Intelligent pantograph-catenary wear detection method and system based on machine vision

Abstract

The invention relates to the related technical field of computer vision analysis, in particular to an intelligent pantograph-catenary wear detection method and system based on machine vision, and the method comprises the steps: obtaining the state and image information of a pantograph in a pantograph-catenary contact area, inverting a fatigue crack propagation path under the control parameters of a pantograph lifting mechanism, extracting spatial-temporal characteristics through three-dimensional convolution, and obtaining a pantograph-catenary wear detection result. According to the method, environment compensation and image information are combined with multi-scale features, a dynamic wear feature tensor is generated for wear offset control, and the technical problems that a fatigue crack propagation path is interfered and influenced by various factors, and the wear detection precision of a pantograph and a contact network is insufficient are solved. The technical effects that the fatigue crack propagation paths of the pantograph rising mechanism and the pantograph falling mechanism are inversed, the abrasion condition of the pantograph net is comprehensively evaluated, remote abrasion real-time detection and maintenance decision making are conducted on the pantograph and the contact net, potential abrasion hidden dangers are found and processed in time, accidents are avoided, and the detection precision and efficiency are improved are achieved.

Description

Technical Field

[0001] This invention relates to the field of computer vision analysis technology, specifically to a machine vision-based intelligent detection method and system for pantograph-catenary wear. Background Technology

[0002] The pantograph-catenary system is used to transmit electrical energy from the traction network to electric locomotives. Due to the harsh working environment and the frequent exposure to strong electrical and mechanical forces, it is prone to various defects. Among these, wear not only affects the performance of the pantograph and the contact network, but can also lead to power supply failures, affecting the normal operation of electric locomotives.

[0003] With the rapid development of computer technology and machine vision, inspection is gradually moving towards online and real-time inspection. By using image equipment to remotely monitor the pantograph and connect it to the subway maintenance center, subway maintenance personnel can detect pantograph faults in a timely and accurate manner. However, in practical applications, there are still problems such as poor adaptability to environmental factors and insufficient ability to identify complex defects.

[0004] In summary, existing technologies suffer from technical problems such as fatigue crack propagation paths being affected by various factors and insufficient accuracy in pantograph and contact wire wear detection. Summary of the Invention

[0005] This application provides a machine vision-based intelligent pantograph-catenary wear detection system, aiming to solve the technical problem in the prior art where the propagation path of fatigue cracks is affected by various factors, resulting in insufficient accuracy in pantograph and catenary wear detection.

[0006] In view of the above problems, the technical solution to achieve the present application is as follows: This application provides a machine vision-based intelligent detection method for pantograph-catenary wear, comprising: acquiring pantograph operating status and surface image information in the pantograph-catenary contact area between the contact wire and the sliding plate; performing finite element simulation inversion on the control parameters of the pantograph raising mechanism corresponding to the upper frame and lower boom, and the control parameters of the pantograph lowering mechanism corresponding to the upper frame and lower boom, respectively, to obtain a first set of fatigue crack propagation paths and a second set of fatigue crack propagation paths under multiple sets of control parameter combinations; performing three-dimensional convolution processing on the first and second sets of fatigue crack propagation paths to extract spatiotemporal correlation features; adding environmental compensation data to correct perception bias based on the pantograph operating status and surface image information, and performing multi-scale feature fusion in combination with the first and second sets of fatigue crack propagation paths to generate a dynamic wear feature tensor under spatiotemporal correlation features; and performing wear offset control based on the dynamic wear feature tensor in combination with the control parameters of the pantograph raising mechanism and the pantograph lowering mechanism.

[0007] In another aspect, this application provides a machine vision-based intelligent pantograph-catenary wear detection system, wherein the system includes: an information acquisition module for acquiring pantograph operating status and surface image information in the pantograph-catenary contact area between the contact wire and the sliding plate; a simulation module for performing finite element simulation inversion on the control parameters of the pantograph raising mechanism corresponding to the upper frame and the lower boom, and the control parameters of the pantograph lowering mechanism corresponding to the upper frame and the lower boom, respectively, to obtain a first fatigue crack propagation path set and a second fatigue crack propagation path set under multiple sets of control parameter combinations; a convolution processing module for performing three-dimensional convolution processing on the first fatigue crack propagation path set and the second fatigue crack propagation path set to extract spatiotemporal correlation features; a feature fusion module for adding environmental compensation data to correct perception bias based on the pantograph operating status and surface image information, and performing multi-scale feature fusion in combination with the first fatigue crack propagation path set and the second fatigue crack propagation path set to generate a dynamic wear feature tensor under spatiotemporal correlation features; and an offset control module for performing wear offset control based on the dynamic wear feature tensor and in combination with the control parameters of the pantograph raising mechanism and the control parameters of the pantograph lowering mechanism.

