Track detection and maintenance method and device based on 3D vision, equipment and medium

Abstract

The invention provides a track detection and maintenance method and device based on 3D vision, equipment and a medium, and the method comprises the steps that a data collection module obtains point cloud data of a ground track in real time, and inputs the point cloud data to a preprocessing module; the preprocessing module preprocesses the point cloud data to obtain track data, and inputs the track data to the feature extraction module; the feature extraction module extracts feature data of the track data and inputs the feature data to the damage detection module; the damage detection module analyzes the feature data to generate detection data and inputs the detection data to the decision output module; and the decision output module generates a maintenance decision and a visualization result according to the detection data and a preset processing rule. The invention provides a real-time online, high-precision and intelligent track detection and maintenance method, the detection efficiency, continuity, stability and precision during track damage detection are remarkably improved, the maintenance efficiency and maintenance intelligence are improved, and the maintenance cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of logistics and warehousing management technology, and in particular to a method, apparatus, equipment and medium for track detection and maintenance based on 3D vision. Background Technology

[0002] In modern logistics warehousing systems, stacker cranes are the core equipment for automated storage, retrieval, and handling of goods. Their operational stability directly determines the efficiency and safety of the entire warehousing system. The stacker crane's running track, as its key load-bearing and guiding foundation, is in excellent condition. However, in actual high-intensity, high-frequency operations, the track system faces two major problems: First, the track surface, due to the continuous friction and impact from the stacker crane's high-speed reciprocating motion, is prone to uneven wear. This not only significantly increases running resistance and reduces energy efficiency but also causes abnormal shaking or deviation during the stacker crane's movement, severely affecting the positional accuracy of goods storage and retrieval, and even posing a significant safety risk of equipment jamming or goods falling. Second, the connecting screws used to fasten the track are prone to gradual loosening or even falling off under long-term vibration loads. This hidden danger directly leads to gaps or misalignments at the track connections, damaging the overall flatness and straightness of the track, further amplifying operational risks.

[0003] Currently, the industry primarily relies on periodic manual inspections or static measurements after shutdowns to monitor track conditions. Both methods suffer from inherent drawbacks such as low detection efficiency, high labor costs, and inability to cover all time periods. In particular, they struggle to provide real-time, continuous monitoring and early warning of dynamic faults such as track wear and loose screws during equipment operation. Consequently, repairs are often only carried out after problems have caused significant malfunctions or performance degradation, leading to high maintenance costs and long unplanned downtime. This severely hinders the continuous, stable, and efficient operation goals pursued by modern logistics and warehousing systems. Summary of the Invention

[0004] This invention provides a method, apparatus, computer equipment, and storage medium for track detection and maintenance based on 3D vision, aiming to solve the problems of discontinuity, instability, low efficiency, and high maintenance costs in existing track detection methods.

[0005] In a first aspect, embodiments of the present invention provide a track detection and maintenance method based on 3D vision. This method is applied to a detection and maintenance system including a data acquisition module, a preprocessing module, a feature extraction module, a damage detection module, and a decision output module. The method includes: the data acquisition module acquiring point cloud data of the ground track in real time and inputting the point cloud data into the preprocessing module; the preprocessing module preprocessing the point cloud data to obtain track data and inputting the track data into the feature extraction module; the feature extraction module extracting feature data from the track data and inputting the feature data into the damage detection module; the damage detection module analyzing the feature data to generate detection data and inputting the detection data into the decision output module; and the decision output module generating maintenance decisions and visualization results based on the detection data and preset processing rules.

[0006] Secondly, embodiments of the present invention also provide a track detection and maintenance device based on 3D vision, which includes a module for performing the above-described method.

[0007] Thirdly, embodiments of the present invention also provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0008] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the above-described method.

[0009] This application provides a 3D vision-based method, apparatus, computer equipment, and storage medium for track inspection and maintenance. By integrating a 3D acquisition device into the bottom of the transport equipment, real-time acquisition of track point cloud data is achieved. The 3D acquisition device collects data in real time as the transport equipment moves, eliminating the need for other equipment and reducing inspection costs. It also continuously and dynamically monitors the real-time status of the track, improving the continuity and real-time performance of the inspection and maintenance system, significantly enhancing the safety of transport equipment and the efficiency of track maintenance. Furthermore, preprocessing and feature extraction are used to obtain track feature data, improving the accuracy of basic data and the efficiency of subsequent analysis. Combined with geometric matching algorithms and machine learning models, defects such as track wear and loose screws are accurately identified, triggering graded warnings and maintenance recommendations. Therefore, this solution provides a real-time, online, high-precision, and intelligent track inspection and maintenance method, significantly improving the efficiency, continuity, stability, and accuracy of track damage detection, while also increasing maintenance efficiency, intelligence, and reducing maintenance costs. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A schematic flowchart illustrating the track detection and maintenance method based on 3D vision provided in an embodiment of the present invention; Figure 2 A schematic diagram of a sub-process of the track detection and maintenance method based on 3D vision provided in an embodiment of the present invention; Figure 3 A schematic diagram of a sub-process of the track detection and maintenance method based on 3D vision provided in an embodiment of the present invention; Figure 4 A schematic diagram of a sub-process of the track detection and maintenance method based on 3D vision provided in an embodiment of the present invention; Figure 5 A schematic diagram of a sub-process of the track detection and maintenance method based on 3D vision provided in an embodiment of the present invention; Figure 6 A schematic diagram of a sub-process of the track detection and maintenance method based on 3D vision provided in an embodiment of the present invention; Figure 7 A schematic diagram of a sub-process of the track detection and maintenance method based on 3D vision provided in an embodiment of the present invention; Figure 8 A schematic block diagram of a track detection and maintenance device based on 3D vision provided in an embodiment of the present invention; Figure 9 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0014] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0015] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0016] Please see Figure 1 This is a schematic flowchart illustrating a 3D vision-based track detection and maintenance method provided in this invention. In this application, the 3D vision-based track detection and maintenance method is applied to the field of automated logistics warehousing, and is particularly suitable for online dynamic monitoring and intelligent maintenance of stacker crane tracks. This method enables: real-time acquisition of track surface point cloud data during daily operation using a 3D radar installed at the bottom of the stacker crane; accurate identification of track wear, loose bolts, and other faults and quantification of damage severity through preprocessing, feature extraction, and damage detection; and generation of maintenance plans and visual reports by combining predictive analysis and decision output. This method represents a technological leap from manual inspection to online monitoring, and from post-construction repair to predictive maintenance, significantly improving the operational safety and continuity of the warehousing system.

