A data analysis-based switch machine fault identification method and system

Abstract

The application discloses a kind of based on data analysis's switch machine fault identification method and system, it is related to switch machine fault identification technical field, including the following steps: before three-dimensional scanning, the whole surface of the connection area of switch rail heel and switch rod is spatially divided, the whole surface is divided into several identical sub-regions;After completing sub-region division, for each sub-region, a full initial scanning is carried out using the preset reference point cloud spatial distribution density, to obtain the basic surface structure data in the sub-region.The application can effectively identify the high complex surface area such as switch rail root, rotating interface and orifice by automatic prediction of sub-region surface complexity using convolutional neural network, and realizes point cloud encryption in high complex area, ensures that local geometric details are fully expressed, and improves the identification ability of typical faults such as micro-cracks, orifice deformation and eccentric wear.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of switch machine fault identification, and particularly relates to a switch machine fault identification method and system based on data analysis. BACKGROUND

[0002] The switch machine (also known as a turnout controller) is a core device in the railway signal system, mainly used to drive and lock the switch part of the turnout (commonly known as "railway turnout"), to realize the safe and smooth transfer of trains from one track to another. Its basic principle is to control the rotation of the switch rail through a power, hydraulic or electro-hydraulic hybrid drive device, to ensure that the switch rail is in close contact with the basic rail or separated, to complete the track conversion, and at the same time, through the locking mechanism, to ensure that the turnout can be firmly locked after positioning, to prevent the switch rail from being displaced due to external force or train impact, to ensure the safety of train operation. Modern switch machines are also equipped with electrical detection, remote control and state feedback functions, which can be linked with the dispatching command center or signal interlocking system, to realize automatic control and real-time monitoring, greatly improving the safety and scheduling efficiency of railway transportation.

[0003] When identifying switch machine faults based on data analysis, image technology is often used to visually monitor and diagnose the key components and working conditions of the switch machine. Common image technologies include video monitoring, infrared imaging, structured light or laser scanning, image recognition and deep learning target detection technology, mainly used to identify common faults in the switch rail, basic rail, switch machine housing, transmission rod, locking device and other parts. For example, high-definition video and image recognition technology can detect insufficient adhesion between the switch rail and the basic rail, switch rail jamming, foreign matter intrusion, switch machine box damage and other external faults; infrared imaging can monitor the temperature rise of switch machine drive motors, locking electromagnets and other components, to assist in identifying potential electrical or mechanical overload faults; three-dimensional scanning or structured light technology can be used to monitor the wear, deformation, misalignment and other structural defects of the turnout and switch mechanism components.

[0004] Three-dimensional scanning technology is used to identify faults in the curved surface of the switch rail heel and the connection area of the switch rod, which mainly functions to accurately detect structural defects such as deformation, wear, misalignment and micro-cracks in the key connection part. The switch rail heel is the direct connection point of the switch rail, switch rod and locking device transmission system, with complex curved surface structure, multiple special-shaped interfaces and transition surfaces, and is the most stressed area during long-term operation of the switch machine, prone to problems such as fatigue bending, hole wear, hole cracking, welding deformation and local misalignment. High-precision three-dimensional scanning can completely obtain the real geometric shape of this area, and by comparing point cloud fitting with the reference model, it can effectively identify small deformations, gap abnormalities and connection instability at the interface between the switch rail heel and the switch rod, providing reliable geometric diagnostic basis for preventing switch rail movement failure, locking abnormalities or poor turnout switching faults.

[0005] The prior art has the following shortcomings:

[0006] In the prior art, when identifying faults of the curved surface of the connection area between the heel of the switch rail and the switch rod by three-dimensional scanning, a uniform distribution density of point clouds is assigned to the curved surface of the connection area based on experience. However, the complexity of the curved surface structure of the connection area between the heel of the switch rail and the switch rod is significantly different, especially in the local areas such as the root of the switch rail, the rotating interface and the orifice, and the geometric structure of these areas is mostly characterized by small-scale features and complex curved surface combinations. Since the uniform distribution density of point clouds cannot be adjusted differently for complex curved surfaces, the distribution of point clouds in the high-complexity areas is insufficient, and the geometric details of the local areas cannot be effectively obtained, so that the typical defects including orifice wear, cracks, slight deformation and gap abnormalities cannot be accurately identified, and faults such as micro-cracks in the root of the switch rail, ovalization of the transmission orifice and eccentric wear are likely to be missed, which may directly threaten the safe and reliable operation of the switch machine.

[0007] The above information disclosed in the background section is only used to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0008] The purpose of the present application is to provide a switch machine fault identification method and system based on data analysis, which can effectively identify high-complexity curved surface areas such as the root of the switch rail, the rotating interface and the orifice by automatically predicting the complexity of the curved surface of the sub-area through a convolutional neural network, and can realize point cloud encryption in high-complexity areas to ensure that the local geometric details are fully expressed and the identification ability of typical faults such as micro-cracks, orifice deformation and eccentric wear is improved. At the same time, the reference point cloud density is maintained in the low-complexity area to avoid redundant sampling and reduce data redundancy and subsequent processing burden, thereby solving the problems of insufficient point clouds in high-complexity areas and waste of scanning resources in low-complexity areas in the prior art to solve the problems in the background.

[0009] To achieve the above purpose, the present application provides the following technical scheme: a switch machine fault identification method based on data analysis, comprising the following steps:

[0010] Before three-dimensional scanning, the overall curved surface of the connection area between the heel of the switch rail and the switch rod is spatially divided, and the overall curved surface is divided into a plurality of sub-areas of the same size;

[0011] After the sub-area division is completed, for each sub-area, a preset reference point cloud spatial distribution density is used for an initial full-range scanning to obtain basic curved surface structure data in the sub-area;

[0012] The key features reflecting the complexity of the curved surface structure are extracted from the acquired point cloud data of each sub-region, and the geometric variation characteristics of the internal curved surface of each sub-region are mined after feature engineering processing of the extracted key features, thereby providing quantitative description for subsequent judgment of the complexity of the curved surface;

[0013] The processed features are input into the pre-trained convolutional neural network model as feature vectors, and the complexity of the curved surface structure of each sub-region is predicted by the neural network, thereby realizing automatic and intelligent judgment of the complexity of the curved surface of the sub-region.

