A calculation method based on contact line profile three-dimensional splicing and profile wear

Abstract

The application provides a calculation method based on contact line profile three-dimensional splicing and profile wear, and belongs to the field of contact line wear analysis. The method comprises the following steps: collecting left and right profile point cloud data, forming profile point cloud sequences connected along the line, and carrying out pretreatment; generating continuous and consistent profile point cloud sequences by preliminary alignment and accurate alignment of adjacent section point clouds, obtaining a complete contact net three-dimensional profile model, and completing three-dimensional splicing; cutting profile sections at fixed intervals on the contact net three-dimensional profile model, and identifying wear surfaces; fitting the profile point clouds of the wear surfaces, calculating wear parameters, realizing wear quantitative analysis, and forming wear change curves along the line direction; and generating three-dimensional profile and wear data. The application meets the engineering application requirements of the rail transit industry on contact net intelligent operation and maintenance, wear prediction and life evaluation, realizes dynamic reconstruction of the real form of the contact line, and high-precision calculation of wear parameters.

Description

Technical Field

[0001] This invention belongs to the field of contact line wear analysis, and particularly relates to a calculation method based on three-dimensional splicing of contact line profiles and profile wear. Background Technology

[0002] With the rapid development of urban rail transit and the continuous increase in operating mileage, the requirements for the reliability and life management of the overhead contact system have significantly increased. The contact wire is a key component for traction power supply of subway trains, and its wear pattern directly affects the current collection stability of the pantograph, the power transmission efficiency, and the safety of line operation. Therefore, accurately obtaining the wear profile of the contact wire and conducting full-line wear analysis is an important foundation for intelligent operation and maintenance and life prediction of the overhead contact system.

[0003] Current wear assessment methods mostly employ two-dimensional cross-sectional models or empirical estimation methods, relying on local measurement points or discrete sampling. These methods cannot reflect the evolution of the three-dimensional continuous profile of the contact wire along the track direction, resulting in limited accuracy of wear results. Furthermore, traditional point cloud registration and modeling techniques suffer from significant error accumulation in complex structures, noise interference, and wear areas, easily leading to splicing misalignment and geometric distortion, making it difficult to achieve engineering-grade accuracy requirements.

[0004] In terms of wear identification, existing methods are often based on threshold segmentation or local contour fitting, which can only identify local wear features and cannot accurately calculate key indicators such as wear surface, wear area, and wear chord length. They lack robustness and efficiency in handling complex wear patterns and large-scale data processing, making it difficult to support the needs of full-line digital maintenance and life management of the overhead contact system.

[0005] Existing technologies mainly rely on two-dimensional cross-sections or local point measurements for wear analysis, which cannot accurately characterize the three-dimensional morphology of the contact wire in real space and its continuous changes along the line. Traditional point cloud registration methods are easily affected by wear areas, noise points, and installation deviations, leading to accumulated splicing errors and geometric distortion, which cannot meet engineering-level accuracy requirements. Wear identification and quantitative calculation often rely on simple thresholds or local fitting, resulting in low identification accuracy, weak noise resistance, and inability to cover complex wear patterns, making it difficult to support the life management needs of the entire contact network. Summary of the Invention

[0006] To address the aforementioned shortcomings in existing technologies, this invention provides a method for calculating contact line profile three-dimensional stitching and profile wear, which solves the problems of insufficient accuracy in three-dimensional profile reconstruction, severe accumulation of point cloud registration errors, poor robustness in wear identification, and difficulty in accurately calculating continuous wear parameters across the entire line in existing contact line wear analysis technologies.