[0017] This application provides a 3D vision-based method, apparatus, computer equipment, and storage medium for track inspection and maintenance. The 3D vision-based track inspection and maintenance method is applied to an inspection and maintenance system including a data acquisition module, a preprocessing module, a feature extraction module, a damage detection module, and a decision output module. The method includes: the data acquisition module acquiring point cloud data of the ground track in real time and inputting the point cloud data into the preprocessing module; the preprocessing module preprocessing the point cloud data to obtain track data and inputting the track data into the feature extraction module; the feature extraction module extracting feature data from the track data and inputting the feature data into the damage detection module; the damage detection module analyzing the feature data to generate detection data and inputting the detection data into the decision output module; and the decision output module generating maintenance decisions and visualization results based on the detection data and preset processing rules.

[0018] This application achieves real-time acquisition of track point cloud data by integrating a 3D acquisition device at the bottom of the transportation equipment. The 3D acquisition device collects data in real time as the transportation equipment moves, eliminating the need for other equipment and reducing inspection costs. It also continuously and dynamically monitors the real-time status of the track, improving the continuity and real-time performance of the inspection and maintenance system, significantly enhancing the safety of transportation and the efficiency of track maintenance. Furthermore, it obtains track feature data through preprocessing and feature extraction to improve the accuracy of basic data and the efficiency of subsequent analysis. Combining geometric matching algorithms and machine learning models, it accurately identifies defects such as track wear and loose screws, triggering graded warnings and maintenance recommendations. Therefore, this solution provides a real-time, online, high-precision, and intelligent track inspection and maintenance method, significantly improving the efficiency, continuity, stability, and accuracy of track damage detection, while also increasing maintenance efficiency, making maintenance more intelligent, and reducing maintenance costs.

[0019] Figure 1 This is a flowchart illustrating the track detection and maintenance method based on 3D vision provided in an embodiment of the present invention. Figure 1 As shown, the method is applied to a detection and maintenance system including a data acquisition module, a preprocessing module, a feature extraction module, a damage detection module, and a decision output module. The method includes the following steps S10-S50.

[0020] S10. The data acquisition module acquires point cloud data of the ground track in real time and inputs the point cloud data into the preprocessing module.

[0021] Specifically, the 3D vision-based track detection and maintenance method operates within a detection and maintenance system comprising a data acquisition module, a preprocessing module, a feature extraction module, a damage detection module, and a decision output module. The data acquisition module is responsible for acquiring and initially verifying the ground track point cloud data. This module can use a 3D line laser profilometer, a structured light camera, a laser scanner, or 3D radar to collect track data from the transport equipment. In this embodiment, 3D radar is used as the core sensor. This 3D radar is installed at a specific location on the bottom of the transport equipment facing the track, and it has high-frequency electromagnetic wave transmission and reception capabilities, covering the entire track area. It utilizes the reciprocating movement of the transport equipment during daily operation to achieve dynamic scanning of the ground track. In this embodiment, the transport equipment refers to a stacker crane.

[0022] More specifically, the 3D radar is fixed to the bottom crossbeam of the transport equipment to ensure that the scanning angle covers the width of the track, and is calibrated regularly (using the laser target method).

[0023] When the transport equipment runs along the track, the 3D radar continuously emits high-frequency electromagnetic waves into the ground track at a preset frequency and receives the signals reflected back from the track surface. Based on the time difference between the transmission and reception of electromagnetic waves and the spatial orientation information, the precise three-dimensional coordinates of each sampling point on the track surface are calculated, thereby generating point cloud data that can comprehensively reflect the geometry of the track. The point cloud data is a dense dataset composed of a large number of points containing spatial coordinate information. These data are like the "digital fingerprint" of the track, which finely depicts the real state of the track at different locations. At this time, the point cloud data includes the point cloud data of the track and the surrounding space.

[0024] The collected point cloud data is then used to generate raw point cloud files in .las / .ply / .txt format; finally, the point cloud files are validated (format checked) to ensure that the data format meets the requirements for subsequent processing.

[0025] Finally, the verified point cloud data is input into the preprocessing module in real time. The preprocessing module is used to preprocess the point cloud data to obtain standard track data, laying the foundation for the work of subsequent modules.

[0026] This embodiment introduces advanced 3D radar technology, which cleverly utilizes the daily operation of transportation equipment to achieve dynamic detection of the track, so as to efficiently and accurately obtain real-time status data of the track, thereby improving the accuracy and efficiency of damage detection, and eliminating the need to add other mobile equipment or manually scan the real-time status data of the track, thus reducing detection costs.

[0027] More specifically, the detection and maintenance system also includes a transmission unit that supports wired / wireless transmission (selecting Wi-Fi or Ethernet depending on the environment) and integrates data encryption and verification mechanisms. In this embodiment, an industrial-grade wireless AP or wireless optical communication is deployed in the warehouse to ensure a transmission rate of ≥100Mbps.

[0028] S20. The preprocessing module preprocesses the point cloud data to obtain track data, and inputs the track data into the feature extraction module.

[0029] Specifically, the point cloud data refers to a dataset of three-dimensional coordinate points captured by 3D radar that characterizes the track surface and surrounding spatial information; the track data refers to standardized data that has been preprocessed to remove invalid information and retain the core features of the track; the preprocessing module is a core module in the detection and maintenance system used to clean, simplify, and standardize the point cloud data to provide qualified data for subsequent feature extraction.

[0030] After receiving the point cloud data from the data acquisition module, the preprocessing module immediately initiates preprocessing to systematically process the raw point cloud data. The preprocessing module uses a preset processing algorithm or model to preprocess the point cloud data, ensuring that the preprocessed data accurately represents the true state of the track. After the complete preprocessing process, standardized track data is obtained, which represents the continuous morphology of the track surface. The preprocessing module then synchronously inputs this track data into the feature extraction module, providing reliable data support for subsequent extraction of track damage-related features.

[0031] In this embodiment, the preprocessing module simplifies the massive point cloud data into a series of standard track data that can characterize the continuous morphology of the track surface, so as to facilitate the accurate extraction of features from the actual track in the subsequent process.

[0032] In one embodiment, such as Figure 2 As shown, step S20 includes steps S21-S23.

[0033] S21. Perform point cloud format conversion and coordinate system unification processing on the point cloud data to generate first data; S22. Perform noise filtering and data simplification on the first data to generate the second data; S23. Perform track segmentation and coordinate alignment processing on the second data to generate the track data.

[0034] Specifically, this embodiment describes the specific preprocessing process of the preprocessing module.