[0014] According to the complexity results of each sub-region predicted by the convolutional neural network model, the point cloud spatial distribution density of the corresponding sub-region is dynamically adjusted. If a sub-region is identified as a high-complexity structure curved surface, the original reference point cloud spatial distribution density is improved based on the complexity prediction result. If a sub-region is identified as a low-complexity structure curved surface, the original reference point cloud spatial distribution density is maintained.

[0015] According to the adjusted density allocation strategy, all sub-regions of the whole switch rail heel and switch rod connection area are re-scanned in all directions.

[0016] Preferably, the regular grid division method is used to divide the space of the whole curved surface of the switch rail heel and switch rod connection area, and the specific steps are as follows:

[0017] Firstly, a three-dimensional coordinate system of the switch rail heel and switch rod connection area is established, and the boundary range and coordinate axis direction of the region to be divided are determined according to the design reference of the switch machine or the preliminary point cloud data, thereby providing a spatial reference for subsequent grid division.

[0018] Secondly, according to the set grid division parameters (such as grid size, unit edge length), the whole region is regularly and uniformly divided along the X, Y and Z directions, and the curved surface space is divided into several cuboid or rectangular sub-regions with equal specifications, each of which has a unique space number and position coordinates.

[0019] Finally, the effective grid sub-regions intersected or contained by the actual curved surface of the switch rail heel and switch rod connection area are screened out by using the grid intersection analysis with the curved surface, and the invalid blank grids are removed, and only the effective sub-regions (the sub-regions participating in the subsequent feature extraction and point cloud density adjustment) are reserved.

[0020] Preferably, key features reflecting the complexity of the curved surface structure are extracted from the acquired point cloud data of each sub-region, wherein the extracted features include the fluctuation degree of the normal vectors of adjacent points in the sub-region and the density of the contour lines in the unit area of the sub-region. After feature engineering processing of the extracted key features, the normal vector fluctuation reference value and the contour line density reference value are generated respectively. The geometric variation characteristics of the internal curved surface of each sub-region are quantified by the normal vector fluctuation reference value and the contour line density reference value, thereby providing a basis for subsequent judgment of the complexity of the curved surface.

[0021] Preferably, the processed normal vector fluctuation reference value and the contour line density reference value are input as feature vectors into a convolutional neural network model trained in advance. The complexity score is generated by the neural network, and the complexity of the curved surface structure of each sub-region is regressed and predicted based on the complexity score, thereby realizing the automatic and intelligent judgment of the complexity of the curved surface of the sub-region.

[0022] Preferably, according to the complexity results of each sub-region predicted by the convolutional neural network model, the point cloud spatial distribution density of the corresponding sub-region is dynamically adjusted. The specific steps are as follows:

[0023] According to the complexity prediction results of the sub-region by the convolutional neural network model, the complexity score of the sub-region is obtained, and a complexity score reference threshold is introduced as a criterion for distinguishing between high complexity curved surfaces and low complexity curved surfaces. In order to realize the continuity of density adjustment, a point cloud density adjustment factor is introduced to determine the point cloud density increment multiple of the region. The calculation formula of the point cloud density adjustment factor is as follows:

[0024]

[0025] , wherein: λ i is the point cloud density adjustment factor of the sub-region, which determines the point cloud density increment multiple of the sub-region, α y is a density enhancement coefficient, which controls the density improvement range of the high complexity region, β y is a nonlinear control parameter, which is used to control the curve growth of the point cloud density adjustment factor, Complexity score is the complexity score of the sub-region output by the convolutional neural network model, T c is the reference threshold of the complexity score. When Complexity score >T c , it is determined as a high complexity structure curved surface, and when Complexity score ≤T c , it is determined as a low complexity structure curved surface.

[0026] Preferably, after obtaining the point cloud density adjustment factor λ iThen, based on the original baseline point cloud spatial distribution density, the actual point cloud spatial distribution density of each sub-region is adaptively adjusted, and the adjustment formula is as follows:

[0027] ρ i =ρ0·λ i

[0028] , where: ρ i ρ0 represents the spatial distribution density of the point cloud after adaptive adjustment for the sub-region, i.e., the adjusted sampling density, while ρ0 represents the baseline spatial distribution density of the point cloud.

[0029] Preferably, the specific steps for generating a reference value for normal vector fluctuation after performing feature engineering processing on the fluctuation degree of the normal vectors of adjacent points within a sub-region are as follows:

[0030] For the point cloud within each sub-region, denoted as point set P, P = {p i For each pair of adjacent points in Point p i For each point in the neighborhood of a given point, calculate the direction cosine of that neighboring point with respect to the normal vector. The expression for this calculation is:

[0031] C ij =|n i ·n j |

[0032] , where: n i and n j Points p i and point p j The unit normal vector, C ij Point p i and point p j The cosine of the angle between the unit normal vectors, C ij ∈[0,1], the closer the value is to 1, the closer the direction is, the closer the value is to 0, the more the direction is 90 degrees apart, the smaller the value is, the more drastic the change of the normal vector, |·| means that only the size of the angle between the unit normal vectors is considered and the orientation is not considered;

[0033] Next, construct the set of direction cosines C for all adjacent point pairs within this sub-region, C = {C ij};

[0034] The normal vector fluctuation reference value is calculated based on the direction cosine set C. The expression for the calculation is as follows:

[0035]

[0036] , where: N vi For the reference value of normal vector fluctuation, (1-C ij) represents the fluctuation intensity of the normal vector of the adjacent point pair, the larger the value, the stronger the fluctuation of the normal vector of the adjacent point pair, (1-C ij )∈[0,1],w ij is the weight of the adjacent edge of the adjacent point pair, which can be taken as where d ij is the distance of the adjacent point pair, which is used to emphasize that the contribution of the locally adjacent point pair fluctuation is greater, and П represents the product of all adjacent point pairs.