[0007] To achieve the above objectives, the technical solution adopted by this invention is: a calculation method based on three-dimensional splicing of contact line profiles and profile wear, comprising the following steps: S1. At fixed travel intervals, sample left and right profile point cloud data to form a profile point cloud sequence connected along the line, and preprocess the profile point cloud. S2. Extract local and global features of the contour point cloud; S3. Based on the preprocessing results and the extracted local and global features, a continuous and consistent profile point cloud sequence is generated by performing preliminary and precise alignment on adjacent cross-sectional point clouds. Based on the continuous and consistent profile point cloud sequence, a complete three-dimensional profile model of the contact network is obtained, and the three-dimensional stitching is completed. S4. Cut the profile section at fixed intervals on the three-dimensional profile model of the contact wire, and identify the wear surface by automatically separating the contact wire and the grid area. S5. Fit the point cloud of the wear surface profile, calculate the wear parameters, realize the quantitative analysis of wear, and form the wear change curve along the line direction. S6. Based on the wear change curve, generate three-dimensional profile and wear data, and complete the calculation of profile wear.

[0008] Furthermore, the initial alignment includes the following steps: Input the preprocessed contour point cloud and the extracted local and global features. By setting the feature threshold of the bus, the points in the contour point cloud that meet the feature threshold are marked as anchor regions, and those that do not meet the feature threshold are marked as interference regions. A bus feature mask is generated to lock the alignment region and remove interference points. Based on global features, global similarity constraints between cross sections are constructed by characterizing the overall geometric shape, spatial orientation, and continuity features along the line direction of adjacent profile cross sections. Based on the locking and elimination results, the feature salience score S is calculated by using the bus assignment probability and the local feature magnitude, and the feature salience score S is weighted and corrected based on the global similarity constraint. With a fixed number of clusters K, points whose weighted feature significance score S is greater than or equal to a preset first threshold are preferentially selected as initial cluster centers, and the highest feature significance score S is clustered using the following formula: d = (1-S) i )×D; Where D represents the Euclidean distance, d represents the weighted clustering distance under the feature saliency constraint, and Si represents the feature saliency score corresponding to the i-th profile point; Based on the focusing results, for clusters with intra-cluster feature saliency scores S greater than or equal to the second threshold, spatial connectivity, and reaching the required number of iterations, dynamic anchor clusters are generated, where each cluster is a bus feature-geometric composite unit. Based on dynamic anchor clusters, topological descriptors are extracted using fixed topological features of the busbar to establish ordered matching relationships between clusters. Based on the ordered matching relationship between clusters, the initial transformation matrix is ​​solved to complete the initial alignment.

[0009] Furthermore, the precise alignment includes the following steps: Using the busbar feature vector of profile A as the standard, the objective function is to minimize the deviation between the busbar feature vector of adjacent profile B after matrix transformation and that of profile A.

[0010] Where T represents the rigid body transformation matrix used for registration. , and All represent weights. and Both represent the slope of the busbar edge of profile A. Indicate the coordinates of the intersection point of profile A. and Both represent the busbar characteristics of profile B after matrix transformation; Use the initial transformation matrix as the initial iteration value; By adjusting the rotation angle and translation components, the objective function value is gradually reduced. When the difference between the objective functions of two iterations is less than the preset value, or when the number of iterations is reached, the iteration stops, the final transformation matrix is ​​output, and precise alignment is achieved.

[0011] Furthermore, the three-dimensional stitching includes the following steps: All two-dimensional profile point clouds are projected onto a unified coordinate system through a final transformation matrix after precise alignment to generate a continuous and consistent sequence of profile point clouds. Based on a continuous and consistent sequence of contour point clouds, a complete three-dimensional model of the overhead contact line is obtained using the Poisson surface reconstruction method.

[0012] Furthermore, step S4 includes the following steps: S401. Cut out profile sections at fixed intervals on the three-dimensional profile model of the overhead contact line; S402. Based on the cut profile section, calculate the curvature value of each point on the profile line to separate the contact line. Points with curvature values ​​greater than the third threshold are classified as contact lines, points with curvature values ​​less than the third threshold but within the fitted standard circle are classified as contact lines, and points with curvature values ​​less than the third threshold and outside the standard circle are classified as busbars and discarded. S403. Based on the separation results, calculate the distance between the points on the contact line profile and the fitted standard circle. If the distance between the points is greater than the threshold T and is within the standard circle, then it is identified as a wear surface.

