linear array.

[0030] Furthermore, the inspection robot is a wheeled robot, and the inspection path of the inspection robot is to drive straight along the track side.

[0031] Furthermore, the defect location module includes an acoustic location part, a visual location part, and an acoustic-visual location part. The acoustic location part uses information such as the phase difference and direction angle of the acoustic signal reaching the sensor array of the inspection robot, and adopts a beamforming algorithm to determine the fault location. The visual location part establishes a coordinate system in three-dimensional space and marks the specific location of the defect. In the acoustic-visual location part, the result of acoustic signal location is mapped into the point cloud coordinate system to correct the location error.

[0032] Compared with the prior art, one or more of the above technical solutions can achieve at least one of the following beneficial effects:

[0033] This detection device is mounted on the inspection robot and does not come into direct contact with the train, avoiding damage or deformation to the object under test caused by physical contact. At the same time, it avoids the disadvantages of difficult information acquisition and subsequent maintenance due to the installation of sensors in complex mechanical structures.

[0034] This detection method is based on the fusion of acoustic signals and visual data, combining the advantages of vision and acoustics, and has strong complementarity. Visual signals can intuitively reflect the appearance and structural changes of the wheels and the train surface as well as the state of some bolts, while acoustic signals can capture the dynamic fault characteristics of internal bearings and the loosening state of bolts. The combination of the two can detect the health status of the train more comprehensively and improve the detection accuracy at the same time. It can improve the efficiency of train health detection, and the fully automated detection greatly saves labor and time costs. Description of the Drawings

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0036] Figure 1 Schematic diagram of the rail vehicle defect detection device provided by the present invention.

[0037] Figure 2 Schematic diagram of the rail vehicle defect detection method provided by the present invention.

[0038] Figure 3 Schematic diagram of the working process of the inspection robot provided by the present invention. Detailed Embodiments

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] Embodiment 1

[0041] The system includes a system trigger module M01, an acoustic signal acquisition module M02, a visual data acquisition module M03, a data transmission module M04, an acoustic signal processing module M05, a visual data processing module M06, a feature fusion module M07, and a defect location module M08. A cloud server is set up. Among them, the acoustic signal processing module M05, the visual data processing module M06, and the feature fusion module M07 are deployed on the cloud server. In addition, the server also includes a scheduling module C01, a storage module C02, a digital twin visualization system C03, and a data transmission module C04. The acoustic signal acquisition module M02, the visual data acquisition module M03, the data transmission module M04, and the defect location module M08 are deployed on the inspection robot. The system trigger module M01 uses a photoelectric sensor and is separately deployed on both sides of the track. The inspection robot is deployed beside the track.

[0042] When the train enters the detection area, the information will be captured in advance by the photoelectric sensors installed on both sides of the track, and the system trigger module M01 starts to work. At the same time, the acoustic signal acquisition module M02 and the visual data acquisition module M03 are triggered to start working.

[0043] After the train enters the detection area, it will maintain a low-speed driving state. Abnormal acoustic signals will be generated by components such as defective bearings, loose bolts, and defective wheels during operation, which are captured by the acoustic signal acquisition module M02 and the signals are transmitted to the data transmission module M04. At the same time, the binocular industrial camera on the robot collects image data of the wheels, the surface of the train, and some bolts, and the visual data acquisition module M03 transmits the data to the data transmission module M04.

[0044] The acoustic signal acquisition module M02 uses a cardioid directional capacitive microphone with a relatively high detection accuracy to capture subtle sound changes within a certain range directly in front. The installation method is in a linear array and is installed on one side of the inspection robot.

[0045] The visual data acquisition module M03 includes a binocular industrial camera to capture image data of the wheels and the surface of the train. Since the train's movement will cause large vibrations, an anti-vibration lens is equipped to provide clear images.

[0046] The data transmission module M04 includes a communication interface module. The communication interface module is used to transmit data to the acoustic signal processing module M05, the visual data processing module M06, and the storage module C02 in the cloud server. The data transmission uses the TCP / IP protocol, and the communication interface module needs to have strong anti-interference ability.

[0047] The acoustic signal processing module M05 performs noise reduction processing on the collected acoustic signals, improves the signal-to-noise ratio of the signals, extracts the characteristic frequencies of the signals and compares them with the theoretical characteristic frequencies for defect diagnosis. If there are defects, the defect information is transmitted to the data transmission module C04 or the feature fusion module M07 according to the location where the defects occur. The visual data processing module M06 analyzes and processes the collected image data, establishes a point cloud model, performs local geometric analysis, and transmits the defect information to the data transmission module C04 or the feature fusion module M07.

