Straddle-type monorail state monitoring method and system

Abstract

The application discloses a straddle-type monorail state monitoring method and system, relates to the technical field of track detection, and has the following process: 3D point cloud data of a contact rail surface is acquired first, and effective data is obtained through preprocessing; the RT matrix dynamic compensation mapping to the same coordinate system is carried out through binocular line laser calibration, then a calculation graph reflecting the distribution characteristic matrix of the 3D point cloud data of the contact rail is generated through regional sampling and noise processing; specific targets and characteristic values of the contact rail and various kinds of accessory workpieces are calibrated in the calculation graph, and the MaxPool global characteristic parameter matrix is constructed through multi-thread filtering and data continuity discrimination fitting; the parallel calculation matrix is constructed based on the matrix, and the state parameters are solved in full resolution and in parallel after the high importance area is segmented; finally, the state evaluation result is output by fusing the related data in multiple modes, and the alarm is triggered when the state is abnormal. The application realizes comprehensive and accurate monitoring and abnormal early warning of the state of the contact rail and various kinds of accessory workpieces, and provides reliable support for safe and stable operation of the straddle-type monorail system.

Description

Technical Field

[0001] This invention relates to the field of track inspection technology, and in particular to a straddle-type monorail condition monitoring method and system. Background Technology

[0002] Straddle-type monorail transit, as a flexible, efficient, and highly adaptable form of rail transportation, is widely used in urban commuting and other scenarios due to its advantages such as small footprint, strong climbing ability, and small turning radius. The contact rail, as the core power supply component of the straddle-type monorail train, directly affects the stability of the train's power supply and driving safety. Meanwhile, the structural integrity and assembly precision of auxiliary components such as expansion joints, insulators, intermediate joints, and segmented insulators are also crucial to the long-term safe operation of the monorail system.

[0003] Currently, state parameter detection algorithms for suspended monorails have emerged in the rail transit field. However, when adapting to the detection needs of straddle-type monorails, many technical limitations have gradually become apparent. Straddle-type monorails and suspended monorails differ significantly in track structure layout, contact rail installation methods, and engineering vehicle operating conditions. Straddle-type monorails often use side or under-mounted contact rails, and their auxiliary components are numerous and densely installed. Furthermore, the maximum operating speed of engineering vehicles can reach 75 km / h, placing higher demands on the response speed and detection resolution of the detection algorithms, making it difficult for existing suspended monorail detection algorithms to adapt.

[0004] Existing state parameter detection algorithms for suspended monorails have the following shortcomings when adapted to straddle-type monorail scenarios: limitations in inter-frame data processing; insufficient anti-interference capabilities in data acquisition and preprocessing; difficulty in balancing efficiency and accuracy in parameter calculation; insufficient multi-source data support for state assessment; and inadequate specific target detection and data continuity processing. Therefore, a technical solution for straddle-type monorail state monitoring is urgently needed to address these shortcomings and ensure the safe and stable operation of straddle-type monorail systems. Summary of the Invention

[0005] In view of this, this application provides a straddle-type monorail condition monitoring method and system to address the shortcomings of the existing technology.

[0006] The first aspect of this application provides a method for monitoring the condition of a straddle-type monorail, comprising:

[0007] Acquire 3D point cloud data of the contact rail surface;

[0008] The 3D point cloud data is preprocessed to obtain valid data;

[0009] The RT matrix is ​​obtained based on binocular laser calibration. The effective data is mapped to the same coordinate system through dynamic compensation. After the data distribution feature matrix is ​​extracted by regional sampling, noise is processed according to the set process to obtain a calculation map reflecting the distribution feature matrix of the 3D point cloud data of the contact rail.

[0010] In the computational graph, all specific targets and their feature values ​​are marked. After multi-threaded parallel filtering, the data continuity is discriminated and fitted to construct the MaxPool global feature parameter matrix. All specific targets include contact rail, expansion joint, intermediate joint, turnout section insulator, main line section insulator, center anchor, cable connection plate and insulator.

[0011] Based on the MaxPool global feature parameter matrix, a parallel parameter calculation matrix is ​​constructed. After the high-importance region is segmented according to the data distribution topology, the contact rail state parameters are solved in parallel using a full-resolution matrix calculation method.

[0012] A multimodal fusion method is adopted to integrate the contact rail status parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data, outputting status assessment results and triggering alarms when abnormalities occur.

[0013] In one possible implementation of the first aspect, obtaining the RT matrix based on binocular laser calibration and mapping the effective data to the same coordinate system through dynamic compensation includes:

[0014] Based on binocular laser calibration technology, the RT matrix describing the relative attitude and position relationship between the sensor coordinate system and the contact rail reference coordinate system is obtained;

[0015] The system receives dynamic compensation parameters in a unified coordinate system provided by the pre-set contact rail parameter compensation module, and maps the effective data to the same coordinate system based on the RT matrix.