[0013] Furthermore, step S5 includes the following steps: S501, Fit the profile point cloud located on the wear surface into a line; S502. Based on the fitting results, find the two intersection points between the wear surface and the standard fitting circle to obtain the wear chord length; S503. Based on the wear chord length, calculate the wear parameters, realize the quantitative analysis of wear, and form a wear change curve along the line direction.

[0014] The beneficial effects of this invention are: This invention proposes a method for calculating the wear of contact wire profiles based on 3D splicing and wear analysis. It involves collecting point cloud data of multiple contact wire profiles, normalizing and denoising the data, and then inputting it into a deep learning network to extract features. A feature-driven point cloud registration algorithm is used for initial alignment, and high-precision registration is achieved through iterative optimization. The multiple point cloud segments are projected onto a unified coordinate system, and a surface reconstruction method is used to obtain a complete 3D contact wire model. Profile sections are then cut at fixed intervals on the reconstructed model. Wear areas are identified using a dual-constraint method of curvature and a reference circle. The wear point set is fitted, and key wear parameters such as wear chord length, wear area, and residual height are calculated. This achieves automated, continuous, and high-precision quantitative analysis of the contact wire wear state. This invention has at least the following beneficial effects: Achieve high-precision 3D stitching of contact line profile point clouds, eliminate wear interference and installation deviations through deep learning feature extraction and dynamic anchor point registration, and realize unified coordinate system reconstruction of the entire contact line; To improve the accuracy and stability of point cloud registration, error accumulation is suppressed through local-global feature fusion and iterative refinement mechanisms to obtain a continuous and complete three-dimensional contact wire model; Improve the accuracy and robustness of wear identification and quantitative calculation, and realize the automatic extraction of wear areas and the accurate calculation of parameters such as wear area, chord length, and residual height; It provides reliable data support for overhead contact line maintenance, wear trend analysis and life prediction, and supports digital operation and maintenance and intelligent decision-making of overhead contact lines. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0017] Example like Figure 1 As shown, this invention provides a calculation method based on three-dimensional splicing of contact line profiles and profile wear, the implementation method of which is as follows: S1. At fixed travel intervals, sample left and right profile point cloud data to form a profile point cloud sequence connected along the line, and preprocess the profile point cloud. S2. Extract local and global features of the contour point cloud; S3. Based on the preprocessing results and the extracted local and global features, a continuous and consistent profile point cloud sequence is generated by performing preliminary and precise alignment on adjacent cross-sectional point clouds. Based on the continuous and consistent profile point cloud sequence, a complete three-dimensional profile model of the contact network is obtained, and the three-dimensional stitching is completed. The initial alignment includes the following steps: inputting the preprocessed contour point cloud and the extracted local and global features; by setting the feature threshold of the bus, marking the points in the contour point cloud that meet the feature threshold as anchor regions and those that do not meet the feature threshold as interference regions; generating a bus feature mask to lock the alignment region and remove interference points. Based on global features, global similarity constraints between cross sections are constructed by characterizing the overall geometric shape, spatial orientation, and continuity features along the line direction of adjacent profile cross sections. Based on the locking and elimination results, the feature salience score S is calculated by using the bus assignment probability and the local feature magnitude, and the feature salience score S is weighted and corrected based on the global similarity constraint. With a fixed number of clusters K, points whose weighted feature significance score S is greater than or equal to a preset first threshold are preferentially selected as initial cluster centers, and the highest feature significance score S is clustered using the following formula: d = (1-S) i )×D; Where D represents the Euclidean distance, d represents the weighted clustering distance under the feature saliency constraint, which is used to measure the comprehensive similarity between point i and the current cluster center; Si represents the feature saliency score corresponding to the i-th profile point, which is used to characterize the reliability of the point belonging to the busbar anchoring region; Based on the focusing results, for clusters with intra-cluster feature saliency scores S greater than or equal to the second threshold, spatial connectivity, and reaching the required number of iterations, dynamic anchor clusters are generated, where each cluster is a bus feature-geometric composite unit. Based on dynamic anchor clusters, topological descriptors are extracted using fixed topological features of the busbar to establish ordered matching relationships between clusters. Based on the ordered matching relationship between clusters, the initial transformation matrix is ​​solved to complete the preliminary alignment. In this step, the global features are mainly used to constrain the overall consistency and adjust the weights of the preliminary alignment results. They do not directly participate in point-level matching, but rather construct global similarity constraints between sections by characterizing the overall geometric shape, spatial orientation, and continuity features along the line direction of adjacent profile sections. Based on this global similarity, the feature saliency score S obtained from the busbar belonging probability and the local feature magnitude is weighted and corrected to suppress the influence of local feature anomalies or mismatches on the alignment results, thereby improving the stability and overall consistency of the preliminary alignment of adjacent section profiles. Precise alignment includes the following steps: Using the busbar feature vector of profile A as the standard, the objective function is to minimize the deviation between the busbar feature vector of adjacent profile B after matrix transformation and that of profile A.