[0048] The feature fusion module M07 uses a two-stream network to perform feature fusion on the received acoustic information and visual information, jointly diagnoses defects, and transmits the information to the data transmission module C04.

[0049] The data transmission module C04 transmits the information to the defect location module M08 on the inspection robot. Its structure is the same as that of the data transmission module M04 on the inspection robot.

[0050] In the defect location module M08, the acoustic part uses the received fault information, phase difference, direction angle and other information, and adopts the beamforming algorithm to determine the fault location. The visual part establishes a coordinate system in the three-dimensional space and marks the specific location of the defect; in the acoustic-visual positioning part, the result of acoustic signal positioning is mapped to the point cloud coordinate system to correct the positioning error. After the positioning is completed, the information is uploaded to the digital twin visualization system C03 in the cloud server.

[0051] In the storage module S02, the acoustic signals are stored in the WAV format, and the point cloud data is stored in the PLY format.

[0052] This embodiment provides an intelligent inspection system and method for rail vehicles integrating acoustics and vision. The detection method steps are as follows:

[0053] S0. Construct the above-mentioned rail vehicle defect detection device.

[0054] S1. Obtain rail vehicle detection data, and the detection data includes: system trigger module sensor data, acoustic signal data, visual image data.

[0055] S2. Process the detected acoustic signals, including the following steps:

[0056] When the system trigger module detects the approach of a train, data collection begins. The autocorrelation noise reduction is performed on the collected acoustic signal data to reduce the influence of random noise in the original signal;

[0057] The mean removal process is carried out on the signal after autocorrelation noise reduction to eliminate the DC component in the signal;

[0058] Perform Hilbert transform to calculate the envelope signal;

[0059] Perform fast Fourier transform on the envelope signal and take the amplitude spectrum as its envelope spectrum;

[0060] The specific steps of S2 are as follows:

[0061] S21. When the system trigger module detects the approach of a train, data collection begins. The autocorrelation noise reduction is performed on the collected acoustic signal data to reduce the influence of random noise in the original signal of the faulty bearing and enhance the periodic characteristics of the bearing fault;

[0062] S211. Autocorrelation noise reduction: The expression of the autocorrelation function is:

[0063]

[0064] where \(x(t)\) is the original signal; \(\tau\) is the time delay step; \(T\) is the signal period; \(R(\tau)\) is the autocorrelation function of the original signal;

[0065] S22. Mean removal process: Take the average value of the signal data after autocorrelation noise reduction, and then subtract the average value from the signal to remove the DC component;

[0066] S23. Calculate the Hilbert amplitude envelope: The Hilbert amplitude envelope can be obtained from the modulus of the analytic signal \(z(t)\) of the original signal \(x(t)\). The expression of the Hilbert amplitude envelope is:

[0067]

[0068] where \(x(t)\) is the original signal, is the Hilbert transform of the original signal;

[0069] S24. Perform Fourier transform to obtain the envelope spectrum: Perform Fourier transform on the envelope signal to obtain the Hilbert envelope spectrum;

[0070] S3. According to the fault characteristic frequency ranges of different components, determine the component where the defect occurs by comparing with the envelope spectrum;

[0071] When defects occur in the wheels or bolts, the extracted feature information is transmitted to the feature fusion module, and decision fusion is performed with the visual feature information through a two-stream neural network for defect diagnosis; when defects occur in the bearings and other components where visual images cannot be collected, the theoretical feature frequencies are calculated to determine the grading thresholds for faults.

[0072] S4. Process the detected visual data, including the following steps:

[0073] When the system trigger module detects the approach of a train, start collecting data; perform Gaussian filtering on the collected image data to remove image noise and reduce the impact of noise on feature point extraction and matching.

[0074] Extract feature points from the denoised data, and use the SIFT algorithm to complete the extraction and description of feature points; based on the feature point matching of SIFT, combine the FLANN algorithm for rough registration to initially align the two images.

[0075] For the point cloud data after rough registration, use the ICP algorithm for fine registration to further improve the registration accuracy and generate a point cloud model.

[0076] The above S4 steps specifically include:

[0077] S41. Gaussian filtering: The core is the Gaussian function, and the two-dimensional form is as follows:

[0078]

[0079] Among them, σ is the standard deviation, which determines the smoothness of the filter. The shape of the Gaussian kernel is a two-dimensional Gaussian distribution, and the weight decreases as the distance from the center point increases.

[0080] By convolving the Gaussian kernel with the image, the smoothing process of the image is realized, and the edge information of the image can be retained as much as possible.

[0081] S42. SIFT algorithm: The SIFT algorithm includes two steps: feature point extraction and feature point description, and can extract stable feature points at different scales and angles; although noise reduction processing is built into the algorithm, it is still necessary to denoise the image in advance under complex working conditions.