[0008] In summary, one or more technical solutions provided in this application achieve the technical effects of inverting the fatigue crack propagation path of the pantograph raising and lowering mechanisms, comprehensively assessing the wear of the pantograph and catenary, enabling real-time remote wear detection and maintenance decisions for the pantograph and catenary, timely identifying and addressing potential wear hazards, preventing accidents, and improving detection accuracy and efficiency. Attached Figure Description

[0009] Figure 1 A flowchart illustrating the intelligent detection method for pantograph-catenary wear based on machine vision is provided for this application; Figure 2 This application provides a schematic diagram of the structure of a machine vision-based intelligent detection system for pantograph-catenary wear.

[0010] Explanation of reference numerals in the attached figures: Information acquisition module M100, simulation module M200, convolution processing module M300, feature fusion module M400, offset control module M500. Detailed Implementation Example

[0011] The present application will now be described in detail with reference to the accompanying drawings, such as... Figure 1 As shown, this application provides a machine vision-based intelligent detection method for bow and cable receptacle wear, wherein the method includes: S1: Obtain pantograph working status and surface image information in the pantograph contact area between the contact wire and the sliding plate; S2: Perform finite element simulation inversion on the control parameters of the pantograph raising mechanism corresponding to the upper frame and the lower boom, and the control parameters of the pantograph lowering mechanism corresponding to the upper frame and the lower boom, respectively, to obtain the first fatigue crack propagation path set and the second fatigue crack propagation path set under multiple sets of control parameter combinations.

[0012] Specifically, acquiring pantograph operating status and surface image information in the pantograph-catenary contact area refers to using a high-resolution camera and imaging system to capture real-time images of the pantograph-catenary's operating status and surface. Image processing algorithms are then used to preprocess the acquired images, including noise reduction, contrast enhancement, and image segmentation, to highlight wear areas and the degree of wear. Finite element simulation inversion is a numerical simulation method based on finite element analysis. By discretizing the actual structure, a finite element model is established, and appropriate loads and boundary conditions are applied to solve the structure's response, thereby obtaining the mechanical properties and damage evolution of the structure under different working conditions. Fatigue crack propagation path refers to the trajectory of the gradual propagation of internal cracks in a material under cyclic loading. By analyzing the crack propagation path, the fatigue life and failure mode of the material can be predicted.

[0013] In the pantograph-catenary contact area between the overhead contact wire and the sliding plate, a high-resolution camera and imaging system are installed to acquire real-time information on the pantograph's working status and surface images. Based on the pantograph's structure and working principle, control parameters for the raising and lowering mechanisms are determined, including air pressure, raising speed, and lowering angle. Based on these control parameters, finite element simulations are performed, setting different combinations of control parameters to simulate the pantograph's stress and crack propagation process under various working conditions. Using finite element analysis software (such as ABAQUS and MARC), a finite element model of the pantograph is established, and corresponding loads and boundary conditions are applied to solve for the first and second fatigue crack propagation path sets under multiple combinations of control parameters.

[0014] By acquiring high-resolution images of the pantograph, a rich data foundation is provided for subsequent wear feature extraction and analysis. The images contain information such as wear marks and cracks on the pantograph surface, which intuitively reflect the wear condition of the pantograph. By using finite element simulation inversion technology, the fatigue crack propagation path of the pantograph raising and lowering mechanism can be predicted, which can help to understand the fatigue damage of the pantograph under different working conditions in advance and provide support for targeted optimization.

[0015] S3: Perform three-dimensional convolution processing on the first fatigue crack propagation path set and the second fatigue crack propagation path set to extract spatiotemporal correlation features; S4: Based on the pantograph's working state and surface image information, add environmental compensation data to correct perception bias, and combine the first fatigue crack propagation path set and the second fatigue crack propagation path set to perform multi-scale feature fusion to generate a dynamic wear feature tensor under spatiotemporal correlation features; S5: Based on the dynamic wear feature tensor, combine the control parameters of the pantograph raising mechanism and the control parameters of the pantograph lowering mechanism to perform wear offset control.