[0037] Preferably, after feature engineering of the density of the contour lines in the unit area, the specific steps of generating the contour line density reference value are as follows:

[0038] First, based on the point cloud data in the sub-region, the contour surface (isosurface) of the sub-region is projected along the direction perpendicular to the normal of the main curved surface (or the self-defined scanning direction), to generate a set of contour lines, denoted as where N c is the total number of contour lines generated in the sub-region, and after the contour lines are generated, the total bending degree of the contour lines is calculated as the geometric complexity index of the contour lines in the sub-region, and the expression is:

[0039]

[0040] where: TCC represents the total bending degree of the contour lines, C u is the u-th contour line, κ(s) is the curvature value at the arc length parameter s on the curve, and ds is the arc length infinitesimal along the contour line;

[0041] In order to comprehensively reflect the number and bending characteristics of the contour lines, the contour line density reference value is defined as follows:

[0042]

[0043] where: C di is the contour line density reference value, A is the actual projection area of the sub-region on the curved surface, γ is an area scaling factor, and 0<γ<1 (generally 0.5-0.8) is taken, which is used to avoid the virtual high phenomenon caused by the natural increase of the length of the contour lines in the sub-region with larger area.

[0044] A switch failure recognition system based on data analysis includes a region division module, an initial scanning module, a feature extraction and feature engineering module, a complexity prediction module, a point cloud density adaptive adjustment module, and an adaptive scanning execution module.

[0045] The region division module divides the overall curved surface of the connection area between the heel of the switch rail and the switch rod into several sub-regions of the same size before three-dimensional scanning.

[0046] An initial scanning module, after completing sub-region division, performs an initial scanning in all directions for each sub-region using a preset reference point cloud spatial distribution density to obtain basic curved surface structure data in the sub-region;

[0047] A feature extraction and feature engineering module extracts key features reflecting the complexity of the curved surface structure from the obtained point cloud data of each sub-region, processes the extracted key features, and mines the geometric variation characteristics of the internal curved surface of each sub-region to provide quantitative description for subsequent judgment of the complexity of the curved surface;

[0048] A complexity prediction module inputs the processed features into a pre-trained convolutional neural network model, performs regression prediction of the complexity of the curved surface structure of each sub-region through the neural network, and realizes automatic and intelligent judgment of the complexity of the curved surface of the sub-region.

[0049] A point cloud density self-adaptive adjustment module dynamically adjusts the point cloud spatial distribution density of the corresponding sub-region according to the complexity results of each sub-region predicted by the convolutional neural network model, and if a sub-region is identified as a high-complexity structure curved surface, the original reference point cloud spatial distribution density is improved based on the complexity prediction result; if a sub-region is identified as a low-complexity structure curved surface, the original reference point cloud spatial distribution density is maintained.

[0050] An adaptive scanning execution module re-performs omnidirectional scanning on all sub-regions of the whole switch rail heel and switch rod connection area according to the adjusted density allocation strategy.

[0051] In the above technical solution, the present application provides the following technical effects and advantages:

[0052] The present application can effectively identify high-complexity curved surface regions such as the switch rail root, the rotating interface, and the orifice through the automatic prediction of the complexity of the curved surface of the sub-region by the convolutional neural network, and can realize point cloud encryption in high-complexity regions to ensure that local geometric details are fully expressed and improve the identification ability of typical faults such as micro-cracks, orifice deformation, and eccentric wear. At the same time, the reference point cloud density is maintained in low-complexity regions to avoid redundant sampling, reduce data redundancy and subsequent processing burden, and solve the problems of insufficient point cloud in high-complexity regions and waste of scanning resources in low-complexity regions in the prior art, thereby providing effective technical support for high-precision and intelligent three-dimensional scanning and fault identification of the switch machine. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0054] Figure 1 A method flow chart of a switch machine fault identification method based on data analysis of the present application.

[0055] Figure 2 A module schematic diagram of a switch machine fault identification system based on data analysis of the present application. DETAILED DESCRIPTION

[0056] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of example implementations to those skilled in the art.

[0057] The present application provides a switch machine fault identification method based on data analysis as shown in Figure 1 The method comprises the following steps:

[0058] Before three-dimensional scanning, the overall curved surface of the switch rail heel and the switch rod connection area is spatially divided, and the overall curved surface is divided into a plurality of sub-regions of the same specification;

[0059] Specific division can be based on the CAD geometric model or the preliminary point cloud model of the switch machine, and the regular grid division (Grid Partitioning) or voxelization method is adopted to divide the complex free curved surface into a plurality of small-scale sub-regions. Each sub-region serves as a basic unit for subsequent curved surface complexity analysis and point cloud density adjustment. The role of this step is to refine the scanning unit of the target area, so that differentiated point cloud density control can be implemented for the structural complexity of different sub-regions in the subsequent process, avoiding the one-size-fits-all uniform scanning strategy for the overall curved surface.

[0060] The regular grid division method is adopted to spatially divide the overall curved surface of the switch rail heel and the switch rod connection area, and the specific steps are as follows:

[0061] First, a three-dimensional coordinate system of the switch rail heel and the switch rod connection area is established, and the boundary range and coordinate axis direction of the region to be divided are determined according to the design reference of the switch machine or the preliminary point cloud data, thereby providing a spatial reference for subsequent grid division;

[0062] Secondly, according to the set grid division parameters (such as grid size, unit edge length), the overall region is regularly and uniformly divided along the X, Y and Z directions, and the curved surface space is divided into a plurality of cuboid or rectangular sub-regions of the same specification. Each sub-region has a unique space number and position coordinates;

[0063] Finally, the effective grid sub-regions intersecting or containing the curved surface of the connection area between the heel of the switch rail and the switch rod are screened out by grid and curved surface intersection analysis, and the invalid blank grids are removed, and only the effective sub-regions (sub-regions participating in subsequent feature extraction and point cloud density adjustment) are reserved.

[0064] Through the above steps, the spatial division of the complex curved surface of the connection area between the heel of the switch rail and the switch rod can be realized, and a basis is provided for subsequent regional feature extraction and adaptive point cloud distribution.

[0065] After completing the sub-region division, a full initial scanning is performed on each sub-region by using a preset reference point cloud spatial distribution density, so as to obtain the basic curved surface structure data in the sub-region;

[0066] The reference point cloud density is usually determined according to the equipment performance and detection accuracy requirements, such as initially setting the point spacing to 1 mm or finer. During the full scanning process, the angle and position of the scanning device are adjusted to ensure that each sub-region is completely scanned from multiple angles without dead angles. The core role of this step is to quickly and comprehensively obtain the curved surface original point cloud data of the connection area between the heel of the switch rail and the switch rod, and to provide necessary three-dimensional spatial information for subsequent feature extraction and complexity analysis.