[0082] S43. FLANN algorithm: After using the SIFT algorithm to extract the feature point descriptors, use the K-nearest neighbor algorithm for matching, and screen out good matches by setting the distance ratio threshold; then screen out good matches; calculate the transformation matrix using the matched feature points, and achieve rough registration according to the transformation matrix.

[0083] S44. ICP algorithm: Use the Kd-Tree to search for corresponding points between the source point cloud P and the target point cloud Q, solve the optimal rotation matrix and translation matrix, and then perform coordinate transformation on the point cloud data according to the rotation matrix and translation matrix to obtain the square of the Euclidean distance as the average error. The average error formula is as follows:

[0084]

[0085] where R is the rotation matrix, t is the translation vector, p i is a point in the source point cloud P, and q i is a point in the target point cloud Q.

[0086] S5. According to the point cloud model generated in step S4, determine the component where the defect occurs;

[0087] If the defect occurs in components such as wheels and bolts, transmit the extracted feature information to the feature fusion module, and perform decision fusion with the acoustic feature information through a two-stream neural network for defect diagnosis; if the defect occurs in components such as the train surface where acoustic signals cannot be collected, perform local geometric analysis on the data in the point cloud model, detect changes in curvature features or surface normal vectors, and determine the defect type.

[0088] S6. The inspection robot uploads the detected defect information to the cloud server, and displays the location information of the fault on the visualization platform through the digital twin system.

[0089] According to the parameters of each component of the rail vehicle, determine the frequency threshold for classifying the defect types of each component, distinguish different defects according to different thresholds, collect acoustic data and visual data under different working conditions, extract feature information, and determine the defect type and defect degree. The specific defect types are shown in Table 1:

[0090] Table 1 Rail vehicle defect occurrence table

[0091]

[0092] Obviously, the above embodiments are only examples for clearly explaining the technical solutions of the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. An intelligent inspection system and method for rail vehicles integrating acoustics and vision, characterized in that, The method includes the following steps: S0. Construct a detection system for acoustic-visual fusion: The system includes: a defect location module that locates defect points of the rail vehicle using acoustic-visual data; an acoustic signal acquisition module that acquires acoustic signal data of each component of the rail vehicle; a visual data acquisition module that acquires image data of the wheels, surface, and some bolts of the rail vehicle; a system trigger module that triggers the visual data acquisition module and the acoustic signal acquisition module to start working by detecting whether a train is approaching; a data transmission module that transmits the data to an acoustic signal processing module and a visual signal processing module; an acoustic signal processing module that processes the acquired acoustic signals, filters out irrelevant information, extracts characteristic frequencies, determines the components where defects occur, and realizes defect classification; a visual signal processing module that processes the acquired image data, filters the image data, extracts feature points, generates a point cloud data model, and determines the location where defects occur after registration for defect detection; S1. Obtain rail vehicle detection data, where the detection data includes: system trigger module sensor data, acoustic signal data, and visual image data: S2. Process the detected acoustic signals, including the following steps: When the system trigger module detects that a train is approaching, start collecting data. Perform autocorrelation noise reduction on the acquired acoustic signal data to reduce the influence of random noise in the original signal; Perform mean removal processing on the signal after autocorrelation noise reduction to eliminate the DC component in the signal; Perform Hilbert transform to calculate the envelope signal; Perform fast Fourier transform on the envelope signal and take the amplitude spectrum as its envelope spectrum; The specific steps of step S2 above include: S21. When the system trigger module detects that a train is approaching, start collecting data; perform autocorrelation noise reduction on the acquired acoustic signal data to reduce the influence of random noise in the original signal; S211. Autocorrelation noise reduction: The expression of the autocorrelation function is: where x(t) is the original signal; τ is the time delay step; T is the signal period; R(τ) is the autocorrelation function of the original signal; S22. Mean removal processing: Take the average value of the signal data after autocorrelation noise reduction, and then subtract the average value from the signal to remove the DC component; S23. Calculate the Hilbert amplitude envelope: The Hilbert amplitude envelope can be obtained from the modulus of the analytic signal z(t) of the original signal x(t), and the expression of the Hilbert amplitude envelope is: where x(t) is the original signal, is the Hilbert transform of the original signal; S24. Perform Fourier transform to obtain the envelope spectrum: Perform Fourier transform on the envelope signal to obtain the Hilbert envelope spectrum; S3. According to the fault characteristic frequency ranges of different components, determine the components where defects occur by comparing with the envelope spectrum; If the defect occurs in the wheels or bolts, transmit the extracted characteristic information to the feature fusion module, and perform decision fusion with the visual feature information through a two-stream neural network for defect diagnosis; if the defect occurs in components such as bearings and other components where visual images cannot be collected, determine the grading threshold of the fault through the calculated theoretical characteristic frequencies; S4. Process the detected visual data, including the following steps: When the system trigger module detects the approach of a train, data collection begins; the collected image data is subjected to Gaussian filtering to remove image noise and reduce the impact of noise on feature point extraction and matching; Feature points are extracted from the denoised data, and the SIFT algorithm is used to complete the extraction and description of feature points; based on the feature point matching of SIFT, the FLANN algorithm is combined for rough registration to preliminarily align two images; For the coarsely registered point cloud data, the ICP algorithm is used for fine registration to further improve the registration accuracy and generate a point cloud model; The specific steps of the above S4 are as follows: S41. Gaussian filtering: The core is the Gaussian function, and the two-dimensional form is as follows: Among them, σ is the standard deviation, which determines the smoothness of the filter; the shape of the Gaussian kernel is a two-dimensional Gaussian distribution, and the weight decreases as the distance from the center point increases; By convolving the Gaussian kernel with the image, the smoothing process of the image is realized, and the edge information of the image can be retained as much as possible; S42. SIFT algorithm: The SIFT algorithm includes two steps: feature point extraction and feature point description, and can extract stable feature points at different scales and angles; although there is built-in noise reduction processing in the algorithm, it is still necessary to perform noise reduction processing on the image in advance under complex working conditions; S43. FLANN algorithm: After using the SIFT algorithm to extract the feature point descriptors, the K-nearest neighbor algorithm is used for matching, and good matches are screened out by setting the distance ratio threshold; then good matches are screened out; the transformation matrix is calculated using the matched feature points, and rough registration is achieved according to the transformation matrix; S44. ICP algorithm: Use the Kd-Tree to search for corresponding points of the source point cloud P and the target point cloud Q, solve the optimal rotation matrix and translation matrix, and then perform coordinate transformation on the point cloud data according to the rotation matrix and translation matrix, and obtain the square of the Euclidean distance as the average error. The average error formula is as follows: where R is the rotation matrix, t is the translation vector, p i is a point in the source point cloud P, and q i is a point in the target point cloud Q; S5. According to the point cloud model generated in step S4, determine the component where the defect occurs; If the defect occurs in the wheel or bolt, the extracted feature information will be transmitted to the feature fusion module, and decision fusion will be performed with the acoustic feature information through a two-stream neural network for defect diagnosis; if the defect occurs in components such as the train surface where acoustic signals cannot be collected, the data in the point cloud model will be subjected to local geometric analysis to detect changes in curvature features or surface normal vectors to determine the type of defect; S6. The inspection robot uploads the detected defect information to the cloud server, and the location information of the fault is displayed on the visualization platform through the digital twin system.