[0035] First, the preprocessing module performs point cloud format conversion and coordinate system unification: the original point cloud data is usually stored in a sensor-specific binary format, which the preprocessing module converts into a universal point cloud data format so that subsequent algorithms can read and process it uniformly; at the same time, since the point clouds collected by the transportation equipment at different times during its movement are located in different local coordinate systems, the preprocessing module calculates the transformation matrix by extracting the overlapping feature regions between adjacent point cloud frames, and converts all point clouds to a unified coordinate system based on the track centerline or global reference point, eliminating the coordinate offset introduced by the change in the transportation equipment's attitude, and the first data is obtained after this processing.

[0036] Subsequently, the preprocessing module performs noise filtering and data simplification on the first data: First, noise filtering is performed on the first data, starting with statistical outlier removal—calculating the average distance between each point and its neighbors, and removing points whose average distance exceeds a global statistical threshold (e.g., mean plus three standard deviations); then, radius outlier removal is performed—a search radius is set for each point, and if the number of neighbors within that radius is less than a set threshold, the point is considered an isolated noise point and removed; finally, Gaussian filtering is used to smooth the point cloud, replacing the position coordinates of each point with the weighted average of its neighbors, with the weights determined by a Gaussian distribution, thereby reducing local surface fluctuations caused by sensor measurement noise and making the point cloud distribution smoother. Noise filtering removes noise and outliers from the point cloud data to improve data quality.

[0037] Then, the first data after noise filtering is simplified: first, voxel mesh downsampling is used—the 3D space is divided into a fixed-size cubic voxel mesh, and only one representative point is retained in each voxel (usually the centroid or center point of all points in the voxel), thereby quickly reducing the number of point clouds; next, curvature-preserving sampling is performed—on the basis of downsampling, the surface curvature changes of each region of the point cloud are calculated, and more points are retained in regions with drastic curvature changes (i.e., significant geometric features), ensuring that key shape information such as the edges of the track and pits are not lost during simplification; finally, uniform sampling is performed—the point cloud space is divided into uniformly spaced meshes to ensure that the retained points are more evenly distributed in space, avoiding subsequent analysis deviations caused by uneven density of the original scan. The second data is obtained after this processing. By simplifying the data, the amount of data is reduced while preserving the core features, thus improving processing efficiency.

[0038] Finally, the preprocessing module performs track segmentation and coordinate alignment on the second data: First, track segmentation is performed on the second data. Ground removal is first performed—by setting a height threshold or using a plane fitting algorithm, non-track point clouds belonging to the ground or the surrounding environment of the track are separated out, focusing on the track target; then, the DBSCAN clustering algorithm is used—based on point cloud density, spatially adjacent and density-connected points are clustered into one class, thereby distinguishing different targets such as tracks, shelf columns, and the structure of the transport equipment itself; then, Region of Interest (ROI) extraction is performed—based on the preset spatial range of the track in the scene, only point cloud data within this range is retained, further focusing on the track target; finally, track recognition is performed—combining the geometric features of the track cross-section (such as the outline shape of I-beams or T-rails) to accurately locate and segment the point cloud belonging to the track from the clustering results, i.e., the track point cloud; through the track segmentation stage, track-related regions are extracted from the above-processed point cloud data.

[0039] Then, coordinate alignment processing is performed on the second data after track segmentation. First, principal component analysis is used to calculate the principal direction of the track point cloud—eigenvalue decomposition is performed on the covariance matrix of the track point cloud, and the eigenvector corresponding to the largest eigenvalue is the extension direction of the track. Then, coordinate transformation is performed—the track point cloud is rotated with a rotation matrix until its principal direction is aligned with the coordinate axes of the world coordinate system. Finally, height normalization is performed—the average height or reference height of the top surface of the track is uniformly subtracted from the height coordinates of the track point cloud, so that the longitudinal extension direction of the track is aligned with the horizontal axis and the height direction is zeroed, thereby obtaining track data with standardized orientation and height reference.

[0040] The preprocessing module inputs the final generated track data into the feature extraction module for subsequent wear detection and analysis.

[0041] S30. The feature extraction module extracts the feature data of the track data and inputs the feature data into the damage detection module.

[0042] Specifically, the feature extraction module extracts key geometric features of the track from the preprocessed track data, providing a basis for damage detection. The track data is standardized point cloud data that has been preprocessed to remove invalid information and retain core track features. The feature data refers to the set of core feature parameters extracted from the track data that characterize the track's state (including potential damage).

[0043] Upon receiving the track data from the preprocessing module, the feature extraction module immediately initiates feature extraction. The entire process strictly focuses on track-related features, avoiding the extraction and processing of irrelevant data. Using a preset feature extraction algorithm, the module performs comprehensive and high-precision feature mining on the track data, accurately extracting core feature parameters related to track damage, such as track surface flatness, track connection gaps, and screw fixing point status. These parameters are then integrated to form complete feature data, ensuring that the extracted feature data comprehensively and accurately reflects the actual condition of the track, providing reliable support for subsequent damage detection.

[0044] After the feature data is extracted, the feature extraction module immediately inputs the feature data into the damage detection module for further detection and judgment of track damage.

[0045] In one embodiment, such as Figure 3 As shown, step S30 may include steps S31-S33.

[0046] S31. Slice along the extension direction of the track data to obtain track cross-sectional data; S32. Perform contour fitting on the track cross-section data and construct an actual track model; S33. Calculate the deviation between the actual orbit model and the standard orbit model, and calculate the feature data based on the deviation.

[0047] In this embodiment, the feature extraction module receives track data from the preprocessing module. This track data is a point cloud of the track surface after coordinate normalization, accurately describing the geometric shape of the track in three-dimensional space. The feature extraction module first slices the point cloud at equal intervals along the extension direction of the track data. Each slice corresponds to a cross-sectional position of the track. The slice thickness is set according to the track size and detection accuracy requirements, ensuring that each slice contains sufficient point cloud information to characterize the cross-sectional contour, thereby obtaining a series of track cross-sectional data. This track cross-sectional data is a subset of point clouds reflecting the lateral geometry of the track at different positions.

[0048] Subsequently, for each track cross-section data, the feature extraction module uses a curve fitting algorithm to fit the contour, specifically employing either polynomial fitting or spline fitting. Polynomial fitting approximates the point cloud distribution of the cross-section by setting a polynomial function of appropriate degree, while spline fitting uses piecewise low-degree polynomials to smoothly connect at the nodes, thus more flexibly describing complex contour shapes. The continuous curve obtained by fitting is the actual track contour at that cross-section.