[0016] Specifically, the three-dimensional convolution processing of the first and second fatigue crack propagation path sets refers to using a three-dimensional convolutional neural network (3D CNN) to extract and analyze features from these path sets. Three-dimensional convolution can perform convolution operations simultaneously in spatial and temporal dimensions, capturing motion information and spatial features between consecutive frames, thereby extracting the spatiotemporal correlation features of crack propagation. Adding environmental compensation data to correct perception bias refers to compensating the acquired image information by introducing environmental parameters (such as temperature, humidity, and illumination) to reduce the impact of environmental factors on the detection results. Multi-scale feature fusion refers to integrating features of different scales and levels to more comprehensively describe the characteristics of crack propagation. Generating dynamic wear feature tensors means representing the fused features in tensor form for easier subsequent analysis and processing. Wear offset control involves actively adjusting the control parameters of the raising and lowering mechanisms to keep the wear position and degree of key components of the bow and catenary under control, avoiding excessive local wear.

[0017] Three-dimensional convolution processing is performed on the first and second fatigue crack propagation path sets, treating these path sets as continuous three-dimensional data blocks. A three-dimensional convolution kernel is used to perform convolution operations in both spatial and temporal dimensions to extract spatiotemporal correlation features such as crack propagation gradient features and multi-frequency band features. Based on the pantograph's working state and surface image information, environmental compensation data is added to correct perception biases. Environmental parameters, such as temperature, humidity, and light intensity, are collected, and an environmental compensation model is established to adjust image information and reduce detection errors caused by environmental factors. Specifically, in outdoor detection systems, changes in lighting affect image quality; environmental compensation can improve image stability and reliability.

[0018] Multi-scale feature fusion is performed by combining the first and second fatigue crack propagation path sets to integrate crack propagation features at different scales, forming a more comprehensive feature description. Specifically, the concat operation in the early fusion method can concatenate different features, increasing the feature dimensionality. A dynamic wear feature tensor under spatiotemporal correlation features is generated, and the fused features are represented in tensor form. The tensor dimension can include time, space, crack depth, etc., which facilitates subsequent analysis and processing. By analyzing the wear features in the tensor and combining the current control parameters of the raising and lowering mechanisms, the future wear offset trend of the pantograph-catenary system is predicted, thereby changing the pantograph-catenary contact force distribution and suppressing wear offset.

[0019] Preferably, by using three-dimensional convolution processing and spatiotemporal correlation feature extraction, the dynamic process and spatiotemporal variation characteristics of crack propagation can be captured, providing a more accurate basis for wear assessment. On the one hand, adding environmental compensation data corrects perception bias, improving the stability and reliability of detection results, enabling the system to better adapt to complex and ever-changing actual environments. Multi-scale feature fusion further integrates feature information at different scales, making the description of wear characteristics more comprehensive and detailed. The generated dynamic wear feature tensor provides rich data support for subsequent maintenance decisions. On the other hand, the configured pantograph-catenary maintenance control instruction set directly transforms the detection results into specific maintenance action guidance, realizing precise maintenance of the pantograph and catenary, improving the efficiency and pertinence of maintenance work, and ensuring the safety and reliability of rail transit operation.

[0020] Furthermore, based on the dynamic wear characteristic tensor, and combined with the control parameters of the raising and lowering mechanisms, wear offset control is performed. The method of this application includes: Based on the dynamic wear feature tensor, the wear anomaly region and wear depth evolution index are determined; according to the wear anomaly region and wear depth evolution index, the feasible region of the control parameters of the raising mechanism and the lowering mechanism is updated, and the feasible region is mapped to the parameter combination that minimizes the wear offset obtained by the search.

[0021] Specifically, the abnormal wear region refers to the area where the wear degree exceeds the normal range by analyzing the data in the dynamic wear feature tensor; the wear depth evolution index is obtained by quantifying the change trend of wear depth over time in the abnormal wear region; the feasible region refers to the set of reasonable ranges of values ​​for the control parameters of the raising mechanism and the lowering mechanism. The parameter combination within this range is considered to meet the system operation requirements and conform to the wear deviation control target.

[0022] The dynamic wear feature tensor is analyzed, and areas with abnormal wear are identified by setting a wear threshold or using clustering algorithms. For example, in tool wear detection, wear points are determined by using a 3D model. The wear depth evolution index is calculated, which can be the rate of change of wear depth over time or other indicators that can reflect the wear development trend. For example, the wear coefficient is estimated by using the Archard model, and the wear depth is calculated based on the wear volume and slip.