[0018] Where T represents the rigid body transformation matrix used for registration, which includes a rotation matrix and a translation vector, used to map the coordinates of profile B to the coordinate system of profile A; , and All represent weights. and Both represent the slope of the busbar edge of profile A. Indicate the coordinates of the intersection point of profile A. and Both represent the busbar characteristics of profile B after matrix transformation; Use the initial transformation matrix as the initial iteration value; By adjusting the rotation angle and translation components, the objective function value is gradually reduced. When the difference between the objective functions of two iterations is less than the preset value, or when the number of iterations is reached, the iteration stops, the final transformation matrix is ​​output, and the precise alignment is completed. The 3D stitching process includes the following steps: projecting all 2D profile point clouds onto a unified coordinate system using a final transformation matrix after precise alignment to generate a continuous and consistent profile point cloud sequence; and using the Poisson surface reconstruction method to obtain a complete 3D model of the contact network based on the continuous and consistent profile point cloud sequence. S4. Cut profile sections at fixed intervals on the three-dimensional profile model of the contact wire, and identify the wear surface by automatically separating the contact wire and the grid area. The implementation method is as follows: S401. Cut out profile sections at fixed intervals on the three-dimensional profile model of the overhead contact line; S402. Based on the cut profile section, calculate the curvature value of each point on the profile line to separate the contact line. Points with curvature values ​​greater than the third threshold are classified as contact lines, points with curvature values ​​less than the third threshold but within the fitted standard circle are classified as contact lines, and points with curvature values ​​less than the third threshold and outside the standard circle are classified as busbars and discarded. S403. Based on the separation results, calculate the distance between the points on the contact line profile and the fitted standard circle. If the distance between the points is greater than the threshold T and is inside the standard circle, they are identified as wear surfaces. In S402, "points with curvature values ​​less than the first threshold and outside the standard circle are classified as busbars and removed" are the grid area. Through this step, the grid area is removed and only the contact line profile is retained. S5. Fit the point cloud of the wear surface profile, calculate the wear parameters, realize the quantitative analysis of wear, and generate a wear variation curve along the track direction. The implementation method is as follows: S501, Fit the profile point cloud located on the wear surface into a line; S502. Based on the fitting results, find the two intersection points between the wear surface and the standard fitting circle to obtain the wear chord length; S503. Based on the wear chord length, calculate the wear parameters, realize the quantitative analysis of wear, and form a wear change curve along the line direction; S6. Based on the wear change curve, generate three-dimensional profile and wear data, and complete the calculation of profile wear.