[0067] The key features reflecting the complexity of the curved surface structure are extracted from the obtained point cloud data of each sub-region, and the geometric variation characteristics of the internal curved surface of each sub-region are mined after feature engineering processing of the extracted key features, thereby providing quantitative description for subsequent judgment of the complexity of the curved surface.

[0068] The key features reflecting the complexity of the curved surface structure are extracted from the obtained point cloud data of each sub-region, and the geometric variation characteristics of the internal curved surface of each sub-region are mined after feature engineering processing of the extracted key features, thereby providing quantitative description for subsequent judgment of the complexity of the curved surface.

[0069] For each sub-region, if the normal vector fluctuation degree of adjacent points in the sub-region is high, it usually indicates that the geometric variation of the curved surface in the sub-region is violent in space, and the structure complexity is high. Specifically, the point cloud normal vector reflects the local orientation of the curved surface at each point. When the normal vector directions of adjacent points change frequently and significantly, it indicates that the curvature distribution of the sub-region is uneven, and there may be complex geometric structures such as sharp edges, concave-convex mutations, twists and bends, or multi-scale micro-features. Especially in the mechanical connection part, the welding transition area or the wear deformation concentrated area, the normal vector fluctuation often becomes a key indicator to reveal the high complexity of the curved surface.

[0070] The specific steps of generating the normal vector fluctuation reference value after feature engineering of the fluctuation degree of the normal vectors of adjacent points in the sub-region are as follows:

[0071] For each point cloud in the sub-region, denoted as point set P, P = {p i}, for each pair of adjacent points Wherein represents the points in the neighborhood of point p i , the directional cosine value of the normal vector of the adjacent point pair is calculated, and the expression for calculation is:

[0072] C ij = |n i ·n j |

[0073] , wherein: n i and n j are the unit normal vectors of point p i and point p j , C ij represents the cosine value of the angle between the unit normal vectors of point p i and point p j , C ij ∈ [0, 1], the value closer to 1 indicates the direction is close, the value closer to 0 indicates the direction difference is 90 degrees, the smaller the value, the more intense the normal vector changes, |·| represents only the size of the angle between the unit normal vectors is concerned and the direction is not concerned (i.e. the internal and external unit normal vectors are not distinguished);

[0074] Then, the directional cosine set C of all adjacent point pairs in the sub-region is constructed, C = {C ij};

[0075] The purpose of this step is to construct the fluctuation basis data set of the local normal vector, which provides the basis for subsequent direct quantification of the fluctuation degree, while avoiding the traditional mean or variance processing method and retaining the directional information of the fluctuation.

[0076] The normal vector fluctuation reference value is calculated based on the directional cosine set C, and the expression for calculation is:

[0077]

[0078] , wherein: N vi is the normal vector fluctuation reference value, (1-C ij ) represents the fluctuation intensity of the normal vector of the adjacent point pair, the larger the value, the stronger the fluctuation of the normal vector of the adjacent point pair, (1-C ij ) ∈ [0, 1], w ij is the weight of the adjacent edge of the adjacent point pair, which can be taken as where d ijis the distance of adjacent point pairs, which is used to emphasize that the contribution of local adjacent point pairs fluctuation is greater, and П represents the product of all adjacent point pairs;

[0079] When there are a large number of normal vector mutations in the sub-region (i.e., C ij is smaller), (1-C ij ) tends to 1, resulting in the normal vector fluctuation reference value close to 1; and when most of the normal vectors of the point pairs are similar, (1-C ij ) is close to 0, and the normal vector fluctuation reference value quickly decays to close to 0. In this way, the normal vector fluctuation reference value can highlight local strong fluctuation phenomena rather than global average changes, and is particularly sensitive to surfaces with obvious edges, concave-convex, small-scale complex features. The role of this step is to directly index and quantify the normal vector changes that are significantly fluctuated, which is used for accurate identification of high-complexity surfaces.

[0080] From the normal vector fluctuation reference value, the greater the performance value of the normal vector fluctuation reference value generated after feature engineering processing of the fluctuation degree of the normal vectors of adjacent points in the sub-region, the more complex the surface geometry change characteristics in the sub-region. The reason is that the normal vector fluctuation reference value is essentially a cumulative quantization of the direction change degree between the normal vectors of adjacent points. When the surface in the sub-region has a large number of curvature mutations, edges, corners, concave-convex, or complex local details, the difference in the direction of the normal vectors of adjacent points will significantly increase, which is manifested as the angle between the normal vectors of adjacent points tends to increase, thereby reducing the cosine value of the normal vector direction, resulting in an increase in the value of (1-C ij ) used in the index calculation, and finally making the normal vector fluctuation reference value overall increase. Therefore, the high and low of the normal vector fluctuation reference value is positively correlated with the severity of the local normal vector change of the surface. The greater the value, the more dramatic the surface fluctuation, the more detailed the surface, and the more complex the structure. Conversely, if the surface is smooth and changes gently, the difference between the normal vectors of adjacent points is small, and the normal vector fluctuation reference value decreases, reflecting that the surface is relatively simple and the structural changes are not obvious.

[0081] The density of the contour lines in the unit region is closely related to the surface geometry change characteristics. If the contour lines are densely distributed, it usually indicates that the surface in the sub-region has dramatic ups and downs, concave-convex changes, or curvature mutations, i.e., the surface geometry change characteristics are relatively complex. The contour line is essentially an isometric section of the surface in a certain direction. When the surface fluctuates gently, the spacing between the contour lines is large, and vice versa. When there are sharp turns, grooves, protrusions, curvature abrupt changes, or small-scale structures densely distributed in the local surface, the contour lines will be in a state of dense, twisted, or even overlapping. Therefore, the contour line density can be used as an important geometric feature to measure the complexity of the surface, which can effectively reflect whether there is a complex surface region with dramatic deformation or rich details in the connection area of the sharp rail heel and the switch rod.

[0082] The specific steps of generating the contour line density reference value after feature engineering processing of the contour line density in a unit area are as follows:

[0083] First, based on the point cloud data in the sub-region, the contour surface (isosurface) of the sub-region is projected along the direction perpendicular to the normal direction of the main curved surface (or the self-defined scanning direction), to generate a set of contour lines, denoted as Where N c is the total number of contour lines generated in the sub-region. After the contour lines are generated, the total bending degree of the contour lines is calculated as the geometric complexity index of the contour lines in the sub-region, and the expression for the calculation is:

[0084]

[0085] Where: TCC represents the total bending degree of the contour lines, C u is the u-th contour line, and κ(s) is the curvature value at the arc length parameter s on the curve. ds is the arc length infinitesimal along the contour line.