[0049] Then, the feature extraction module integrates the actual track contours of all cross sections to construct a complete actual track model that represents the geometry of the current track surface. The actual track model is a three-dimensional track surface model composed of a series of continuous cross-sectional contours.

[0050] Simultaneously, the feature extraction module calls a preset standard track model—this model is constructed based on track design drawings or a reference point cloud collected in the track's brand-new state, representing the ideal geometric shape of the track in a wear-free state. After spatially registering and aligning the actual track model with the standard track model, the feature extraction module calculates the spatial positional difference between the two point by point or section by section to obtain the deviation value. The deviation value refers to the amount of offset of the actual contour relative to the ideal contour in the normal direction, and this offset accurately reflects the amount of wear or deformation of the track surface.

[0051] Finally, based on the calculated deviation value, the feature extraction module further statistically calculates and analyzes the geometric feature parameters of the track, including the maximum wear depth of each cross-section along the longitudinal direction of the track, the lateral width and longitudinal extension length of the wear area, the wear volume—obtained by integrating the wear area of ​​each cross-section along the longitudinal direction, and the offset position of the wear area relative to the track centerline. The feature extraction module integrates and encodes the above multi-dimensional geometric feature parameters to form feature data containing wear depth, width, length, volume, and position information, and inputs this feature data into the damage detection module for subsequent wear pattern recognition and severity assessment.

[0052] This embodiment extracts the geometric feature parameters of the track using the feature extraction model. The process is accurate and efficient, providing precise data for subsequent damage detection, thereby enabling precise maintenance of the track.

[0053] S40. The damage detection module analyzes the feature data to generate detection data and inputs the detection data into the decision output module.

[0054] Specifically, the core function of the damage detection module is to identify track wear areas and quantitatively assess the degree of wear, providing a precise basis for decision-making. The feature data refers to the set of track geometric feature parameters calculated based on the deviation between the actual track model and the standard model, output by the feature extraction module. The detection data refers to the standardized data generated by the damage detection module after analyzing the feature data, which characterizes the track damage area, damage type, damage degree, and overall track damage status.

[0055] After receiving the feature data input by the feature extraction module, the damage detection module immediately starts the analysis. This module uses a preset analysis algorithm to perform a comprehensive and detailed analysis of the feature data. Combining this with core parameters such as the deviation values ​​of various track positions contained in the feature data, it accurately identifies damaged areas on the track. Simultaneously, based on preset damage assessment standards, it quantitatively assesses the degree of damage to the identified damaged areas, clarifying the specific quantitative indicators of the damage.

[0056] Throughout the analysis process, the damage detection module continuously verifies the analysis results to avoid detection errors caused by data deviations, ensuring that it can comprehensively and accurately reflect the track damage and overall damage status, and ultimately generate complete and standardized detection data.

[0057] After the detection data is generated, the damage detection module immediately inputs the detection data into the decision output module, providing reliable detection support for the subsequent decision output module to make relevant decisions such as track maintenance and fault warning.

[0058] In one embodiment, such as Figure 4As shown, the detection data includes the location of damage, the type of damage, and the degree of damage. Step S40 may include steps S41-S44.

[0059] S41. Locate the damaged location from the feature data based on a preset positioning algorithm; S42. Based on preset classification rules, classify the damaged location according to the feature data to generate the corresponding damage type; S43. Select the corresponding preset analysis method to identify the degree of damage based on the damage type; S44. Input the damaged location, the damaged type, and the damaged extent into the decision output module.

[0060] Specifically, this embodiment describes the specific process of damage detection: First, the damage detection module locates the damage location from the feature data based on a preset positioning algorithm. In this embodiment, the preset positioning algorithm uses a threshold segmentation method to mark continuous spatial locations in the feature data where the wear depth exceeds a set threshold as candidate damage areas. Then, through connected component analysis, spatially adjacent candidate points are merged into independent damage areas, and the spatial coordinate range and geometric center position of each damage area are recorded to obtain accurate damage location information.

[0061] After locating the damage, the damage detection module classifies each damage location based on preset classification rules to generate a corresponding damage type. The preset classification rules include multi-level discrimination logic based on the geometric shape and spatial location of the damage location. The damage detection module first extracts the shape features of each damage location—including the elongation of the area, the ratio of area to perimeter, the offset distance relative to the track centerline, and the sharpness of the edges. If the damage location is located on the top surface of the track and its shape is strip-like and extends longitudinally along the track, it is classified as track wear—that is, track surface wear or peeling. If the damage location is located near the bolt mounting holes on both sides of the track and presents a circular or elliptical discrete distribution, it is classified as bolt anomaly—that is, local dents caused by loose, broken, or missing bolts. If the damage location does not meet the above two characteristics and manifests as an isolated point or block anomaly, it is classified as other types of faults—such as foreign objects attached to the track surface or local deformation.

[0062] Subsequently, the damage detection module selects a preset analysis method based on the damage type corresponding to each damage location to identify the degree of damage at that location. This process is divided into two stages: First, the damage degree quantification stage, where different damage types correspond to different preset analysis methods. For example, for track-related damage, the preset analysis methods include calculating the average depth, maximum depth, wear area, and wear volume of the wear region—obtained by integrating the difference between the point cloud within the region and the fitted reference surface. For bolt-related damage, the preset analysis methods include calculating the local subsidence depth around the bolt hole, the missing height of the bolt head relative to the track surface, and the local track misalignment caused by bolt loosening. For other types of faults, the preset analysis methods select the corresponding algorithm based on the specific abnormal morphology, such as foreign object height detection or local deformation curvature calculation. The specific analysis methods for different damage types mentioned above are merely examples; specific methods are not limited here. Developers can use different analysis methods to quantify the degree of damage based on different application scenarios.

[0063] After quantifying the damage level, the system enters the damage level identification stage. This involves the damage detection module assessing the severity of each damaged location. Based on the damage type and severity parameters, the detection results are compared with a preset multi-level threshold system. For example, in this embodiment, the threshold system includes a baseline threshold and a severity threshold. When the damage level is below the baseline threshold, it is considered a minor fault, and the system only logs the information without triggering a processing procedure. When the damage level exceeds the baseline threshold but is below the severity threshold, it is considered a moderate fault, and the system generates a planned shutdown suggestion, scheduling maintenance during a period that does not affect core operations. When the damage level exceeds the severity threshold, it is considered a critical fault, and the system triggers an emergency shutdown warning, requiring immediate interruption of equipment operation for emergency repairs. The different maintenance rules corresponding to different damage levels described above are merely examples; specific maintenance rules are not limited here. Developers can implement different maintenance rules based on different usage scenarios and damage levels.