[0019] In this embodiment, the calculation process and principle of the present invention are as follows: First, a line laser sensor scans along the target track section or contact line cross-section to collect left and right profile point cloud data, and sampling is triggered at fixed travel intervals to form a continuous profile point cloud sequence along the track direction. Simultaneously, the data is stored in real-time to the processing unit. Then, the collected profile point clouds are denoised, scaled, and the number of points is unified to provide a reliable foundation for subsequent registration and splicing. Next, the moving average and grey relational analysis methods are used to perform preliminary registration of adjacent cross-section point clouds. Then, high-precision fine registration is achieved through iterative optimization, generating a continuous and consistent profile point cloud sequence. Finally, a complete three-dimensional profile model of the contact network is obtained through a surface reconstruction algorithm, realistically restoring the contact line or rail cross-section. Morphology: A profile section is cut at fixed intervals on the generated 3D model of the contact network. A wear identification algorithm based on curvature thresholds and standard circle constraints automatically separates the contact line and the grid structure area, and identifies the wear surface. The algorithm effectively handles uneven local wear, measurement noise, and anomalies, ensuring separation accuracy and robustness. Subsequently, the profile point cloud of the wear surface is fitted, and key parameters such as wear chord length, wear area, and residual height are calculated through geometric intersection to achieve quantitative wear analysis. A wear variation curve can be generated along the line direction for equipment condition evaluation, life prediction, and maintenance decisions. The final generated 3D profile and wear data can be used for contact network maintenance optimization, structural offset analysis, and life prediction, achieving continuous and high-precision contact line wear monitoring and management. The specific 3D stitching and wear calculation methods are as follows: I. Three-dimensional splicing of the outline 1. Global registration (initial alignment) ① Extracting profile features using the improved Pointnet++ model (1) Model training Point cloud preprocessing: Collect sufficient profile point cloud data, perform scale normalization, noise reduction and other processing on the original profile point cloud, and unify the number of points cloud by FPS sampling or interpolation; Model Adjustment: The Set Abstraction module of Pointnet++ is retained in the improved Pointnet model to capture features from local to global through multi-level abstraction; Dataset customization: The contour point cloud is divided into training set, validation set and test set in a ratio of 7:2:1, and the data is augmented by rotating, scaling and other methods to avoid overfitting; Targeted training: The model is trained using the Adam optimizer and the mean squared error (MSE) function, with a batch size of 8-32 and an initial number of iterations of 100-300, combined with an early stopping mechanism to prevent overfitting. Evaluation and optimization: The model is evaluated using metrics such as root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). If the feature discrimination is low, we can try to modify the global feature dimension or introduce more fine-grained local features.

[0020] (2) Feature extraction The contour point cloud data to be reconstructed in 3D is input into the trained improved point net model Pointnet++ for processing, and the local and global features of the contour point cloud are extracted and output.

[0021] ② Perform preliminary alignment based on extracted profile features (1) Anchor cluster matching relationship is established by using the BF-DAR algorithm (dynamic anchor registration algorithm dominated by bus features) and the improved K-means weighted clustering algorithm. Input preparation: Input the preprocessed contour point cloud and features extracted by the improved Pointnet++ model; Generate bus feature mask: Set the feature threshold of the bus (edge ​​slope range, intersection coordinate deviation ≤ 0.01mm), mark the points in the point cloud that meet the feature threshold as anchoring regions, and mark the points that do not meet the threshold as interference regions, output the feature mask of the bus, and accurately lock the registration core region; Use the improved K-means weighted clustering algorithm to generate dynamic anchor clusters: 1) Calculate the feature significance score: Quantify the registration value of each point according to the formula S=0.6×busage probability p+0.4×local feature magnitude P; 2) Cluster initialization: Fix the number of clusters K=3~5, and preferentially select points with S≥0.8 as the initial cluster centers; 3) Weighted clustering: using the improved distance formula d = (1 - S) i Multiplying the Euclidean distance makes it easier for high-scoring points to cluster together; 4) Cluster constraints and termination: The mean S within the cluster must be ≥0.7 and the space must be continuous. Iteration stops when the distance moved from the cluster center is ≤0.001mm or after 20 iterations. Topology matching: For each dynamic anchor cluster, extract the fixed topological features of the busbar, including the slope distribution of points within the cluster, the coordinates of intersection points, and the distance between adjacent clusters. Construct a 128-dimensional topological descriptor, calculate the similarity of the topological descriptors of anchor clusters in adjacent profiles (cosine similarity ≥ 0.9), and establish matching relationships according to the linear topological order of the busbar. (2) Initial alignment is achieved by solving the initial transformation matrix using the singular value decomposition (SVD) algorithm.