[0023] Based on the location, size, and wear depth evolution index of the abnormal wear area, combined with finite element simulation data and historical experience data, the feasible domains of the control parameters for the pantograph raising and lowering mechanisms are redefined. Specifically, if a certain abnormal wear area shows excessively rapid local wear, the range of values ​​for the pantograph raising pressure parameter is narrowed to avoid exacerbating wear due to excessive pressure. Based on the wear state information obtained from previous detection and analysis, the scope for subsequent optimization search of control parameters is defined, effectively reducing the blindness and computational load of parameter search. Reasonably defining the feasible domain improves the efficiency of control parameter optimization, enabling faster and more accurate identification of the control parameter combination that minimizes wear offset, thereby achieving effective control of pantograph-catenary wear offset, extending the service life of the pantograph-catenary equipment, and ensuring the stable operation of the rail transit power supply system.

[0024] By identifying abnormal wear areas and wear depth evolution indices.

[0025] Furthermore, the method of this application also includes: making wear maintenance decisions for the pantograph and overhead contact line. An iterative search is performed within the updated feasible region, with the minimization of wear offset as the reward function; during the iterative search process, the smoothness of the control parameters of the raising mechanism and the smoothness of the control parameters of the lowering mechanism are used as penalty terms.

[0026] Specifically, iterative search refers to the process of gradually approaching the optimal solution within the updated feasible domain by repeatedly trying different combinations of control parameters for the raising and lowering mechanisms. The reward function is a mathematical function used to evaluate the quality of each search result. Here, minimizing the wear offset is used as the reward function, meaning that the smaller the wear offset, the higher the "reward," indicating that the parameter combination is better. The penalty term is a mechanism for "deducting points" from search results that do not meet expectations. The smoothness of the raising and lowering mechanism control parameters is used as a penalty term to avoid drastic fluctuations in the control parameters during the adjustment process.

[0027] Within the updated and determined feasible domain, intelligent optimization algorithms (such as genetic algorithms, particle swarm optimization algorithms, etc.) are used to randomly or according to a certain strategy select initial combinations of control parameters, which are then substituted into the calculation model of the reward function and penalty term. Minimizing the wear offset is used as the reward function. By establishing a pantograph-catenary wear prediction model (such as a model based on finite element analysis or machine learning), the wear offset under this parameter combination is obtained; the smaller the wear offset, the higher the reward.

[0028] Simultaneously, the variation range of the control parameters for the raising and lowering mechanisms is calculated between adjacent iterations. The greater the variation range, the higher the value of the penalty term. Through continuous iteration, based on the reward and penalty calculated each time, the direction of the parameter combination for the next search is adjusted, gradually moving towards the direction that maximizes the reward (minimizes wear offset) and minimizes the penalty (high parameter smoothness).

[0029] By introducing a reward function and a penalty term, when searching for the optimal combination of control parameters, we can not only focus on reducing the wear offset but also take into account the stability of parameter adjustment. In practice, if the parameter smoothness penalty term is not considered, the control parameters will be frequently and significantly adjusted, leading to increased vibration of the pantograph-catenary system and accelerating wear. By balancing the reward and penalty, the system can effectively reduce the wear offset under the premise of stable operation, significantly improving the practicality and effectiveness of the intelligent pantograph-catenary wear detection and control scheme.

[0030] Furthermore, by performing finite element simulation inversions to obtain a first fatigue crack propagation path set under multiple combinations of control parameters and a second fatigue crack propagation path set under multiple combinations of control parameters, the method of this application also includes: Based on the control parameters of the raising and lowering mechanisms, a first-class control parameter combination matrix and a second-class control parameter combination matrix are set respectively; based on the first-class and second-class control parameter combination matrices, an orthogonal array is configured to obtain the multiple sets of control parameter combinations.

[0031] Specifically, based on the control parameters of the pantograph raising and lowering mechanisms, two sets of control parameter combination matrices are set up respectively. This means that two different sets of parameter combination matrices are constructed according to the different control parameters during the pantograph raising and lowering processes, which are used for subsequent simulation and optimization analysis. Orthogonal array configuration based on the first and second control parameter combination matrices is to obtain multiple representative control parameter combinations through orthogonal experimental design methods, which are used to evaluate the influence of different parameter combinations on the fatigue crack propagation path of the pantograph.

[0032] Two types of control parameter combination matrices are set up. The first type of control parameter combination matrix corresponds to the control parameters of the pantograph raising mechanism, such as air pressure and raising speed. These parameters have a significant impact on the stress distribution and crack propagation of the pantograph during the raising process. The second type of control parameter combination matrix corresponds to the control parameters of the pantograph lowering mechanism, such as lowering angle and damping coefficient. These parameters also have a significant impact on the stress distribution and crack propagation of the pantograph during the lowering process.