[0064] Finally, the damage detection module integrates the above analysis results into detection data, which includes a structured set of information such as the damage location coordinates, damage type label, damage degree quantification parameters, and severity level for each damaged location. This detection data is then input into the decision output module for subsequent early warning and maintenance decisions.

[0065] In another embodiment, when the damage detection module analyzes the feature data, it first locates the damage position based on threshold segmentation and classifies the wear pattern according to the geometric characteristics of the damage position: if the wear is uniformly distributed laterally along the track and the depth changes gradually, it is identified as a uniform wear pattern; if the wear is concentrated in a certain section of the track and has an irregular shape, it is identified as a local wear pattern; if the wear mainly occurs on both sides of the track and extends in a band, it is identified as an edge wear pattern; if the wear manifests as discrete deep pit-like depressions, it is identified as a pitting wear pattern. Based on the identified wear patterns, the damage detection module further determines the corresponding damage type: uniform wear and local wear, because they occur in the bearing area of ​​the top surface of the track, are classified as track wear damage; edge wear, if it appears near bolt holes and extends in a band, is often accompanied by bolt loosening and is classified as bolt abnormality damage; pitting wear, because it manifests as fatigue spalling of surface material, is classified as other types of damage. The module then selects the corresponding parameter calculation method for different damage types, quantifies indicators such as wear depth, area, and volume, and finally generates detection data containing damage location, damage type, and damage degree.

[0066] This embodiment calculates the location, type, and extent of track damage using the damage detection module to accurately identify defects such as track wear and loose screws. This enables the decision output module to trigger tiered warnings and more precise maintenance recommendations, significantly improving the accuracy of maintenance plans generated by the detection and maintenance system.

[0067] In one embodiment, such as Figure 5 As shown, the damage types include track wear and bolt abnormality, and step S43 may include steps S431-S432.

[0068] S431. If the damage type is track wear, the track surface deviation value is calculated based on the feature data and preset reference data using a geometric matching algorithm, and the degree of damage is identified based on the track surface deviation value and a preset threshold. S432. If the damage type is a bolt anomaly, the bolt anomaly type is identified based on the feature data using a preset machine learning model, and then the corresponding damage level is matched based on the anomaly type.

[0069] Specifically, the preset benchmark data refers to standard track model data pre-stored in the system database; the geometric matching algorithm refers to an algorithm that calculates track surface deviation based on ICP (Iterative Closest Point) registration between point cloud data and standard track models; the preset machine learning model refers to a trained classifier (such as SVM or CNN) used to identify abnormal features of screw connection parts; and the preset threshold refers to multi-level judgment criteria set by the system (such as wear depth threshold and screw displacement threshold) used to trigger graded alarms and determine the degree of damage.

[0070] In this embodiment, after receiving feature data from the feature extraction module, the damage detection module first determines the damage type corresponding to each damaged location according to a preset classification rule. If the damage type is determined to be track wear, the damage detection module calls a geometric matching algorithm to identify the degree of damage: the geometric matching algorithm uses an iterative nearest-point registration method to spatially align the feature data of the current track point cloud with the reference data of the standard track model pre-stored in the system database. By iteratively optimizing the optimal transformation matrix between the two point sets, the actual track model formed by the current point cloud and the standard track model are maximized to overlap. After registration is completed, the algorithm calculates the spatial deviation of the current point cloud relative to the standard track model point by point to obtain the track surface deviation value at each position on the track surface—that is, the difference between the measured height and the standard height, where the measured height refers to the height value of the actual track model and the standard height refers to the height value of the standard track model.

[0071] Subsequently, the damage detection module identifies the degree of damage based on the comparison between the track surface deviation value and a preset threshold. In this embodiment, the preset threshold includes a first wear threshold and a second wear threshold, with the first wear threshold being less than the second wear threshold. If the track surface deviation value exceeds the second wear threshold, it is determined to be severe wear, corresponding to level 3 damage (i.e., severe severity). If it is below the first wear threshold, it is determined to be slight wear, corresponding to level 1 damage (i.e., slight severity). If the track surface deviation value is between the first and second wear thresholds, it is determined to be moderate wear, corresponding to level 2 damage (i.e., moderate severity). The above comparison method is merely illustrative, and specific comparison methods are not limited here.

[0072] If the damage type is determined to be bolt anomaly, the damage detection module calls a preset machine learning model to identify the degree of damage: First, it extracts a local point cloud subset of the bolt connection area from the feature data, and analyzes the degree of abrupt change in point cloud density, surface smoothness change, and point cloud continuity features in this area; These features are then input into a pre-trained machine learning model—this model is built based on support vector machines or convolutional neural networks and is trained on a large amount of sample data of bolts in normal and abnormal states. It can identify different anomaly types such as deformation, loosening, fracture, and displacement. Therefore, the machine learning model can identify the bolt anomaly type based on the feature data, and the anomaly type includes deformation, loosening, displacement, and fracture.

[0073] After the machine learning model outputs the classification results of bolt anomaly types, the damage detection module matches the corresponding damage level according to the anomaly type: In this embodiment, the anomaly type of deformation is classified as Level 1 damage (i.e., primary minor degree), the anomaly type of loosening is classified as Level 2 damage (i.e., secondary minor degree), the anomaly type of offset is classified as Level 3 damage (i.e., moderate degree), and the anomaly type of fracture is classified as Level 4 damage (i.e., severe degree).

[0074] After completing the above analysis, the damage detection module integrates the damage type, damage degree, and pre-located damage location of each damage location into complete detection data, and inputs it into the decision output module.

[0075] This embodiment uses different identification rules to identify the degree of damage of different damage types, accurately identifying defects such as track wear and loose screws, which can improve the identification accuracy and intelligence level, thereby improving the detection accuracy and intelligence of the detection and maintenance system.

[0076] Furthermore, this embodiment can also set graded alarms according to different degrees of damage, thereby further improving the intelligence of the detection and maintenance system.

[0077] In one embodiment, such as Figure 6 As shown, the detection and maintenance system also includes a predictive analysis module, and steps S433-S435 are included after step S43.

[0078] S433. Determine whether the degree of damage is considered severe. S434. If the severity is not severe, the detection data is input to the predictive analysis module. The predictive analysis module generates a predictive analysis result based on the detection data and historical fault data. The predictive analysis result is then input to the decision output module so that the decision output module generates maintenance decisions and visualization results based on the predictive analysis result and preset processing rules. S435. If the severity level is specified, the detection data is input to the decision output module so that the decision output module generates maintenance decisions and visualization results based on the detection data and preset processing rules.