[0022] Advantages: The Pointnet++ improved model uses hierarchical feature extraction to adapt to the disorder of point clouds, capturing both details and global structure with high feature recognition; the BF-DAR dynamic anchoring and registration algorithm uses the bus as its core, with feature masks accurately eliminating interference such as contact line wear and noise, dynamic anchor clusters adapt to bus installation deviations, and topological descriptors effectively ensure orderly and unambiguous matching; the Singular Value Decomposition (SVD) algorithm has a fast solution speed and strong stability, with initial alignment errors controlled at a low level, laying a solid foundation for subsequent fine-tuning.

[0023] 2. Registration refinement (precise alignment) The Levenberg-Marquardt algorithm (LM) is used to refine the transformation matrix for accurate alignment.

[0024] Using the busbar feature vector of profile A as the standard, the objective function is to minimize the deviation between the transformed busbar feature vector of adjacent profile B and A. ; Where: T represents the rigid body transformation matrix used for registration, which includes a rotation matrix and a translation vector, used to map the coordinates of profile B to the coordinate system of profile A; , and All represent weights. and Both represent the slope of the busbar edge of profile A. Indicate the coordinates of the intersection point of profile A. and Both represent the busbar characteristics of profile B after matrix transformation. Standard value; Values ​​to be optimized; Equal weights are applied because each feature of the busbar is equally important for registration.

[0025] The initial matrix output by the Singular Value Decomposition (SVD) algorithm is used as the initial iteration value of the Levenberg-Marquardt algorithm (LM). The objective function value is gradually reduced by adjusting the rotation angle and translation components. When the difference between the objective functions of two iterations is less than 0.001 mm, or when the number of iterations reaches 50, the convergence condition is met, and the iteration is stopped. The final transformation matrix is ​​output to complete the precise alignment.

[0026] Advantages: The initial values ​​are derived from reliable solutions of the Singular Value Decomposition (SVD) algorithm, avoiding the Levenberg-Marquardt algorithm's (LM) getting stuck in local optima, resulting in a more accurate convergence direction; the bus feature constraint objective function aligns with the core requirements of engineering registration, focusing on key benchmark features; the LM algorithm has fast convergence speed and high accuracy, fully meeting the accuracy requirements of 3D stitching. 3. Three-dimensional stitching All two-dimensional contour point clouds are projected onto a unified coordinate system using a precisely aligned transformation matrix, and a complete three-dimensional model of the contact network is obtained through the Poisson surface reconstruction method.

[0027] Advantages: Poisson surface reconstruction automatically fills in the gaps in the point cloud, and the generated model is continuous and crack-free with a smooth surface. It can accurately preserve the contact line wear details and busbar straight line features without information loss due to excessive smoothing. Compared with traditional point cloud models, smooth surfaces reduce noise interference in subsequent section extraction and improve the reliability of wear calculation.

[0028] II. Contact Line Wear Calculation 1. Wear and tear identification The stitched 3D model was processed using the CWWI contact wire wear identification algorithm: Cross-section cutting: Input the spliced ​​3D model and cut cross-sections at fixed intervals along the line direction; Dual-constraint separation of wear surface and busbar: 1) First constraint: Calculate the curvature value of each point on the cross-sectional profile line. Points with curvature values ​​> the third threshold are classified as contact line areas, while points with curvature values ​​≤ the third threshold are subject to the second constraint for further determination. 2) Second constraint: Fit a standard circle of the contact line based on the parameters of the unworn contact line. Points with a first constraint curvature value ≤ the third threshold but still within the standard circle area are classified as the contact line area. Points with a first constraint curvature value ≤ the third threshold and outside the standard circle area are classified as the busbar and removed. Identify the wear surface: Calculate the shortest distance d from each point on the contact line to the standard fitted circle, set a threshold T, and mark the points where d > T and are within the standard circle as points on the wear surface.