[0033] In the control optimization model of the pantograph, the control parameter variables include gain coefficients and integral coefficients, which need to be optimized in the form of combination matrices. Using orthogonal arrays, representative parameter combinations are selected from the first and second class control parameter combination matrices. Orthogonal array design can ensure that all possible combinations of parameters are covered within a limited number of experiments, improving experimental efficiency and the reliability of results. Specifically, when performing static optimization of the pantograph, it is necessary to consider the combination of multiple parameters. Orthogonal array configuration can effectively screen out the key parameter combinations.

[0034] By configuring orthogonal arrays, multiple sets of different control parameter combinations are obtained. These combinations will be used for subsequent finite element simulation inversion to analyze the fatigue crack propagation path of the pantograph under different working conditions. In the above steps, by setting two types of control parameter combination matrices and configuring orthogonal arrays, the influence of different parameter combinations on the fatigue crack propagation of the pantograph can be systematically explored, providing a scientific basis for subsequent simulation analysis and optimization, ensuring the comprehensiveness and reliability of the results, and helping to more accurately assess the wear condition of the pantograph, thereby formulating reasonable maintenance strategies.

[0035] Furthermore, the method of this application also includes: Based on the control parameters of the bow lifting mechanism, a first parameter optimization space is constructed; the partial derivatives of the crack propagation rate with respect to the air pressure value and bow lifting speed in the control parameters of the bow lifting mechanism are calculated to determine the redundant control parameter combinations in the low sensitivity region; in the first parameter optimization space, the redundant control parameter combinations are used to perform redundancy elimination, and a control parameter combination matrix is ​​configured.

[0036] Specifically, constructing a first parameter optimization space based on the control parameters of the lifting mechanism refers to establishing a range and framework for parameter optimization based on the control parameters of the lifting mechanism (such as air pressure and lifting speed) for subsequent optimization analysis. Calculating the partial derivatives of the crack propagation rate with respect to the air pressure and lifting speed control parameters refers to mathematically determining the sensitivity of the crack propagation rate to changes in these control parameters. Identifying redundant control parameter combinations in the low-sensitivity region means finding parameter combinations that have a relatively small impact on the crack propagation rate and using these redundant control parameter combinations for redundancy elimination in the first parameter optimization space. Configuring a control parameter combination matrix means removing these redundant parameter combinations from the optimization space to obtain a more concise and effective control parameter combination matrix for subsequent analysis and optimization.

[0037] Based on the control parameters of the pantograph lifting mechanism (such as air pressure and lifting speed), the reasonable range and step size of these parameters are determined, and a parameter optimization space is constructed. This space covers all possible combinations of control parameters, providing a foundation for subsequent optimization analysis. Partial derivatives of the crack propagation rate with respect to air pressure and lifting speed are calculated to obtain the degree of influence of these parameters on the crack propagation rate. By analyzing the magnitude of the partial derivatives, it is determined which parameter combinations are in the low-sensitivity region, i.e., combinations with less impact on the crack propagation rate. Specifically, in the pantograph control optimization model, by constructing a PD controller for the pantograph-catenary contact force with the goal of minimizing the contact force output error, the influence of different control parameter combinations on the output error can be calculated, thereby determining redundant parameter combinations.

[0038] In the constructed first parameter optimization space, redundant control parameter combinations are eliminated, and parameter combinations that have a significant impact on crack propagation rate are retained, forming a class of control parameter combination matrices. This improves the relevance and effectiveness of parameter combinations and reduces unnecessary calculations. In the above steps, by constructing the parameter optimization space and calculating partial derivatives, the influence of the control parameters of the pantograph lifting mechanism on crack propagation rate can be systematically analyzed, providing a scientific basis for subsequent simulation and optimization. The control parameter combination matrix obtained after eliminating redundant parameter combinations not only improves computational efficiency but also ensures the accuracy and reliability of the analysis results, helping to more accurately assess the wear of the pantograph and thus formulate reasonable maintenance strategies.

[0039] Furthermore, the method of this application also includes: Define the crack propagation rate function: The crack propagation rate The air pressure value in the control parameters of the bow lifting mechanism First-order partial derivatives The crack propagation rate The bow lifting speed in the control parameters of the bow lifting mechanism First-order partial derivatives ,in, Used to characterize crack length Used to characterize the number of cycles.