[0016] In one possible implementation of the first aspect, the data distribution feature matrix extracted through regional sampling includes:

[0017] Based on the single-frame 3D point cloud data mapped to the same coordinate system, the contact rail surface is sampled by a differential network and texture extraction operator to extract a data distribution feature matrix that characterizes the surface feature distribution of the contact rail.

[0018] In one possible implementation of the first aspect, preprocessing the 3D point cloud data to obtain effective data includes:

[0019] Based on prior data, invalid values ​​are set to zero for the surface depth map and surface grayscale map corresponding to the 3D point cloud data, respectively.

[0020] The surface depth map and surface grayscale map, after invalid values ​​have been set to zero, are filtered for valid values ​​using a logical AND operation, as follows:

[0021]

[0022] For valid data, This is a surface depth map. This is a grayscale image of the surface.

[0023] In one possible implementation of the first aspect, processing noise according to a set process includes: sequentially processing large-scale outliers and local noise.

[0024] In one possible implementation of the first aspect, all specific targets and their feature values ​​are identified in the computational graph, and all feature values ​​are filtered in parallel by multiple threads, including:

[0025] Based on the computational graph reflecting the distribution characteristics of the 3D point cloud data of the contact rail, the spatial topology, surface texture distribution and data gradient change characteristics carried by the computational graph are analyzed and evaluated to identify all specific targets and their corresponding feature values.

[0026] Based on the aforementioned feature values, a multi-threaded parallel architecture is used to filter and label the preprocessed and feature-extracted multi-frame data matrix, select the effective data that matches the feature values ​​corresponding to various specific targets, and accurately store all specific targets and their corresponding feature values ​​in the matrix pointed to by the target category pointer, denoted as the target matrix.

[0027] In one possible implementation of the first aspect, discriminative fitting of data continuity to construct the MaxPool global feature parameter matrix includes:

[0028] The target matrix is ​​evaluated and discrete invalid data is filtered out using data continuity.

[0029] After data continuity discrimination and filtering, the effective data corresponding to all specific targets are obtained and line segment fitting and curve segment fitting are performed respectively to obtain the corresponding fitting parameters.

[0030] Based on all fitting parameters and feature information of various specific targets, the MaxPool global feature parameter matrix is ​​constructed.

[0031] In one possible implementation of the first aspect, the high-importance area includes the contact rail surface and side, intermediate joint weld, and expansion joint.

[0032] In one possible implementation of the first aspect, the fusion of the contact rail status parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data, to output a status assessment result and trigger an alarm in case of an anomaly, includes:

[0033] By integrating parallel solutions, contact rail state parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data are obtained. Through multi-dimensional feature cross-validation and comprehensive analysis, state assessment results are generated.

[0034] The status assessment results cover three levels: normal operation status, Level 1 abnormal status, and Level 2 abnormal status.

[0035] When the status assessment result indicates normal operation, no action is taken;

[0036] When the status assessment result is a Level 1 abnormal state, an early warning message is output and the abnormal data is recorded and archived.

[0037] When the status assessment result is a level 2 abnormal state, the hierarchical alarm mechanism is automatically triggered, and the abnormal location, abnormal type and characteristic parameters are synchronously fed back to the monitoring terminal.

[0038] A second aspect of this application provides a straddle-type monorail condition monitoring system, comprising:

[0039] The data acquisition module is used to acquire 3D point cloud data of the contact rail surface;

[0040] The preprocessing module is used to preprocess the 3D point cloud data to obtain effective data;

[0041] The feature extraction module is used to obtain the RT matrix based on binocular laser calibration, map the effective data to the same coordinate system through dynamic compensation, extract the data distribution feature matrix through regional sampling, process noise according to the set process, and obtain a calculation map reflecting the distribution feature matrix of the 3D point cloud data of the contact rail.

[0042] The matrix construction module is used to mark all specific targets and their feature values ​​in the calculation graph. All feature values ​​are filtered in parallel by multiple threads and the data continuity is discriminated and fitted to construct the MaxPool global feature parameter matrix. All specific targets include contact rail, expansion joint, intermediate joint, turnout section insulator, main line section insulator, center anchor, cable connection plate and insulator.

[0043] The state parameter calculation module is used to construct a parallel parameter calculation matrix based on the MaxPool global feature parameter matrix. After dividing the high-importance region according to the data distribution topology, it uses a full-resolution matrix calculation method to solve the contact rail state parameters in parallel.

[0044] The multimodal monitoring module is used to fuse the contact rail status parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data using a multimodal fusion method, output status assessment results, and trigger alarms when abnormalities occur.