[0029] Advantages: High separation accuracy due to dual constraints of curvature threshold and standard circle range; wear area determination is based on distance threshold as the quantitative standard, eliminating subjective errors and ensuring good consistency of results; the algorithm has low computational load and fast processing speed, adapting to the high-efficiency processing needs of batch cross-sections, and has outstanding anti-interference ability.

[0030] 2. Wear Calculation The Random Sampling Consensus Algorithm (RANSAC) is used to fit the point cloud of the wear surface into a line, find the two intersection points of the wear surface and the standard fitted circle, obtain the wear chord length, and then calculate parameters such as wear area and residual height.

[0031] Advantages: The Random Sampling Consensus Algorithm (RANSAC) has strong noise and outlier resistance capabilities, can accurately fit the wear surface, and avoid interference points affecting parameter accuracy; parameters such as wear chord length, area, and residual height are derived based on geometric principles, with sufficient theoretical support and accurate calculation results.

[0032] In summary, this invention extracts local and global structural features of point clouds using a deep learning model, utilizes stable busbar geometric features as a registration benchmark, and reduces the accumulation of registration errors through optimized solutions to achieve three-dimensional continuous reconstruction of the true spatial morphology of the contact line. Wear identification is based on the geometric features that cause the cross-section to deviate from the ideal circular contour due to wear of the contact line. Curvature changes and the distance from the point to the reference circle are used as the judgment criteria. Key geometric quantities of wear are obtained through fitting and intersection methods, thereby completing the quantitative measurement of wear based on mathematical and geometric principles and accurately reflecting the true degree of wear of the contact line.

[0033] Compared with existing contact wire wear analysis methods that rely on simplified two-dimensional models or local point cloud stitching, this invention achieves significant improvements in accuracy and efficiency through three-dimensional high-precision profile stitching and automated wear calculation. First, based on deep learning network feature extraction combined with a bus-dominated dynamic anchoring registration algorithm, high-precision continuous stitching of contact wire point clouds is achieved, significantly reducing three-dimensional morphological distortion caused by registration errors in traditional two-dimensional or local stitching methods, thereby improving the accuracy and reliability of wear calculation. Second, by employing a wear identification algorithm based on curvature thresholds and standard circle constraints, and a method for calculating wear chord length, area, and residual height using geometric fitting, wear data can be automatically and quantitatively acquired, effectively avoiding manual measurement errors and repeated on-site inspections, thus reducing maintenance costs. Simultaneously, this invention can generate continuous wear variation curves along the line direction, providing a scientific basis for equipment condition assessment, life prediction, and maintenance decisions, improving maintenance efficiency and extending equipment lifespan. Overall, this invention achieves high precision, automation, and continuous monitoring of contact wire wear through the synergistic effect of key technical solutions, providing data support for the life management of the contact network system, and possessing both engineering application value and economic benefits.