2. The method for detecting defects of rail vehicles based on multi-source information according to claim 1, characterized in that The sampling frequency of the acoustic signal should be greater than 25 kHz, and the sampling time should be more than 5 s.

3. The method for detecting defects of rail vehicles based on multi-source information according to claim 1, wherein, The camera for collecting images should select a binocular industrial camera, use an anti-vibration lens, the pixel requirement is not less than 3 million, the frame rate is not less than 100 fps, and the train running speed does not exceed 20 m / s.

4. The method for detecting defects of rail vehicles based on multi-source information according to claim 1, characterized in that A high-precision clock is embedded in the system acquisition module to ensure that the timestamps of the two acquisition modules match and ensure data alignment.

5. The method for detecting defects of rail vehicles based on multi-source information according to claim 1, wherein, The system trigger module uses a photoelectric sensor, which is installed on both sides of the track. When a train approaches, it blocks the signal reception to judge whether the train enters the field.

6. The method for detecting defects of rail vehicles based on multi-source information according to claim 1, characterized in that, The installation method of the acoustic sensor is installed in the form of a linear array on the inspection robot.

7. The method for detecting defects of rail vehicles based on multi-source information according to claim 1, characterized in that, The inspection robot is a wheeled robot, and the inspection path of the inspection robot is to drive straight along the side of the track.

8. The method for detecting defects of rail vehicles based on multi-source information according to claim 1, wherein, The defect location module includes an acoustic location part, a visual location part, and an acoustic-visual location part; the acoustic location part uses information such as the phase difference and direction angle of the sound signal reaching the sensor array of the inspection robot, and adopts a beamforming algorithm to determine the fault location; the visual location part establishes a coordinate system in three-dimensional space and marks the specific location of the defect; In the acoustic-visual location part, the result of acoustic signal location is mapped into the point cloud coordinate system to correct the location error.

Images