[0086] This step directly uses the shape features of the contour lines to measure the curvature variation of the contour lines in space. If the curved surface is steep and fluctuates, and the concave-convex is rich, the contour lines are dense and the bending degree is high, and the total bending degree of the contour lines naturally increases, reflecting the complexity of the geometric variation of the sub-region curved surface.

[0087] In order to comprehensively reflect the number and bending characteristics of the contour lines, the contour line density reference value is defined as follows:

[0088]

[0089] Where: C di is the contour line density reference value, A is the actual projection area of the sub-region on the curved surface, and γ is an area scaling factor, which is taken as 0<γ<1 (generally taken as 0.5-0.8), to avoid the virtual high phenomenon caused by the natural increase of the length of the contour lines in the sub-region with large area;

[0090] The contour line density reference value not only considers the morphological complexity (curvature bending degree) of the contour lines, but also realizes area normalization through A γ to avoid confusion between large-area simple curved surfaces and small-area high-complexity curved surfaces in the index. A high contour line density reference value indicates that the contour lines in the unit area are both numerous and curved, strongly indicating that the curved surface is a high-complexity structure; a low contour line density reference value indicates that the contour lines are sparse and the morphology is flat, indicating that the curved surface is relatively simple.

[0091] The greater the performance value of the contour density reference value generated after the feature engineering processing of the contour density of the unit region in the sub-region, the more complex the geometric variation characteristics of the surface in the sub-region, and vice versa. The fundamental reason is that the contour density reference value integrates the number of contours in the unit region and the curvature of the contour itself. If the geometric variation of the surface in the sub-region is severe, it is often accompanied by features such as ups and downs, concave and convex, broken lines, and sudden changes, which are manifested as tight distribution of contours in space and significant increase in curve curvature, resulting in a significant increase in the value of the contour density reference value after integration and accumulation. When the surface in the sub-region is relatively flat or changes slowly, the contours are sparse and the curvature changes gently, resulting in a small value of the contour density reference value. Therefore, the contour density reference value can objectively reflect the strength of the local geometric variation of the surface, and the greater the performance value, the more complex the surface features and the more delicate the structure.

[0092] The processed features are input into a pre-trained convolutional neural network (CNN) model, and the neural network is used to regress and predict the surface structure complexity of each sub-region, achieving automatic and intelligent determination of the surface complexity of the sub-region.

[0093] The processed normal vector fluctuation reference value and contour density reference value are input into a pre-trained convolutional neural network model as feature vectors, and the neural network is used to generate a complexity score, based on which the surface structure complexity of each sub-region is regressed and predicted, achieving automatic and intelligent determination of the surface complexity of the sub-region.

[0094] The pre-trained convolutional neural network model refers to the convolutional neural network (CNN) used in the complexity determination method of the present application, which is not trained temporarily after the sub-region feature extraction, but is a deep neural network model with strong surface complexity feature recognition and regression prediction ability based on a large number of switch rail connecting area surface point cloud data with known complexity labels through a systematic training process in advance. The design of the convolutional neural network is usually around the surface complexity determination task, and its structure includes basic modules such as input layer, several groups of convolutional layer, pooling layer, feature fusion layer, fully connected layer and output layer, which can perform feature extraction, feature combination and nonlinear mapping learning on the input feature vector (i.e. normal vector fluctuation reference value and contour line density reference value). The core role of the convolutional neural network is to use its ability to handle spatial correlation, graphical features and local feature combinations to establish an implicit relationship model between feature indicators (input) and surface complexity (output). In this technical solution, the CNN can learn the internal nonlinear mapping rules between normal vector fluctuation and contour line density and actual surface complexity, and can effectively handle the situation of complex feature distribution and high spatial overlap of features to complete the complexity regression prediction. The meaning of pre-training is that the neural network model has completed repeated training, optimization and verification based on historical data before the deployment or implementation of the present application, has stable prediction ability, can be directly deployed in the complexity prediction process, and realizes the rapid reasoning from feature input to complexity prediction output.

[0095] Further, the pre-trained convolutional neural network model is usually built on a large number of actually collected or simulated generated point cloud data sets of the switch rail heel and the switch rod connection area of the switch machine. These training data sets not only contain a large number of real curved surface samples of high complexity, low complexity, multi-scale, and multi-morphology, but also are manually or semi-automatically labeled by professional engineers or according to three-dimensional curved surface morphology standards for each curved surface area, and are given corresponding curved surface complexity labels (for example, complexity classification, complexity score). In the process of building the model, first, the curved surface samples are divided into a plurality of sub-regions consistent with the actual detection process, and the normal vector fluctuation reference value and the contour line density reference value of each sub-region are calculated to form a feature input set; then, through end-to-end learning of the convolutional neural network, the model gradually fits the corresponding relationship between the input features and the target complexity label. In the training process, mean square error (MSE) or mean absolute error (MAE) is usually used as the loss function, combined with optimizers such as Adam and SGD, to continuously optimize the network parameters until the model reaches the ideal prediction accuracy and generalization ability on the validation set. The trained neural network model is regarded as a pre-trained model, which can be directly called in the actual fault detection and point cloud processing process to automatically determine the complexity of unknown sub-regions, without the need for retraining or parameter adjustment in actual application, thereby significantly improving the response speed and judgment accuracy of the detection system.

[0096] The pre-trained convolutional neural network model not only plays a core role in the curved surface complexity intelligent determinator, but also has many technical advantages. First, pre-training can effectively solve the nonlinear, multi-scale coupling problem between features and outputs in curved surface complexity determination. For example, the normal vector fluctuation reference value and the contour line density reference value are not simply linearly related to the curved surface complexity, but often exhibit complex nonlinear, multi-feature combination characteristics. For example, even if the normal vector fluctuation reference value is low at the heel of the switch rail, the contour line density reference value is abnormally high, which still shows the characteristics of complex curved surface. Conventional linear models or threshold determinations cannot accurately model such implicit relationships. The convolutional neural network can automatically learn and extract the deep relationship between the two and complexity by relying on the mechanism of spatial feature extraction by convolution kernel + multi-layer nonlinear activation + feature fusion, effectively solving the misjudgment or omission problem caused by complex curved surface feature distribution. Second, pre-training helps to decouple the complex model training process and the actual detection process, so that the detection process only needs to perform forward inference to quickly output the complexity prediction result of the sub-region, which has the advantages of high efficiency, stability, and real-time, greatly improving the closed-loop efficiency of point cloud collection-complexity recognition-adaptive scanning.