[0045] Its beneficial effects are as follows: This invention discloses a straddle-type monorail condition monitoring method and system. The process is as follows: First, acquire 3D point cloud data of the contact rail surface and preprocess it to obtain effective data; dynamically compensate and map the RT matrix calibrated by binocular laser to the same coordinate system, and then generate a calculation map reflecting the distribution characteristics of the 3D point cloud data of the contact rail through regional sampling and noise processing; mark specific targets and feature values ​​such as the contact rail and various auxiliary workpieces in the calculation map, and construct the MaxPool global feature parameter matrix through multi-threaded filtering and data continuity discrimination fitting; construct a parallel computing matrix based on this matrix, segment the high-importance region, and solve the state parameters in parallel at full resolution; finally, fuse relevant data in multiple modes, output the state assessment result, and trigger an alarm when there is an anomaly. This invention effectively solves the problems of insufficient anti-interference capability, difficulty in balancing efficiency and accuracy, lack of multi-source data support, and imperfect target detection and data continuity processing when existing algorithms are adapted to straddle-type monorails. It is suitable for high-speed operation of engineering vehicles at 75km / h and scenarios with dense layout of auxiliary workpieces. It can also ensure the detection accuracy of minor defects through full-resolution calculation and multi-modal fusion, realize comprehensive and accurate monitoring and anomaly early warning of the contact rail and various auxiliary workpieces, and provide reliable support for the safe and stable operation of straddle-type monorail systems. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0047] Figure 1 This is a schematic flowchart of a straddle-type monorail condition monitoring method provided in an embodiment of this application;

[0048] Figure 2 This is a schematic diagram of multimodal data fusion monitoring provided in an embodiment of this application;

[0049] Figure 3 This is a schematic diagram of the composition of a straddle-type monorail condition monitoring system provided in an embodiment of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0052] Example 1

[0053] Existing state parameter detection algorithms for suspended monorails have the following shortcomings when adapted to straddle-type monorail scenarios:

[0054] The inter-frame data processing mode has limitations. The original algorithm processes inter-frame data independently without establishing the correlation between them. This results in a limited visual inspection window, which cannot fully cover the continuous layout of the straddle-type contact rail and its auxiliary workpieces, and is prone to missing defects across frames. At the same time, the number of calculation points per operation is not dynamically adapted according to the trigger frequency and the size of the inspection window. The computational scale is insufficient and cannot meet the resolution requirements of the smallest workpiece under the 75km / h operating condition. It is also unable to accurately capture small defects such as scratches and grooves on the contact rail surface.

[0055] The data acquisition and preprocessing are not strong enough to resist interference. The contact rail data acquisition process is susceptible to multiple interferences, such as sensor noise, data jitter during the operation of the engineering vehicle, reflection on the metal surface of the contact rail, and defects such as surface scratches, grooves, and dirt. All of these can lead to invalid data and noise in the acquired surface depth map and surface grayscale map. The existing algorithm does not have layered processing logic designed for such random and sequentially sensitive noise, which seriously affects the accuracy of subsequent parameter calculations.

[0056] Efficiency and accuracy in parameter calculation are difficult to balance. Straddle-type monorail monitoring needs to simultaneously meet the requirements of rapid response under high-speed operation and high accuracy in defect detection, but existing algorithms lack an efficient parallel computing architecture. If serial calculation or low-resolution data processing is used to improve efficiency, it will result in insufficient feature extraction accuracy for high-importance areas such as the contact rail surface, intermediate joint welds, and expansion joints. If the amount of computation is simply increased to ensure accuracy, it will cause computational delays and cannot meet the real-time detection requirements of high-speed engineering vehicles.

[0057] The condition assessment lacks sufficient multi-source data support. Existing algorithms mostly rely on single-type data for condition assessment, without integrating multi-modal information such as data acquisition timestamps, high-definition imaging data of contact rails and auxiliary workpieces, and surface defect detection results. This results in a lack of comprehensiveness and relevance in condition assessment, which easily leads to misjudgment and omission.

[0058] The detection of specific targets and the processing of data continuity are not perfect. Straddle-type monorails need to accurately identify a variety of specific targets such as contact rails, expansion joints, and insulators. However, existing algorithms do not have a refined marking and filtering mechanism designed for the feature differences of multiple targets, and lack the discrimination and fitting processing of data continuity. They cannot effectively identify small deformations that are continuous across frames, resulting in low target recognition accuracy and insufficient completeness of state parameter calculation.

[0059] Therefore, this application provides a straddle-type monorail condition monitoring method, such as... Figure 1 As shown, it includes:

[0060] Acquire 3D point cloud data of the contact rail surface;

[0061] The 3D point cloud data is preprocessed to obtain valid data;

[0062] The RT matrix is ​​obtained based on binocular laser calibration. The effective data is mapped to the same coordinate system through dynamic compensation. After the data distribution feature matrix is ​​extracted by regional sampling, noise is processed according to the set process to obtain a calculation map reflecting the distribution feature matrix of the 3D point cloud data of the contact rail.