[0040] Specifically, define the crack propagation rate function. ,in, This represents the crack propagation rate. To calculate the crack propagation rate, the effect of the crack propagation rate on the air pressure value in the control parameters of the lifting mechanism is determined. First-order partial derivatives Bow speed First-order partial derivatives The sensitivity of crack propagation rate to these control parameters, variables characterizing crack length and cycle number, are used to quantify crack propagation and fatigue life.

[0041] Based on the mechanics of materials and fatigue crack propagation theory, a crack propagation rate function is established, expressed as: ,in, Used to characterize crack length Used to characterize the number of cycles; the crack propagation rate function is respectively compared with the air pressure value. and draw speed Taking the first-order partial derivative, we get: The partial derivatives represent the sensitivity of the crack propagation rate to the air pressure and the pantograph lifting speed. By calculating these partial derivatives, it is possible to determine which control parameters have a greater impact on the crack propagation rate. In the above steps, by defining the crack propagation rate function and calculating the partial derivatives, the sensitivity of the crack propagation rate to the control parameters of the pantograph lifting mechanism is quantified, providing a scientific basis for subsequent parameter optimization and redundancy elimination, and helping to more accurately assess the fatigue life and wear of the pantograph.

[0042] Furthermore, the method of this application also includes: The crack propagation rate The air pressure value in the control parameters of the bow lifting mechanism and draw speed Second-order partial derivatives: Based on the air pressure value and draw speed The second-order partial derivatives are used to capture the interaction effects under the second-order partial derivatives, and the boundary adaptive adjustment is performed on the low-sensitivity region.

[0043] Specifically, the second-order partial derivatives of the crack propagation rate with respect to the air pressure and lifting speed in the control parameters of the lifting mechanism refer to the second-order partial derivatives of the crack propagation rate function with respect to the air pressure and lifting speed. These second-order partial derivatives can capture the interactive effect of the crack propagation rate on the control parameters, that is, the combined influence of the two parameters changing simultaneously on the crack propagation rate. Based on these second-order partial derivatives, boundary adaptive adjustment is performed on the low-sensitivity region. That is, according to the results of the second-order partial derivatives, the range of parameter combinations that have a smaller impact on the crack propagation rate is dynamically adjusted, thereby optimizing the parameter combination matrix.

[0044] In the crack propagation rate function Based on this, calculate the air pressure value. and draw speed Second-order partial derivatives: These second-order partial derivatives can capture the interaction effect of crack propagation rate on air pressure and lift velocity, that is, the combined effect of the simultaneous change of the two parameters on crack propagation rate; by analyzing the magnitude and sign of the second-order partial derivatives, the interaction effect between air pressure and lift velocity can be determined. A positive value indicates that increases in both air pressure and lifting speed contribute to an increase in the crack propagation rate; if... A negative value indicates that the increase in air pressure and bow lifting speed will jointly inhibit the increase in crack propagation rate.

[0045] Based on the results of the second-order partial derivatives, the boundary of the low-sensitivity region is adaptively adjusted. Specifically, if a certain parameter combination is in the low-sensitivity region and its second-order partial derivatives indicate that the region has little impact on the crack propagation rate, the boundary of the region can be appropriately expanded to reduce unnecessary calculations and analyses. In particular, by dynamically adjusting the range of parameter combinations, the analysis focuses on the region that has a greater impact on the crack propagation rate, thereby improving computational efficiency and the accuracy of the results.

[0046] By calculating the second-order partial derivatives and capturing the interaction effects, we can gain a more comprehensive understanding of the sensitivity of crack propagation rate to the control parameters of the pantograph lifting mechanism. This provides a more refined basis for subsequent parameter optimization and redundancy elimination. Boundary adaptive adjustment further optimizes the parameter combination matrix, ensuring the efficiency and relevance of the analysis. This helps to more accurately assess the wear of the pantograph, thereby formulating reasonable maintenance strategies and improving the reliability and safety of the system.

[0047] Furthermore, the method of this application also includes: Based on the control parameters of the raising mechanism, a second parameter optimization space is constructed; the damping coefficient fluctuation range and the lowering angle variation range are introduced; in the first parameter optimization space, the damping coefficient fluctuation range and the lowering angle variation range are used for data filtering, and a combination matrix of two types of control parameters is configured.