[0097] The convolutional neural network model is not specifically limited, and can realize the generation of the normal vector fluctuation reference value N vi and the contour line density reference value C di The convolutional neural network model can be used for comprehensive analysis to generate a complex score Complexity score In order to realize the technical scheme of the present application, a specific implementation is provided; the expression for generating the complex score Complexity score is as follows: Complexity score = h w *N vi +h l *C di , wherein h w , h l are preset proportion coefficients of the normal vector fluctuation reference value N vi and the contour line density reference value C di , and h w , h l are greater than 0.

[0098] The preset proportion coefficients (i.e., h w and h l ) refer to coefficients for assigning weights to different input feature indicators in the calculation process of generating the complex score. These coefficients are constant values that are set in advance in the model training or system design stage according to experience, data statistics or tuning results, and are used to balance the influence degree of the normal vector fluctuation reference value and the contour line density reference value on the final complex score. Specifically, the role of the preset proportion coefficients is to control the contribution weight of the two indicators to the score result, for example, if the normal vector fluctuation reference value can better reflect the surface complexity in a certain scene, h w can be set to be larger to enhance its influence; on the contrary, if the contour line density is a more critical judgment basis, the weight of h l can be appropriately increased. Since both of the two coefficients are "greater than 0", and the values are adjustable, the system can select appropriate proportion coefficient combinations through experimental optimization, so that the generated complex score is more consistent with the actual structure complexity distribution, and the accuracy and adaptability of intelligent discrimination are improved.

[0099] As can be seen from the complex score, the greater the performance value of the normal vector fluctuation reference value generated after feature engineering processing of the fluctuation degree of the normal vectors of adjacent points in the sub-region, the greater the performance value of the contour line density reference value generated after feature engineering processing of the density of the contour lines in the unit region in the sub-region, that is, the greater the performance value of the complex score generated by the convolutional neural network model trained in advance for the regression prediction of the surface structure complexity of each sub-region, which indicates that the surface geometric change characteristics in the sub-region are more complex, and vice versa.

[0100] According to the complexity of each sub-region predicted by the convolutional neural network (CNN) model, the point cloud spatial distribution density of the corresponding sub-region is dynamically adjusted. If a certain sub-region is identified as a high-complexity structure surface, the original baseline point cloud spatial distribution density is improved based on the complexity prediction result to improve the ability to capture surface details. If a certain sub-region is identified as a low-complexity structure surface, the original baseline point cloud spatial distribution density is maintained to ensure detection accuracy while avoiding global over-encryption, optimizing scanning resource allocation, achieving high-complexity point cloud encryption, and maintaining baseline density in low-complexity areas, balancing accuracy and efficiency.

[0101] According to the complexity of each sub-region predicted by the convolutional neural network (CNN) model, the point cloud spatial distribution density of the corresponding sub-region is dynamically adjusted. If a certain sub-region is identified as a high-complexity structure surface, the original baseline point cloud spatial distribution density is improved based on the complexity prediction result to improve the ability to capture surface details. If a certain sub-region is identified as a low-complexity structure surface, the original baseline point cloud spatial distribution density is maintained to ensure detection accuracy while avoiding global over-encryption, optimizing scanning resource allocation, achieving high-complexity point cloud encryption, and maintaining baseline density in low-complexity areas, balancing accuracy and efficiency.

[0102] According to the complexity of each sub-region predicted by the convolutional neural network (CNN) model, the point cloud spatial distribution density of the corresponding sub-region is dynamically adjusted. If a certain sub-region is identified as a high-complexity structure surface, the original baseline point cloud spatial distribution density is improved based on the complexity prediction result to improve the ability to capture surface details. If a certain sub-region is identified as a low-complexity structure surface, the original baseline point cloud spatial distribution density is maintained to ensure detection accuracy while avoiding global over-encryption, optimizing scanning resource allocation, achieving high-complexity point cloud encryption, and maintaining baseline density in low-complexity areas, balancing accuracy and efficiency.

[0103]

[0104] , wherein: λ i is the point cloud density adjustment factor of the sub-region, which determines the point cloud density increment multiple of the sub-region, α y is the density enhancement coefficient, which controls the density improvement amplitude of the high-complexity region, β y is a non-linear control parameter for controlling the curve growth of the point cloud density adjustment factor, Complexity score is the complexity score of the sub-region output by the convolutional neural network model, T c is the reference threshold of the complexity score. When Complexity score >T c , it is determined as a high-complexity structure surface. When Complexity score ≤T c , it is determined as a low-complexity structure surface.

[0105] The purpose of this step is to design a smooth and controllable point cloud density adjustment factor to achieve a non-linear mapping between complexity and point cloud density, so that the higher the complexity, the more significant the point cloud encryption, while the low-complexity region remains unchanged, avoiding abrupt encryption problems caused by hard threshold, and improving the flexibility and intelligence of density adjustment.

[0106] In the step of obtaining the point cloud density adjustment factor λ iAfter that, based on the original reference point cloud spatial distribution density, the actual point cloud spatial distribution density of each sub-region is adaptively adjusted, and the adjustment formula is as follows:

[0107] ρ i = ρ0· λ i

[0108] , wherein: ρ i is the point cloud spatial distribution density of the sub-region after adaptive adjustment, that is, the adjusted sampling density, and ρ0 is the reference point cloud spatial distribution density;

[0109] The purpose of this step is to directly increase the point cloud density of the high complexity region according to the relative relationship between the sub-region complexity score and the reference threshold, that is, to multiply the original reference density by the encryption coefficient λ i >1; and for the low complexity region, when Complexity score ≤T c , λ i =1, so that the reference density remains unchanged, thereby realizing point cloud encryption in the high complexity region and maintaining the reference density in the low complexity region, effectively balancing the detail capture capability and the scanning resource utilization efficiency.