[0079] Specifically, the predictive analysis module is the core module of the inspection and maintenance system that performs trend prediction and risk assessment based on the inspection data and historical data. The historical fault data refers to the track's past damage, faults, and maintenance-related data stored in the system database. The predictive analysis results include wear development trends, track remaining service life (RUL), and operational risk level. The maintenance decision refers to the maintenance plan generated by the system (e.g., routine maintenance or emergency maintenance), which includes fault information (including damage location, damage type, damage severity, etc.) and maintenance work orders. The visualization results refer to the inspection and prediction-related visualization data that is easy for maintenance personnel to view. The preset processing rules refer to the system's preset maintenance processing standards corresponding to different damage levels.

[0080] In this embodiment, the detection and maintenance system further includes a predictive analysis module. After the damage detection module completes the damage degree identification, it first determines whether the damage degree at each damaged location is severe—severity refers to the damage parameters exceeding a preset severe threshold, posing an imminent threat to the system's operational safety, requiring emergency shutdown.

[0081] If the damage is determined to be non-serious, i.e., minor or moderate damage, the damage detection module will input the detection data, including the location, type, and extent of the damage, into the predictive analysis module.

[0082] After receiving the detection data, the predictive analysis module first retrieves the historical fault data of the track section from the system database—that is, the time series data of parameters such as wear depth, area, and volume recorded in previous detections at this location—and performs time series analysis on the detection data and historical data—that is, analyzes the evolution of wear parameters in chronological order to identify their periodicity, trend, and other temporal characteristics.

[0083] Subsequently, the predictive analysis module performs trend analysis on the integrated data (i.e., historical fault data): it uses linear regression to fit the linear trend of wear parameters changing over time, and uses moving average to smooth short-term random fluctuations to obtain the medium- and long-term trend line of wear development.

[0084] Based on trend analysis, the predictive analysis module inputs the detection data and historical fault data into the pre-trained XGBoost machine learning prediction model. This model is built on the gradient boosting decision tree algorithm and achieves high-precision regression prediction by integrating multiple weak learners. It can capture the nonlinear characteristics of wear development and the coupled influence of multiple factors, and output the damage prediction values ​​for multiple future time points.

[0085] Based on trend analysis and model prediction results, the prediction analysis module further calculates the remaining service life of the track—that is, the remaining running time or number of operation cycles from the current moment until the wear reaches the preset failure threshold.

[0086] Subsequently, the predictive analysis module conducts a risk assessment—combining the predicted damage value with the remaining service life, calculating the probability distribution of track failures during future operating cycles using probability analysis methods, and comprehensively evaluating the track operation risk level based on the probability distribution of failures and the severity of their consequences.

[0087] The predictive analysis module integrates the above analysis results into a predictive analysis result—which includes structured information such as wear development trend curve, future damage prediction value, remaining service life estimate, and risk level, and inputs this result into the decision output module.

[0088] After receiving the predictive analysis results, the decision output module generates maintenance decisions and visualization results based on the predictive analysis results and preset processing rules: maintenance decisions include scheduling planned maintenance windows and formulating routine maintenance plans; visualization results include displaying wear prediction curves and remaining life countdowns in the form of trend charts on the monitoring interface, marking high-risk damage areas and damage types with heat maps, and generating inspection reports containing suggested maintenance times, and pushing them to the operation and maintenance terminal.

[0089] If the damage is determined to be severe, the damage detection module will directly input the detection data into the decision output module. The decision output module will select the corresponding preset processing rules based on the damage location, damage type, and damage severity to generate maintenance decisions and visualization results. The maintenance decisions at this time include immediately triggering multi-level audible and visual alarms, pushing emergency shutdown commands to the maintenance personnel's terminals, and automatically locking the operation permissions of the transportation equipment in the fault area. The visualization results include highlighting the location of the critical fault on the monitoring interface, popping up a detailed fault pop-up window containing the damage type and severity, and generating an emergency repair work order to send to the maintenance department.

[0090] This embodiment classifies operations according to the severity of damage. When the damage is not severe (minor or moderate), the predictive analysis module predicts the future maintenance cycle based on historical fault data, enabling the decision output module to generate a maintenance plan (such as "track section A needs to be inspected within 3 days"). This allows maintenance personnel to choose to perform track maintenance during periods when the transportation equipment is not in operation, avoiding impacting the operation of the transportation equipment and thus affecting transportation efficiency. It can be seen that the detection and maintenance system is highly intelligent.

[0091] If the damage is severe, the decision output module retrieves information based on the detection data and the handling solutions for problems in the database to generate an operation and maintenance report and output it to the user terminal, enabling emergency maintenance (such as shutdown for repair) to avoid safety hazards or more serious damage that would increase maintenance costs.

[0092] S50. The decision output module generates maintenance decisions and visualization results based on the detection data and preset processing rules.

[0093] Specifically, the decision output module is the core module of the detection and maintenance system that generates maintenance decisions and visualization results based on detection data; the preset processing rules refer to the system's preset operation and maintenance processing standards corresponding to different damage levels; the maintenance decisions refer to the routine maintenance or emergency maintenance plans generated by the system; and the visualization results refer to the detection-related visualization data that is easy for operation and maintenance personnel to view.

[0094] In this embodiment, after receiving detection data from the damage detection module or prediction analysis results from the prediction analysis module, the decision output module generates maintenance decisions and visualization results based on preset processing rules.

[0095] In the specific process, the decision output module first performs maintenance option evaluation on the received data: based on the current damage status and predicted development trend, it enumerates multiple optional maintenance schemes - including immediate shutdown for maintenance, planned maintenance, monitoring only without intervention, etc., and performs cost-benefit analysis on each scheme, that is, comprehensively evaluates the balance between the manpower cost, spare parts cost, downtime loss and the extended service life of the equipment after the implementation of the scheme, and quantifies the input-output ratio of each scheme.

[0096] After the assessment is completed, the decision output module applies a decision rule engine—that is, a pre-coded conditional logic rule library. This rule library contains decision logic such as "force shutdown if the damage level is severe", "arrange a weekly maintenance plan if the remaining lifespan is less than 30 days", and "include in the monthly inspection list if the risk level is low". Based on the input data, the corresponding rules are matched to select the maintenance plan that meets the constraints and has the best cost-effectiveness from the assessment plan.

[0097] Based on the selected maintenance plan, the decision output module generates a maintenance plan—clearly defining the specific maintenance execution time window, the required spare parts list, the required personnel configuration and work hour arrangement, etc., forming a structured maintenance task list.