[0048] Specifically, constructing a second parameter optimization space based on the control parameters of the raising mechanism means, on the basis of the previously constructed first parameter optimization space, further considering other relevant control parameters, such as the damping coefficient and the lowering angle, to construct a more comprehensive parameter optimization range. Introducing the damping coefficient fluctuation range and the lowering angle variation range is to consider the changes of these parameters in actual operation, making the optimization results more consistent with the actual situation. Using these ranges in the first parameter optimization space for data filtering and configuring a second type of control parameter combination matrix means selecting parameter combinations that meet the requirements of the damping coefficient fluctuation range and the lowering angle variation range from the first parameter optimization space to form a new control parameter combination matrix for subsequent analysis and optimization.

[0049] Based on the existing control parameters of the pantograph lifting mechanism, further consideration is given to parameters such as damping coefficient and pantograph lowering angle to construct a more comprehensive second parameter optimization space. This space covers all possible combinations of control parameters, including air pressure, pantograph lifting speed, damping coefficient, and pantograph lowering angle, providing a more comprehensive foundation for subsequent optimization analysis. Specifically, in the integrated optimization model of high-speed pantographs, multiple influencing factors need to be comprehensively considered, including pole length parameters, cross-sectional parameters, and control parameters, to construct a comprehensive optimization space to ensure that the performance of the pantograph is comprehensively improved.

[0050] The damping coefficient fluctuation range refers to the possible range of changes in the damping coefficient during pantograph operation, and the pantograph descent angle variation range refers to the possible range of changes in the pantograph descent angle during pantograph descent. The introduction of these parameters makes the optimization model closer to the actual working conditions and can better reflect the performance of the pantograph under different working conditions. Specifically, in the pantograph dynamic optimization model, it is necessary to consider the influence of parameters such as damping coefficient and pantograph descent angle on the dynamic performance of the pantograph. By introducing the variation range of these parameters, the performance of the pantograph can be evaluated more comprehensively.

[0051] In the first parameter optimization space, based on the introduced damping coefficient fluctuation range and pantograph angle variation range, the original parameter combinations are screened and filtered, retaining those parameter combinations that meet these range requirements and eliminating those that do not, forming a second-class control parameter combination matrix. This improves the relevance and effectiveness of the parameter combinations, reduces unnecessary calculations and analyses, and makes the subsequent optimization process more efficient and accurate. For example, in the optimization algorithm, data filtering can ensure that the focus of the analysis is on the parameter combinations that have a greater impact on the pantograph performance, thereby improving the reliability and practicality of the optimization results.

[0052] By constructing a second parameter optimization space and introducing the damping coefficient fluctuation range and the pantograph lowering angle variation range, the performance of the pantograph under different working conditions can be considered more comprehensively. The combined matrix of the two types of control parameters after data filtering not only improves the calculation efficiency, but also ensures the accuracy and reliability of the analysis results. This helps to more accurately assess the wear of the pantograph, thereby formulating reasonable maintenance strategies and improving the reliability and safety of the system.

[0053] Furthermore, the method of this application involves performing three-dimensional convolution processing on the first fatigue crack propagation path set and the second fatigue crack propagation path set to extract spatiotemporal correlation features. Gradient features of the propagation path are extracted along the time, space, and crack depth dimensions, and a three-dimensional convolution kernel is set. Based on the three-dimensional convolution kernel, the first fatigue crack propagation path set and the second fatigue crack propagation path set are decomposed in the frequency domain to extract multi-frequency band features, including high-frequency abrupt change features and low-frequency trend features. Based on the multi-frequency band features, spatiotemporal correlation features are determined using an attention gating mechanism.

[0054] Specifically, extracting gradient features along the time, space, and crack depth dimensions of the crack propagation path refers to calculating gradients along the crack propagation path in the time dimension (cycle number), space dimension (crack location), and crack depth dimension, respectively, to capture the changing trend and direction of crack propagation. Setting a three-dimensional convolution kernel means constructing a three-dimensional convolution kernel for convolution operations in three-dimensional space to extract multi-dimensional features. Frequency domain decomposition based on the three-dimensional convolution kernel means transforming the set of crack propagation paths from the time domain to the frequency domain to analyze features under different frequency components. Multi-frequency features include high-frequency abrupt change features and low-frequency trend features, reflecting the rapid and slow changes in crack propagation, respectively. Determining spatiotemporal correlation features using an attention gating mechanism means using an attention mechanism to weight the extracted multi-frequency features, highlighting important spatiotemporal features and suppressing unimportant features, thereby improving the model's attention to key information.

[0055] Along the crack propagation path, gradient features are extracted along three dimensions: time, space, and crack depth. For example, in the spatiotemporal feature analysis of traffic flow networks, features are extracted along multiple dimensions such as time and space to capture the dynamic changes of traffic flow. A three-dimensional convolution kernel is set, with a size of 3×3×3 or 5×5×5, to perform convolution operations in three-dimensional space and extract multi-dimensional features. The three-dimensional convolution kernel can simultaneously capture feature changes in the time, space, and depth dimensions, improving the comprehensiveness and accuracy of feature extraction.