[0063] In the computational graph, all specific targets and their feature values ​​are marked. After multi-threaded parallel filtering, the data continuity is discriminated and fitted to construct the MaxPool global feature parameter matrix. All specific targets include contact rail, expansion joint, intermediate joint, turnout section insulator, main line section insulator, center anchor, cable connection plate and insulator.

[0064] Based on the MaxPool global feature parameter matrix, a parallel parameter calculation matrix is ​​constructed. After the high-importance region is segmented according to the data distribution topology, the contact rail state parameters are solved in parallel using a full-resolution matrix calculation method.

[0065] A multimodal fusion method is adopted to integrate the contact rail status parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data, outputting status assessment results and triggering alarms when abnormalities occur.

[0066] This invention discloses a straddle-type monorail condition monitoring method, comprising a complete process of RGBD data acquisition (point cloud data), point cloud preprocessing, contact rail feature extraction, feature matrix construction, condition parameter calculation, and condition assessment and early warning. The working principle of this embodiment will be explained in detail below:

[0067] RGBD data acquisition:

[0068] With the goal of comprehensive coverage, strong synchronization, and adaptability to high-speed operating conditions, it provides high-quality multimodal raw data for subsequent processing, solving the problems of incomplete data acquisition coverage and inability to capture minute defects in existing technologies.

[0069] Using an RGBD sensor as the core of data acquisition, and leveraging its ability to simultaneously acquire three-dimensional spatial coordinates and two-dimensional image information, the system captures in real time 3D point cloud data (including the spatial position and shape contour information of the contact rail and its associated workpieces), surface depth map, and surface grayscale map of the contact rail surface.

[0070] The sensor acquisition frequency is dynamically matched with the engineering vehicle's high-speed operation at 75km / h, ensuring that during the train's rapid movement, the continuously acquired multi-frame data can fully cover the contact rail and densely arranged auxiliary workpieces such as expansion joints and insulators, avoiding the omission of cross-frame features due to insufficient acquisition frequency; at the same time, the acquisition resolution is set according to the smallest workpiece size to ensure the original data of minor defects such as scratches and grooves on the contact rail surface can be captured.

[0071] Point cloud preprocessing:

[0072] This method specifically suppresses multiple interferences and filters out pure and valid data, solving the problems of insufficient anti-interference capability of existing technologies and invalid data affecting subsequent calculations.

[0073] Invalid value zeroing principle: Based on prior data (sensor or hardware communication protocol / user manual), invalid data (including sensor blind zone data, distorted data caused by extreme reflection, and meaningless data caused by dirt and occlusion) that exceed the reasonable range in the surface depth map and surface grayscale map are marked as zero, clearly distinguishing between valid data and invalid data, and avoiding invalid data from participating in subsequent calculations.

[0074] Valid value selection principle: A dual verification mechanism is established through logical AND operations. Data at a spatial location is only considered valid if both depth and grayscale data are valid. The principle leverages the complementarity of depth and grayscale data. Depth data reflects spatial morphology, while grayscale data reflects surface characteristics. The simultaneous validity of both significantly eliminates interference from a single data dimension, ultimately achieving precise purification of valid data.

[0075] Contact rail feature extraction:

[0076] To address the issues of spatial inconsistency in multi-source data, fuzzy feature extraction, and severe noise interference, the purified data is transformed into structured features, providing a reliable basis for target calibration.

[0077] Coordinate system one, based on binocular laser calibration technology, establishes a mapping relationship between the sensor coordinate system and the contact rail reference coordinate system through a calibration board, and calculates the RT matrix. The core function of the RT matrix is ​​to quantify the attitude difference and position difference between the sensor acquisition viewpoint and the actual installation viewpoint of the contact rail, so as to achieve spatial alignment of data under different acquisition viewpoints. It receives dynamic compensation parameters provided by the contact rail parameter compensation module, and combines them with the RT matrix to uniformly map all effective data to the contact rail reference coordinate system, ensuring that data acquired at different times and locations have spatial uniformity and avoiding feature misalignment caused by coordinate system confusion.

[0078] For region sampling and feature extraction, a feature extraction strategy of differentiation and multi-operator collaboration is adopted for single-frame 3D point cloud data after unifying the coordinate system: the sampling area is divided according to the spatial distribution of the contact rail and its auxiliary workpieces (such as the contact rail body area, joint connection area, and insulator installation area) to avoid cross-interference of features of different components; the edge gradient features within the area are captured by the differential network (such as the morphological abrupt change at the joint weld and the contour transition of the contact rail side), and the surface texture features are extracted by the texture extraction operator. These multi-dimensional feature quantities are converted into numerical values ​​and integrated to form a data distribution feature matrix that characterizes the surface feature distribution of the contact rail.