Claims

1. A calculation method based on three-dimensional splicing of contact line profiles and profile wear, characterized in that, Includes the following steps: S1. At fixed travel intervals, sample left and right profile point cloud data to form a profile point cloud sequence connected along the line, and preprocess the profile point cloud. S2. Extract local and global features of the contour point cloud; S3. Based on the preprocessing results and the extracted local and global features, a continuous and consistent profile point cloud sequence is generated by performing preliminary and precise alignment on adjacent cross-sectional point clouds. Based on the continuous and consistent profile point cloud sequence, a complete three-dimensional profile model of the contact network is obtained, and the three-dimensional stitching is completed. The initial alignment includes the following steps: Input the preprocessed contour point cloud and the extracted local and global features. By setting the feature threshold of the bus, the points in the contour point cloud that meet the feature threshold are marked as anchor regions, and those that do not meet the feature threshold are marked as interference regions. A bus feature mask is generated to lock the alignment region and remove interference points. Based on global features, global similarity constraints between cross sections are constructed by characterizing the overall geometric shape, spatial orientation, and continuity features along the line direction of adjacent profile cross sections. Based on the locking and elimination results, the feature salience score S is calculated by using the bus assignment probability and the local feature magnitude, and the feature salience score S is weighted and corrected based on the global similarity constraint. With a fixed number of clusters K, points whose weighted feature significance score S is greater than or equal to a preset first threshold are preferentially selected as initial cluster centers, and the highest feature significance score S is clustered using the following formula: d=(1-S i )×D; Where D represents the Euclidean distance, d represents the weighted clustering distance under the feature saliency constraint, and Si represents the feature saliency score corresponding to the i-th profile point; Based on the focusing results, for clusters with intra-cluster feature saliency scores S greater than or equal to the second threshold, spatial connectivity, and reaching the required number of iterations, dynamic anchor clusters are generated, where each cluster is a bus feature-geometric composite unit. Based on dynamic anchor clusters, topological descriptors are extracted using fixed topological features of the busbar to establish ordered matching relationships between clusters. Based on the ordered matching relationship between clusters, the initial transformation matrix is ​​solved to complete the initial alignment; S4. Cut the profile section at fixed intervals on the three-dimensional profile model of the contact wire, and identify the wear surface by automatically separating the contact wire and the grid area. S5. Fit the point cloud of the wear surface profile, calculate the wear parameters, realize the quantitative analysis of wear, and form the wear change curve along the line direction. S6. Based on the wear change curve, generate a three-dimensional profile and wear data, and complete the calculation of profile wear.

2. The calculation method based on three-dimensional splicing of contact line profile and profile wear according to claim 1, characterized in that, The precise alignment includes the following steps: Using the busbar feature vector of profile A as the standard, the objective function is to minimize the deviation between the busbar feature vector of adjacent profile B after matrix transformation and that of profile A. Where T represents the rigid body transformation matrix used for registration. , and All represent weights. and Both represent the slope of the busbar edge of profile A. Indicate the coordinates of the intersection point of profile A. and Both represent the busbar characteristics of profile B after matrix transformation; Use the initial transformation matrix as the initial iteration value; By adjusting the rotation angle and translation components, the objective function value is gradually reduced. When the difference between the objective functions of two iterations is less than the preset value, or when the number of iterations is reached, the iteration stops, the final transformation matrix is ​​output, and precise alignment is achieved.

3. The calculation method based on three-dimensional splicing of contact line profile and profile wear according to claim 2, characterized in that, The three-dimensional stitching includes the following steps: All two-dimensional profile point clouds are projected onto a unified coordinate system through a final transformation matrix after precise alignment to generate a continuous and consistent sequence of profile point clouds. Based on a continuous and consistent sequence of contour point clouds, a complete three-dimensional model of the overhead contact line is obtained using the Poisson surface reconstruction method.

4. The calculation method based on three-dimensional splicing of contact line profile and profile wear according to claim 1, characterized in that, S4 includes the following steps: S401. Cut out profile sections at fixed intervals on the three-dimensional profile model of the overhead contact line; S402. Based on the cut profile section, calculate the curvature value of each point on the profile line to separate the contact line. Points with curvature values ​​greater than the third threshold are classified as contact lines, points with curvature values ​​less than the third threshold but within the fitted standard circle are classified as contact lines, and points with curvature values ​​less than the third threshold and outside the standard circle are classified as busbars and discarded. S403. Based on the separation results, calculate the distance between the points on the contact line profile and the fitted standard circle. If the distance between the points is greater than the threshold T and is within the standard circle, then it is identified as a wear surface.

5. The calculation method based on three-dimensional splicing of contact line profile and profile wear according to claim 1, characterized in that, S5 includes the following steps: S501, Fit the profile point cloud located on the wear surface into a line; S502. Based on the fitting results, find the two intersection points between the wear surface and the standard fitting circle to obtain the wear chord length; S503. Based on the wear chord length, calculate the wear parameters, realize the quantitative analysis of wear, and form a wear change curve along the line direction.

Images