[0110] According to the adjusted density distribution strategy, all sub-regions in the whole switch rail heel and switch rod connection area are re-scanned in all directions;

[0111] In this stage, the high complexity region will be sampled with higher point cloud density, and the low complexity region will maintain the reference density, so as to obtain a set of point cloud data with variable precision, local encryption and global efficiency. The finally obtained point cloud data can completely and accurately reflect the tiny geometric features of the complex regions such as the switch rail heel, the rotating interface and the orifice, effectively improving the recognition accuracy of typical faults such as cracks, abrasion, orifice deformation and switch rail heel eccentricity. The adaptive scanning modeling process oriented to complexity is completed, and a high-quality point cloud data set meeting the subsequent fault recognition requirements is generated.

[0112] By the above-mentioned data analysis-based switch failure identification method, the complexity difference of the curved surface of the switch rail heel and the connection area of the switch rod can be accurately perceived and the point cloud density can be self-adaptively regulated, and the quality of the three-dimensional scanning data and the reliability of the failure identification are significantly improved. Specifically, the method can effectively identify the high-complexity curved surface regions such as the switch rail root, the rotating interface and the orifice by automatically predicting the complexity of the curved surface of the sub-regions through the convolutional neural network, and the point cloud can be encrypted in the high-complexity regions to ensure that the local geometric details are fully expressed and the identification capability for typical faults such as micro-cracks, orifice deformation and eccentric wear is improved; at the same time, the reference point cloud density is maintained in the low-complexity regions to avoid redundant sampling and reduce data redundancy and subsequent processing burden. The scheme takes into account the detection accuracy and system efficiency, solves the problems of insufficient point cloud in the high-complexity region and waste of scanning resources in the low-complexity region in the prior art, and provides effective technical support for high-precision and intelligent three-dimensional scanning and failure identification of the switch machine.

[0113] The application provides a data analysis-based switch failure identification system as shown in the accompanying drawings. Figure 2 The application provides a data analysis-based switch failure identification system as shown in the accompanying drawings.

[0114] The region division module divides the overall curved surface of the switch rail heel and the connection area of the switch rod into a plurality of sub-regions of the same size in space before three-dimensional scanning.

[0115] The initial scanning module performs an initial scanning on each sub-region in a full range by using a preset reference point cloud spatial distribution density after the sub-regions are divided, so as to obtain basic curved surface structure data in the sub-region.

[0116] The feature extraction and feature engineering module extracts key features capable of reflecting the complexity of the curved surface structure from the obtained point cloud data of each sub-region, and processes the extracted key features through feature engineering to mine the geometric variation characteristics of the internal curved surface of each sub-region, thereby providing quantitative description for subsequent judgment of the complexity of the curved surface.

[0117] The complexity prediction module inputs the processed features into a convolutional neural network model trained in advance, and performs regression prediction on the complexity of the curved surface structure of each sub-region through the neural network, thereby realizing automatic and intelligent judgment of the complexity of the curved surface of the sub-region.

[0118] The point cloud density self-adaptive adjustment module dynamically adjusts the point cloud spatial distribution density of the corresponding sub-region according to the complexity result of each sub-region predicted by the convolutional neural network model, if a certain sub-region is identified as a high-complexity structure surface, the original benchmark point cloud spatial distribution density is improved based on the complexity prediction result; if a certain sub-region is identified as a low-complexity structure surface, the original benchmark point cloud spatial distribution density is maintained.

[0119] The adaptive scanning execution module re-performs omnidirectional scanning on all sub-regions of the whole switch rail heel and switch rail connecting area according to the adjusted density distribution strategy.

[0120] The embodiment of the present application provides a kind of based on data analysis's switch machine fault identification method, is realized by the above-mentioned based on data analysis's switch machine fault identification system, and the specific method and process of a kind of based on data analysis's switch machine fault identification system are described in the above-mentioned embodiment of based on data analysis's switch machine fault identification method, and details are not repeated here.

[0121] The above formula is dimensionless to calculate its numerical value, the formula is obtained by collecting a large amount of data to simulate the nearest real situation, and the preset parameter in the formula is set by the person skilled in the art according to the actual situation.

[0122] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0123] The above only describes some exemplary embodiments of the present application by way of illustration, without doubt, for ordinary skilled in the art, without deviating from the spirit and scope of the present application, the described embodiments can be modified in various ways. Therefore, the above drawings and description are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

Claims

1. A method for fault identification of switch machines based on data analysis, characterized in that, Includes the following steps: Before 3D scanning, the overall curved surface of the connection area between the switch rail and the switch rod is spatially divided into several sub-regions of the same size. After the sub-regions are divided, a full-range initial scan is performed on each sub-region using a preset benchmark point cloud spatial distribution density to obtain the surface structure data within that sub-region. For the point cloud data of each sub-region that has been acquired, key features that can reflect the complexity of the surface structure are extracted. After feature engineering processing of the extracted key features, the geometric change characteristics of the surface inside each sub-region are explored to provide a quantitative description for subsequent judgment of the surface complexity. The processed features are input as feature vectors into a pre-trained convolutional neural network model. The neural network performs regression prediction on the surface structure complexity of each sub-region, thereby realizing the automated and intelligent determination of the surface complexity of the sub-region. Based on the complexity results of each sub-region predicted by the convolutional neural network model, the spatial distribution density of the point cloud in the corresponding sub-region is dynamically adjusted. If a sub-region is identified as a highly complex structural surface, the spatial distribution density of the original baseline point cloud is improved based on its complexity prediction results. If a sub-region is identified as a low-complexity structure surface, the original reference point cloud spatial distribution density is maintained. Based on the adjusted density allocation strategy, all sub-regions of the entire connection area between the switch rail and the switch bar are re-scanned in all directions. Based on the complexity results of each sub-region predicted by the convolutional neural network model, the spatial distribution density of the point cloud in the corresponding sub-region is dynamically adjusted. The specific steps are as follows: Based on the complexity prediction results of the convolutional neural network model for the sub-region, a complexity score is obtained for that sub-region. A complexity score reference threshold is introduced as a criterion to distinguish between highly complex and low-complexity surfaces. A point cloud density adjustment factor is introduced to determine the point cloud density increment multiple for that region. The formula for calculating the point cloud density adjustment factor is as follows: in: This is the point cloud density adjustment factor for the sub-region. The density enhancement coefficient controls the degree of density increase in highly complex regions. This is a nonlinear control parameter used to control the curve growth of the point cloud density adjustment factor. A complex score for a sub-region output by a convolutional neural network model. As a reference threshold for complex scoring, when When it is determined to be a highly complex structural surface, When the surface is classified as a low-complexity structure, it is determined to be a surface with a low complexity. Obtain the point cloud density adjustment factor for the sub-region. Then, based on the original baseline point cloud spatial distribution density, the actual point cloud spatial distribution density of each sub-region is adaptively adjusted, and the adjustment formula is as follows: in: This refers to the adaptively adjusted spatial distribution density of the point cloud in the sub-region, i.e., the adjusted sampling density. The baseline cloud spatial distribution density.