[0079] Layered noise reduction: Layered processing logic is designed according to noise scale and characteristics to avoid cross-influence of noise on feature accuracy: First, large-scale discrete points are processed by statistical filtering algorithms to remove discrete points caused by vehicle vibration and sudden sensor failures (such noise is spatially dispersed, large in scale, and has a significant impact on overall features); then, local noise is processed by local smoothing algorithms to suppress minor noise caused by scratches and dirt on the contact rail surface (such noise is concentrated and small in scale, and needs to be moderately suppressed while preserving core features), finally obtaining a computational graph with high signal-to-noise ratio and high discriminability (essentially a visualization and structured carrier of the data distribution feature matrix).

[0080] Feature matrix construction:

[0081] To address the issues of low accuracy in specific target recognition, isolated data between frames, and fragmented features in existing technologies, a structured matrix that comprehensively covers global features is constructed.

[0082] The multi-target calibration principle is based on three core features carried by the computational graph (spatial topology, surface texture distribution, and data gradient changes). It utilizes the inherent feature differences of different specific targets for accurate calibration: spatial topological differences, contact rails exhibit a linear topology with interconnected lines, insulators exhibit an independent columnar topology, and intermediate joints exhibit a locally dense clustered topology, which initially distinguishes target types through topological morphology; texture and gradient differences, contact rail metal surfaces have high reflectivity and gentle gradient changes, while weld seams show obvious gradient abrupt changes, and insulator surfaces have weak reflectivity and uniform texture, which refine target identification through these features; finally, eight specific targets (contact rails, expansion joints, turnout segment insulators, mainline segment insulators, center anchors, cable connection plates, and insulators) and their corresponding feature values ​​are calibrated.

[0083] The principle of multi-threaded parallel filtering is based on the calibrated target feature values. It uses a multi-threaded parallel architecture to process multiple frames of data matrices in batches: each thread corresponds to a specific target and synchronously filters valid data that matches the target feature value, avoiding the efficiency bottleneck caused by serial processing; the filtered target-feature value corresponding data is accurately stored in the matrix pointed to by the target category pointer (target matrix), and the pointer index enables fast retrieval of different target data, laying the foundation for structured storage for subsequent processing.

[0084] The principle of data continuity discrimination and fitting utilizes the temporal correlation of inter-frame data to determine the continuity of the target matrix. If the point cloud data of a certain target is continuously distributed across multiple frames and meets the spatial connection requirements, it is determined to be valid continuous data; if the data is discretely distributed and has no temporal correlation, it is determined to be invalid noise data and is discarded, thus solving the problem of missed detection caused by isolated inter-frame data. Targeted fitting is performed for the morphological characteristics of different targets. Linear targets such as contact rails and mainline segment insulators are fitted with straight lines, while nonlinear targets such as welds and expansion joints are fitted with curve segments. Through fitting operations, discrete point cloud data is transformed into continuous geometric parameters, achieving structured feature extraction.

[0085] By integrating the fitting parameters (geometric parameters), original feature values ​​(texture, gradient parameters), and category labeling information of all targets, a MaxPool global feature parameter matrix is ​​constructed. The core function of the matrix is ​​global feature normalization and integration. Through MaxPool operations, key features of various targets (such as maximum wear, maximum offset, and core morphological parameters) are retained, while redundant data is compressed to form a global feature geometry that can comprehensively characterize the state of the contact rail and all auxiliary workpieces, providing a unified data foundation for subsequent parameter calculations.

[0086] State parameter calculation:

[0087] Breaking through the bottleneck of existing technologies that make it difficult to balance efficiency and accuracy, this technology adapts to high-speed operating conditions while ensuring the accuracy of parameter calculations in highly important areas.

[0088] The principle of constructing the parallel parameter calculation matrix is ​​based on the MaxPool global feature parameter matrix. The calculation tasks are split according to the target type and regional importance to construct the parallel parameter calculation matrix. Each row of the matrix corresponds to an independent calculation task (such as contact rail wear calculation, intermediate joint weld gap calculation, and insulator verticality calculation), and each column corresponds to the feature parameters required by the task. The matrix structure realizes clear splitting and parallel scheduling of calculation tasks.

[0089] The principle of high-importance region segmentation is based on the data distribution topology, which accurately segments high-importance regions such as the contact rail surface and side rails, intermediate joint welds, and expansion joints. These regions are directly related to the safety of monorail operation. The core principle of segmentation is task decoupling. Each segmented region corresponds to an independent computing task. There is no data overlap and no computational interference between regions, providing independently executable task units for parallel computing.