[0098] Subsequently, the decision output module generates a maintenance decision report, integrating the detection data, predictive analysis results, maintenance option evaluation process, selected maintenance scheme, and detailed maintenance plan into an HTML-formatted detection and decision report. This report is presented in web page format and can be viewed in various terminal browsers, containing text descriptions and data tables.

[0099] Finally, the decision output module provides visualization output, creating 2D charts to intuitively display information such as wear trend curves, remaining life countdown, maintenance plan timeline, and fault location heatmap. These visualization charts, along with the HTML report, are pushed to the maintenance personnel's terminal or the warehouse management system monitoring interface for maintenance decision-making reference.

[0100] This embodiment generates accurate, intelligent, and efficient maintenance decisions through the decision output module, and forms complete visualized results, providing accurate and intuitive handling basis for track operation and maintenance work.

[0101] In one embodiment, such as Figure 7 As shown, steps S50 is followed by steps S51-S53.

[0102] S51. Feedback the maintenance decision to the preprocessing module to optimize the preprocessing strategy of the preprocessing module; S52. Feed the detection data back to the feature extraction module to optimize the feature extraction strategy of the feature processing module; S53. Feed back the prediction analysis results to the damage detection module to optimize the analysis method or threshold of the damage detection module.

[0103] In this embodiment, after the decision output module generates maintenance decisions and visualization results based on the detection data or predictive analysis results, the system further executes a feedback optimization process, feeding the output results back to the front-end module to adjust the processing strategy, forming a closed-loop self-optimizing intelligent detection mechanism.

[0104] First, the decision output module feeds back the generated maintenance decision to the preprocessing module as quality control feedback. That is, based on the accuracy and execution effect of this decision, the quality of the previous preprocessing is evaluated. If it is found that improper preprocessing leads to deviations in subsequent analysis, adjustment instructions are generated to optimize the preprocessing strategy of the preprocessing module. For example, the filtering parameters are adjusted to remove specific types of noise more effectively, or the coordinate registration threshold is modified to improve alignment accuracy.

[0105] Meanwhile, the decision output module feeds back the information on damage location, damage type, and damage extent contained in the detection data to the feature extraction module as feature optimization feedback. After receiving this feedback, the feature extraction module analyzes the key features of wear that were successfully identified and the features that were not effectively captured in this detection. Based on this, it adjusts the feature extraction strategy, such as increasing the extraction weight of certain sensitive features (e.g., edge sharpness, local curvature) or introducing new feature dimensions to improve the sensitivity and accuracy of subsequent wear detection.

[0106] In addition, the predictive analysis module feeds back its generated predictive analysis results—including wear development trend curves, future damage predictions, remaining service life estimates, and risk levels—to the damage detection module as model update feedback. After receiving this feedback, the damage detection module compares the prediction results with the current detection data and analyzes whether the classification threshold or analysis model used by the wear detection module is still suitable for the current wear evolution pattern. If a systematic deviation is found between the predicted trend and the historical model, the model update mechanism is triggered, and the analysis method parameters or damage severity assessment threshold of the damage detection module are automatically adjusted. For example, the severe wear depth threshold is adjusted from the original 3 mm to 2.8 mm to provide an earlier warning of accelerated wear evolution.

[0107] Through the above three-level feedback optimization, the system realizes closed-loop adaptive optimization of decision guidance front-end processing, detection feedback feature extraction, and prediction correction damage detection model, enabling each module to continuously iterate and evolve in terms of processing strategy, feature dimension, and analysis threshold, thereby continuously improving the detection accuracy and early warning timeliness of track damage.

[0108] This application achieves real-time acquisition of track point cloud data by integrating 3D radar at the bottom of the transportation equipment. The 3D radar collects data in real time as the transportation equipment moves, eliminating the need for other equipment and reducing detection costs. It also continuously monitors the real-time status of the track, improving the continuity of the detection and maintenance system. Furthermore, it obtains track feature data through preprocessing and feature extraction to improve the accuracy of basic data and the efficiency of subsequent analysis. Combining geometric matching algorithms and machine learning models, it accurately identifies defects such as track wear and loose screws, triggering graded early warnings and maintenance suggestions. This significantly improves the detection efficiency, continuity, stability, and accuracy of track damage detection, while also increasing maintenance efficiency, making maintenance more intelligent, and reducing maintenance costs. In addition, it provides an intelligent predictive maintenance solution for automated warehousing systems.

[0109] Figure 8 This is a schematic block diagram of a track detection and maintenance device 300 based on 3D vision provided in an embodiment of the present invention. Figure 8 As shown, corresponding to the above-described 3D vision-based track detection and maintenance method, the present invention also provides a 3D vision-based track detection and maintenance device 300. This 3D vision-based track detection and maintenance device 300 includes a single module for performing the above-described 3D vision-based track detection and maintenance method, and the device can be configured in a computer device. Specifically, please refer to... Figure 8 The 3D vision-based track detection and maintenance device 300 includes a data acquisition module 301, a preprocessing module 302, a feature extraction module 303, a damage detection module 304, and a decision output module 305.

[0110] The data acquisition module 301 is used to acquire point cloud data of the ground track in real time and input the point cloud data into the preprocessing module; Preprocessing module 302 is used to preprocess the point cloud data to obtain track data, and input the track data into the feature extraction module; Feature extraction module 303 is used to extract feature data from the track data and input the feature data into the damage detection module; The damage detection module 304 is used to analyze the feature data to generate detection data and input the detection data to the decision output module; The decision output module 305 is used to generate maintenance decisions and visualization results based on the detection data and preset processing rules.

[0111] In one embodiment, the damage detection module 304 includes a positioning module, a classification module, an identification module, and an output module.

[0112] The positioning module is used to locate the damaged location from the feature data based on a preset positioning algorithm; The classification module is used to classify the damaged location according to the feature data based on the preset classification rules, so as to generate the corresponding damage type; The identification module is used to select a corresponding preset analysis method to identify the degree of damage based on the damage type; The output module is used to input the damage location, the damage type, and the damage extent into the decision output module.

[0113] In one embodiment, the identification module includes a first identification module and a second identification module.

[0114] The first identification module is used to calculate the track surface deviation value based on the feature data and preset reference data according to the geometric matching algorithm if the damage type is track wear, and then identify the degree of damage according to the track surface deviation value and preset threshold. The second identification module is used to identify the bolt abnormality type based on the feature data according to a preset machine learning model if the damage type is a bolt abnormality type, and then match the corresponding damage degree based on the abnormality type.