2. The method for identifying switch machine faults based on data analysis according to claim 1, characterized in that, The overall curved surface of the connection area between the switch rail heel and the switch bar is spatially divided using a regular mesh generation method. The specific steps are as follows: Establish a three-dimensional spatial coordinate system for the connection area between the switch rail and the switch bar. Based on the preliminary point cloud data, determine the boundary range and coordinate axis direction of the area to be divided, providing a spatial reference for subsequent mesh division. Based on the set grid division parameters, the entire region is regularly and uniformly divided along the X, Y, and Z directions, dividing the curved space into several cubes of equal size. Each sub-region has a unique spatial number and position coordinates. By using mesh-surface intersection analysis, valid mesh sub-regions that intersect with or are contained within the actual switch rail heel and switch bar connection area are selected, invalid blank meshes are removed, and only valid sub-regions are retained.

3. The method for fault identification of a switch machine based on data analysis according to claim 1, characterized in that, For the point cloud data of each sub-region that has been acquired, key features that can reflect the complexity of the surface structure are extracted. The extracted features include the degree of fluctuation of the normal vector of adjacent points in the sub-region and the density of contour lines in a unit area of ​​the sub-region. After feature engineering processing of the extracted key features, reference values ​​for normal vector fluctuation and contour line density are generated respectively. The geometric change characteristics of the surface inside each sub-region are quantified by the reference values ​​for normal vector fluctuation and contour line density.

4. The method for fault identification of a switch machine based on data analysis according to claim 3, characterized in that, The processed normal vector fluctuation reference value and contour line density reference value are used as feature vectors and input into a pre-trained convolutional neural network model. The neural network generates a complex score, and the surface structure complexity of each sub-region is predicted by regression based on the complex score, so as to realize the automated and intelligent determination of the surface complexity of the sub-region.

5. The method for identifying switch machine faults based on data analysis according to claim 3, characterized in that, The specific steps for generating reference values ​​for normal vector fluctuation after performing feature engineering processing on the fluctuation degree of the normal vectors of adjacent points within a sub-region are as follows: The point cloud within each sub-region is denoted as a point set. P , For each pair of adjacent points ,in Point For each point in the neighborhood of a given point, calculate the direction cosine of that neighboring point with respect to the normal vector. The expression for this calculation is: in: and Points and points The unit normal vector, Point and points The cosine of the angle between the unit normal vectors. ; Construct the set of direction cosines for all adjacent point pairs within this sub-region. C , ; Based on direction cosine set C The normal vector fluctuation reference value is calculated using the following expression: in: This serves as a reference value for the normal vector fluctuation. This represents the fluctuation intensity of adjacent points relative to the normal vector. , This represents the weight of the adjacent point relative to its adjacent edges. This represents the product of all adjacent pairs of points.

6. The method for fault identification of a switch machine based on data analysis according to claim 3, characterized in that, The specific steps for generating contour line density reference values ​​after performing feature engineering processing on the density of contour lines within a unit area are as follows: Based on the point cloud data within the sub-region, a contour projection is performed on the surface of the sub-region along a direction perpendicular to the normal of the main surface, generating a set of contour lines, denoted as . ,in The total number of contour lines generated within this sub-region is given. After the contour lines are generated, the total curvature of the contour lines is calculated as an index of the geometric complexity of the contour lines within the sub-region. The calculation expression is as follows: in: Indicates the total curvature of the contour lines. For the first u contour lines, Arc length parameter on the curve s The curvature value at that point, ds Let the arc length be a infinitesimal element along the contour line; The density reference value of contour lines comprehensively reflects the quantity and curvature characteristics of contour lines. The formula for calculating the density reference value of contour lines is as follows: in: This is a reference value for dense contour lines. A This represents the actual projected area of ​​the sub-region on the curved surface. Let be the area scaling factor, and take . .

7. A data analysis-based switch machine fault identification system, used to implement the data analysis-based switch machine fault identification method according to any one of claims 1-6, characterized in that, It includes a region segmentation module, an initial scanning module, a feature extraction and feature engineering module, a complexity prediction module, a point cloud density adaptive adjustment module, and an adaptive scanning execution module; The region division module spatially divides the overall curved surface of the connection area between the switch rail heel and the switch bar into several sub-regions of the same size before 3D scanning. After the initial scanning module completes the sub-region division, it performs a full-range initial scan for each sub-region using a preset benchmark point cloud spatial distribution density to obtain the surface structure data within that sub-region. The feature extraction and feature engineering module extracts key features that reflect the complexity of the surface structure from the acquired point cloud data of each sub-region. After performing feature engineering on the extracted key features, it explores the geometric change characteristics of the surface inside each sub-region, providing a quantitative description for subsequent judgment of the surface complexity. The complexity prediction module takes the processed features as feature vectors and inputs them into a pre-trained convolutional neural network model. The neural network performs regression prediction on the surface structure complexity of each sub-region, realizing the automated and intelligent determination of the surface complexity of the sub-region. The point cloud density adaptive adjustment module dynamically adjusts the spatial distribution density of the point cloud in the corresponding sub-region based on the complexity results predicted by the convolutional neural network model. If a sub-region is identified as a highly complex structural surface, the original baseline point cloud spatial distribution density is increased based on its complexity prediction results. If a sub-region is identified as a low-complexity structure surface, the original reference point cloud spatial distribution density is maintained. The adaptive scan execution module re-scans all sub-regions of the entire switch rail heel and switch bar connection area according to the adjusted density allocation strategy.

Images