[0090] The principle of full-resolution matrix calculation and parallel solution employs a full-resolution calculation method, preserving the original data accuracy of 3200 points per frame without downsampling, ensuring that key features such as minor wear on the contact rail surface and minor gaps in the weld are not lost, thus guaranteeing the accuracy of parameter calculation. Relying on a multi-threaded parallel architecture, the state parameters of each segmented region are solved simultaneously: the wear amount and dynamic offset of the contact rail, the weld gap size of the intermediate joint, the assembly deviation of the expansion joint, and the verticality of the insulator. The synergistic principle of parallel architecture and full-resolution calculation avoids a single task consuming a large amount of computing power through decoupling, and improves the overall computing efficiency through parallel scheduling, ultimately achieving the dual goals of real-time response under high-speed conditions of 75km / h and accurate calculation with minimal defect resolution.

[0091] Status assessment and early warning:

[0092] This addresses the problem of insufficient single-source data support in existing technologies, which leads to missed or incorrect judgments, and enables comprehensive assessment and accurate early warning of the status of contact rails and auxiliary workpieces.

[0093] Multimodal data fusion principle (e.g.) Figure 2As shown, the system integrates six key data categories and enhances assessment reliability through multi-dimensional feature cross-validation: contact rail state parameters obtained through parallel solving (core functional parameters); synchronously acquired timestamp information (spatiotemporal location of associated data, facilitating anomaly tracing); contact rail surface parameters (surface roughness, wear distribution, reflecting surface condition); high-definition imaging data of the contact rail and its associated workpieces (visible light images and 3D point cloud data, providing intuitive visual evidence); and surface defect data (defect location, size, shape, and grayscale gradient features, reflecting local damage status). The core of the fusion mechanism is complementary data verification. For example, when state parameters indicate wear is close to a threshold, high-definition imaging data is used to confirm the visual characteristics of the wear area, and defect data is used to determine whether wear damage exists, avoiding misjudgments caused by a single data dimension.

[0094] The status assessment and early warning grading principle is based on industry standards and pre-set normal parameter thresholds according to operation and maintenance requirements. It generates three levels of status assessment results through multi-dimensional feature cross-validation: normal operation (all parameters are within the threshold range), Level 1 abnormal state (some parameters are close to the threshold but there is no safety risk), and Level 2 abnormal state (critical parameters exceed the threshold or there is significant damage). Differentiated early warning logic: no action is taken in the normal state to ensure no redundant system response; Level 1 abnormal state outputs early warning prompts and records abnormal data and corresponding timestamps, providing a basis for preventative operation and maintenance; Level 2 abnormal state automatically triggers a tiered alarm mechanism (such as audible and visual alarms and remote push notifications), synchronously feeding back the abnormal location (based on timestamp and spatial coordinate association), abnormal type (such as wear amount, weld cracking), and characteristic parameters to the monitoring terminal, ensuring rapid response from operation and maintenance personnel and minimizing safety risks.

[0095] In some embodiments, obtaining the RT matrix based on binocular laser calibration and mapping the effective data to the same coordinate system through dynamic compensation includes:

[0096] Based on binocular laser calibration technology, the RT matrix describing the relative attitude and position relationship between the sensor coordinate system and the contact rail reference coordinate system is obtained;

[0097] The system receives dynamic compensation parameters in a unified coordinate system provided by the pre-set contact rail parameter compensation module, and maps the effective data to the same coordinate system based on the RT matrix.

[0098] In some embodiments, the data distribution feature matrix extracted through regional sampling includes:

[0099] Based on the single-frame 3D point cloud data mapped to the same coordinate system, the contact rail surface is sampled by a differential network and texture extraction operator to extract a data distribution feature matrix that characterizes the surface feature distribution of the contact rail.

[0100] In some embodiments, preprocessing the 3D point cloud data to obtain valid data includes:

[0101] Based on prior data, invalid values ​​are set to zero for the surface depth map and surface grayscale map corresponding to the 3D point cloud data, respectively.

[0102] The surface depth map and surface grayscale map, after invalid values ​​have been set to zero, are filtered for valid values ​​using a logical AND operation, as follows:

[0103]

[0104] For valid data, This is a surface depth map. This is a grayscale image of the surface.

[0105] In some embodiments, processing noise according to a set process includes: sequentially processing large-scale outliers and local noise.

[0106] In some embodiments, all specific targets and their feature values ​​are identified in the computational graph, and all feature values ​​are filtered in parallel using a multi-threaded process, including:

[0107] Based on the computational graph reflecting the distribution characteristics of the 3D point cloud data of the contact rail, the spatial topology, surface texture distribution and data gradient change characteristics carried by the computational graph are analyzed and evaluated to identify all specific targets and their corresponding feature values.

[0108] Based on the aforementioned feature values, a multi-threaded parallel architecture is used to filter and label the preprocessed and feature-extracted multi-frame data matrix, select the effective data that matches the feature values ​​corresponding to various specific targets, and accurately store all specific targets and their corresponding feature values ​​in the matrix pointed to by the target category pointer, denoted as the target matrix.

[0109] In some embodiments, discriminative fitting of data continuity to construct the MaxPool global feature parameter matrix includes:

[0110] The target matrix is ​​evaluated and discrete invalid data is filtered out using data continuity.