[0115] In one embodiment, the 3D vision-based track detection and maintenance device 300 further includes a predictive analysis module and an optimization module.

[0116] A predictive analysis module is used to determine whether the degree of damage is severe. If it is not severe, the detection data is input to the predictive analysis module, which generates a predictive analysis result based on the detection data and historical fault data. The predictive analysis result is then input to the decision output module, so that the decision output module generates maintenance decisions and visualization results based on the predictive analysis result and preset processing rules. If it is severe, the detection data is input to the decision output module, so that the decision output module generates maintenance decisions and visualization results based on the detection data and preset processing rules. The optimization module is used to feed back the maintenance decision to the preprocessing module to optimize the preprocessing strategy of the preprocessing module; feed back the detection data to the feature extraction module to optimize the feature extraction strategy of the feature processing module; and feed back the predictive analysis results to the damage detection module to optimize the analysis method or threshold of the damage detection module.

[0117] In one embodiment, the preprocessing module 302 includes a conversion module, a simplification module, and an extraction module.

[0118] The conversion module is used to perform point cloud format conversion and coordinate system unification processing on the point cloud data to generate the first data. A simplification module is used to perform noise filtering and data simplification on the first data to generate the second data; An extraction module is used to perform track segmentation and coordinate alignment processing on the second data to generate the track data.

[0119] In one embodiment, the feature extraction module 303 includes a slicing module, a fitting module, and a calculation module.

[0120] The slicing module is used to slice along the extension direction of the track data to obtain track cross-sectional data; The fitting module is used to perform contour fitting on the track cross-section data and construct an actual track model. The calculation module is used to calculate the deviation between the actual orbit model and the standard orbit model, and to calculate the feature data based on the deviation.

[0121] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned 3D vision-based track detection and maintenance device and its modules can be referred to the corresponding descriptions in the foregoing method embodiments. For the sake of convenience and brevity, these will not be repeated here.

[0122] The aforementioned 3D vision-based track detection and maintenance device 300 can be implemented as a computer program, which can, for example... Figure 9 It runs on the computer device shown.

[0123] Please see Figure 9 , Figure 9 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a terminal or a server. The terminal can be an electronic device with communication functions, such as a smartphone, tablet, laptop, desktop computer, personal digital assistant, or wearable device. The server can be a standalone server or a server cluster composed of multiple servers.

[0124] See Figure 9 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0125] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a 3D vision-based track detection and maintenance method.

[0126] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0127] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a 3D vision-based track detection and maintenance method.

[0128] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0129] The processor 502 is used to run the computer program 5032 stored in the memory to implement the steps of the above-mentioned 3D vision-based track detection and maintenance method.

[0130] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0131] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0132] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein the computer program includes program instructions. When executed by a processor, the program instructions cause the processor to perform the steps of the above-described 3D vision-based track detection and maintenance method.

[0133] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0134] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0135] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0136] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0137] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0138] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for track detection and maintenance based on 3D vision, characterized in that, The method, applied to a detection and maintenance system including a data acquisition module, a preprocessing module, a feature extraction module, a damage detection module, and a decision output module, comprises: The data acquisition module acquires point cloud data of the ground track in real time and inputs the point cloud data into the preprocessing module; The preprocessing module preprocesses the point cloud data to obtain track data, and then inputs the track data into the feature extraction module; The feature extraction module extracts feature data from the track data and inputs the feature data into the damage detection module; The damage detection module analyzes the feature data to generate detection data, and inputs the detection data into the decision output module; The decision output module generates maintenance decisions and visualization results based on the detection data and preset processing rules.

2. The method according to claim 1, characterized in that, The detection data includes the location of damage, the type of damage, and the extent of damage. The step of the damage detection module analyzing the feature data to generate detection data includes: The damaged location is located from the feature data based on a preset positioning algorithm; Based on preset classification rules, the damaged location is classified according to the feature data to generate the corresponding damage type; Based on the damage type, select the corresponding preset analysis method to identify the degree of damage; The location of the damage, the type of damage, and the degree of damage are input into the decision output module.

3. The method according to claim 2, characterized in that, The damage types include track wear and bolt abnormalities. The step of selecting a corresponding preset analysis method to identify the degree of damage based on the damage type includes: If the damage type is track wear, the track surface deviation value is calculated based on the feature data and preset reference data using a geometric matching algorithm, and then the degree of damage is identified based on the track surface deviation value and a preset threshold. If the damage type is a bolt anomaly, the bolt anomaly type is identified based on the feature data using a preset machine learning model, and then the corresponding damage level is matched based on the anomaly type.

4. The method according to claim 2, characterized in that, The detection and maintenance system further includes a predictive analysis module, and after the step of selecting a corresponding preset analysis method to identify the degree of damage based on the damage type, it includes: Determine whether the degree of damage is considered severe; If the severity is not high, the detection data is input to the predictive analysis module. The predictive analysis module generates predictive analysis results based on the detection data and historical fault data. The predictive analysis results are then input to the decision output module so that the decision output module generates maintenance decisions and visualization results based on the predictive analysis results and preset processing rules. If the severity level is specified, the detection data is input to the decision output module, so that the decision output module generates maintenance decisions and visualization results based on the detection data and preset processing rules.

5. The method according to claim 4, characterized in that, After the step of generating maintenance decisions and visualization results based on the detection data and preset processing rules, the decision output module further includes: The maintenance decision is fed back to the preprocessing module to optimize the preprocessing strategy of the preprocessing module; The detection data is fed back to the feature extraction module to optimize the feature extraction strategy of the feature processing module; The predictive analysis results are fed back to the damage detection module to optimize the analysis method or threshold of the damage detection module.

6. The method according to claim 1, characterized in that, The preprocessing module preprocesses the point cloud data to obtain track data, including the following steps: The point cloud data is converted to a point cloud format and processed using a coordinate system to generate the first data. The first data is subjected to noise filtering and data simplification to generate the second data; The second data is subjected to track segmentation and coordinate alignment to generate the track data.

7. The method according to claim 1, characterized in that, The steps for the feature extraction module to extract feature data from the orbital data include: Slice along the extension direction of the track data to obtain track cross-sectional data; The track cross-section data is used to perform contour fitting and construct an actual track model; The deviation between the actual orbit model and the standard orbit model is calculated, and the feature data is calculated based on the deviation.

8. A track detection and maintenance device based on 3D vision, characterized in that, Includes a module for performing the method as described in any one of claims 1-7.

9. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A storage medium, characterized in that, The storage medium stores a computer program, which includes program instructions that, when executed by a processor, can implement the method as described in any one of claims 1-7.

Images