[0111] After data continuity discrimination and filtering, the effective data corresponding to all specific targets are obtained and line segment fitting and curve segment fitting are performed respectively to obtain the corresponding fitting parameters.

[0112] Based on all fitting parameters and feature information of various specific targets, the MaxPool global feature parameter matrix is ​​constructed.

[0113] In some embodiments, the high-importance area includes the contact rail surface and sides, intermediate joint welds, and expansion joints.

[0114] In some embodiments, fusing the contact rail status parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data to output a status assessment result and trigger an alarm when an anomaly occurs includes:

[0115] By integrating parallel solutions, contact rail state parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data are obtained. Through multi-dimensional feature cross-validation and comprehensive analysis, state assessment results are generated.

[0116] The status assessment results cover three levels: normal operation status, Level 1 abnormal status, and Level 2 abnormal status.

[0117] When the status assessment result indicates normal operation, no action is taken;

[0118] When the status assessment result is a Level 1 abnormal state, an early warning message is output and the abnormal data is recorded and archived.

[0119] When the status assessment result is a level 2 abnormal state, the hierarchical alarm mechanism is automatically triggered, and the abnormal location, abnormal type and characteristic parameters are synchronously fed back to the monitoring terminal.

[0120] Example 2

[0121] Based on the straddle-type monorail condition monitoring method provided in Embodiment 1 of this application, correspondingly, Embodiment 2 of this application also provides a straddle-type monorail condition monitoring system, such as... Figure 3 As shown, it includes:

[0122] The data acquisition module is used to acquire 3D point cloud data of the contact rail surface;

[0123] The preprocessing module is used to preprocess the 3D point cloud data to obtain effective data;

[0124] The feature extraction module is used to obtain the RT matrix based on binocular laser calibration, map the effective data to the same coordinate system through dynamic compensation, extract the data distribution feature matrix through regional sampling, process noise according to the set process, and obtain a calculation map reflecting the distribution feature matrix of the 3D point cloud data of the contact rail.

[0125] The matrix construction module is used to mark all specific targets and their feature values ​​in the calculation graph. All feature values ​​are filtered in parallel by multiple threads and the data continuity is discriminated and fitted to construct the MaxPool global feature parameter matrix. All specific targets include contact rail, expansion joint, intermediate joint, turnout section insulator, main line section insulator, center anchor, cable connection plate and insulator.

[0126] The state parameter calculation module is used to construct a parallel parameter calculation matrix based on the MaxPool global feature parameter matrix. After dividing the high-importance region according to the data distribution topology, it uses a full-resolution matrix calculation method to solve the contact rail state parameters in parallel.

[0127] The multimodal monitoring module is used to fuse the contact rail status parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data using a multimodal fusion method, output status assessment results, and trigger alarms when abnormalities occur.

[0128] The specific principles and execution processes of each unit in the straddle-type monorail status monitoring system disclosed in Embodiment 2 of this application are the same as those of the straddle-type monorail status monitoring method disclosed in Embodiment 1 of this application. Please refer to the corresponding parts of the straddle-type monorail status monitoring method disclosed in Embodiment 1 of this application, and they will not be repeated here.

[0129] Those skilled in the art will further 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, computing 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 implementation should not be considered beyond the scope of this application.

[0130] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0131] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for monitoring the condition of a straddle-type monorail, characterized in that, include: Acquire 3D point cloud data of the contact rail surface; The 3D point cloud data is preprocessed to obtain valid data; The RT matrix is ​​obtained based on binocular laser calibration. The effective data is mapped to the same coordinate system through dynamic compensation. After the data distribution feature matrix is ​​extracted by regional sampling, noise is processed according to the set process to obtain a calculation map reflecting the distribution feature matrix of the 3D point cloud data of the contact rail. In the computational graph, all specific targets and their feature values ​​are marked. After all feature values ​​are filtered in parallel by multiple threads, the data continuity is discriminated and fitted to construct the MaxPool global feature parameter matrix. All specific targets include contact rails, expansion joints, intermediate joints, turnout section insulators, mainline section insulators, center anchors, cable connection plates, and insulators; Based on the MaxPool global feature parameter matrix, a parallel parameter calculation matrix is ​​constructed. After the high-importance region is segmented according to the data distribution topology, the contact rail state parameters are solved in parallel using a full-resolution matrix calculation method. A multimodal fusion method is adopted to integrate the contact rail status parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data, outputting status assessment results and triggering alarms when abnormalities occur; In the computational graph, all specific targets and their feature values ​​are labeled. After all feature values ​​are filtered in parallel by multiple threads, the data continuity is discriminated and fitted to construct the MaxPool global feature parameter matrix, including: Based on the computational graph reflecting the distribution characteristics of the 3D point cloud data of the contact rail, the spatial topology, surface texture distribution and data gradient change characteristics carried by the computational graph are analyzed and evaluated to identify all specific targets and their corresponding feature values. Based on the aforementioned feature values, a multi-threaded parallel architecture is used to filter and label the preprocessed and feature-extracted multi-frame data matrix, select the effective data that matches the feature values ​​corresponding to various specific targets, and accurately store all specific targets and their corresponding feature values ​​in the matrix pointed to by the target category pointer, denoted as the target matrix; The target matrix is ​​evaluated using data continuity, and discrete invalid data is filtered out. After data continuity discrimination and filtering, the effective data corresponding to all specific targets are obtained and line segment fitting and curve segment fitting are performed respectively to obtain the corresponding fitting parameters. Based on all fitting parameters and feature information of various specific targets, the MaxPool global feature parameter matrix is ​​constructed.

2. The straddle-type monorail condition monitoring method according to claim 1, characterized in that, The RT matrix is ​​obtained based on binocular laser calibration, and the effective data is mapped to the same coordinate system through dynamic compensation, including: Based on binocular laser calibration technology, the RT matrix describing the relative attitude and position relationship between the sensor coordinate system and the contact rail reference coordinate system is obtained; The system receives dynamic compensation parameters in a unified coordinate system provided by the pre-set contact rail parameter compensation module, and maps the effective data to the same coordinate system based on the RT matrix.

3. The straddle-type monorail condition monitoring method according to claim 1, characterized in that, The data distribution feature matrix extracted through regional sampling includes: Based on the single-frame 3D point cloud data mapped to the same coordinate system, the contact rail surface is sampled by a differential network and texture extraction operator to extract a data distribution feature matrix that characterizes the surface feature distribution of the contact rail.

4. The straddle-type monorail condition monitoring method according to claim 1, characterized in that, The 3D point cloud data is preprocessed to obtain valid data including: Based on prior data, invalid values ​​are set to zero for the surface depth map and surface grayscale map corresponding to the 3D point cloud data, respectively. The surface depth map and surface grayscale map, after invalid values ​​have been set to zero, are filtered for valid values ​​using a logical AND operation, as follows: For valid data, This is a surface depth map. This is a grayscale image of the surface.

5. The straddle-type monorail condition monitoring method according to claim 1, characterized in that, The noise processing procedure includes sequentially processing large-scale outliers and local noise.

6. The straddle-type monorail condition monitoring method according to claim 1, characterized in that, The high-importance areas include the contact rail surface and sides, intermediate joint welds, and expansion joints.

7. The straddle-type monorail condition monitoring method according to claim 1, characterized in that, By integrating the contact rail status parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data, the system outputs a status assessment result and triggers an alarm when an anomaly occurs, including: By integrating parallel solutions, contact rail state parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data are obtained. Through multi-dimensional feature cross-validation and comprehensive analysis, state assessment results are generated. The status assessment results cover three levels: normal operation status, Level 1 abnormal status, and Level 2 abnormal status. When the status assessment result indicates normal operation, no action is taken; When the status assessment result is a Level 1 abnormal state, an early warning message is output and the abnormal data is recorded and archived. When the status assessment result is a level 2 abnormal state, the hierarchical alarm mechanism is automatically triggered, and the abnormal location, abnormal type and characteristic parameters are synchronously fed back to the monitoring terminal.

8. A straddle-type monorail condition monitoring system, implemented by the straddle-type monorail condition monitoring method as described in claim 1, characterized in that, include: The data acquisition module is used to acquire 3D point cloud data of the contact rail surface; The preprocessing module is used to preprocess the 3D point cloud data to obtain effective data; The feature extraction module is used to obtain the RT matrix based on binocular laser calibration, map the effective data to the same coordinate system through dynamic compensation, extract the data distribution feature matrix through regional sampling, process noise according to the set process, and obtain a calculation map reflecting the distribution feature matrix of the 3D point cloud data of the contact rail. The matrix construction module is used to identify all specific targets and their feature values ​​in the computation graph. After all feature values ​​are filtered in parallel by multiple threads, the data continuity is discriminated and fitted to construct the MaxPool global feature parameter matrix. All specific targets include contact rails, expansion joints, intermediate joints, turnout section insulators, mainline section insulators, center anchors, cable connection plates, and insulators; The state parameter calculation module is used to construct a parallel parameter calculation matrix based on the MaxPool global feature parameter matrix. After dividing the high-importance region according to the data distribution topology, it uses a full-resolution matrix calculation method to solve the contact rail state parameters in parallel. The multimodal monitoring module is used to fuse the contact rail status parameters, data acquisition timestamps, contact rail surface parameters, high-definition imaging data of the contact rail and its associated workpieces, and surface defect data using a multimodal fusion method, output status assessment results, and trigger alarms when abnormalities occur.

Images