[0056] Based on a three-dimensional convolution kernel, frequency domain decomposition is performed on the first fatigue crack propagation path set and the second fatigue crack propagation path set. For example, in fatigue crack propagation acoustic emission signal processing, frequency domain decomposition is performed using wavelet transform and other methods to extract features under different frequency components. Multi-band features are extracted, including high-frequency abrupt change features and low-frequency trend features. High-frequency abrupt change features reflect rapid changes and local abrupt changes in the crack propagation process, while low-frequency trend features reflect the overall trend and slow changes in crack propagation.

[0057] Based on the extracted multi-band features, an attention gating mechanism is constructed. For example, in the prediction model that integrates the attention mechanism, the attention mechanism is constructed by using global average pooling layer, fully connected layer and sigmoid layer to weight the features. Through the attention gating mechanism, spatiotemporal correlation features are determined. The attention mechanism can highlight important spatiotemporal features and suppress unimportant features, making the model pay more attention to the key areas and time periods of crack propagation, thereby improving the accuracy and robustness of the model.

[0058] By extracting gradient features along the propagation path in the dimensions of time, space, and crack depth, and setting a three-dimensional convolution kernel, the multidimensional features of crack propagation can be comprehensively captured. Frequency domain decomposition and multi-band feature extraction further analyze the different frequency components of crack propagation, providing a foundation for refined feature description. The attention gating mechanism determines the spatiotemporal correlation features, improving the model's focus on key information, enhancing the model's feature expression ability and prediction accuracy, and helping to more accurately assess the wear of the pantograph, thereby formulating reasonable maintenance strategies.

[0059] In summary, the beneficial effects of the embodiments of this application are: By utilizing the pantograph-catenary contact area between the contact wire and the sliding plate to acquire pantograph operating status and surface image information; finite element simulation inversion is performed on the control parameters of the pantograph raising mechanism corresponding to the upper frame and lower boom, and the control parameters of the pantograph lowering mechanism corresponding to the upper frame and lower boom, respectively, to obtain a first fatigue crack propagation path set and a second fatigue crack propagation path set under multiple sets of control parameter combinations. Three-dimensional convolution processing is then performed to extract spatiotemporal correlation features. Based on the pantograph operating status and surface image information, environmental compensation data is added to correct perception biases. Multi-scale feature fusion is then performed using the first and second fatigue crack propagation path sets to generate a dynamic wear feature tensor under spatiotemporal correlation features for wear offset control. This application provides a machine vision-based intelligent pantograph-catenary wear detection method and system, inverts the fatigue crack propagation paths of the raising and lowering mechanisms, comprehensively assesses the pantograph-catenary wear condition, performs real-time remote wear detection and maintenance decisions for the pantograph and contact wire, promptly identifies and addresses potential wear hazards, avoids accidents, and improves detection accuracy and efficiency. Example

[0060] Based on the same inventive concept as the machine vision-based intelligent detection method for bow and cable wear in the foregoing embodiments, such as Figure 2 As shown in the figure, this application provides a machine vision-based intelligent detection system for pantograph-catenary wear, wherein the system includes: The information acquisition module M100 is used to acquire pantograph working status and surface image information in the pantograph-catenary contact area between the contact wire and the sliding plate.

[0061] The simulation module M200 is used to perform finite element simulation inversion on the control parameters of the raising mechanism corresponding to the upper frame and the lower boom, and the control parameters of the lowering mechanism corresponding to the upper frame and the lower boom, respectively, to obtain the first fatigue crack propagation path set and the second fatigue crack propagation path set under multiple sets of control parameter combinations.

[0062] The convolution processing module M300 is used to perform three-dimensional convolution processing on the first fatigue crack propagation path set and the second fatigue crack propagation path set to extract spatiotemporal correlation features.

[0063] The feature fusion module M400 is used to add environmental compensation data to correct perception bias based on the pantograph's working state and surface image information, and to perform multi-scale feature fusion by combining the first fatigue crack propagation path set and the second fatigue crack propagation path set to generate a dynamic wear feature tensor under spatiotemporal correlation features.

[0064] The offset control module M500 is used to perform wear offset control based on the dynamic wear feature tensor and in combination with the control parameters of the bow lifting mechanism and the bow lowering mechanism.

[0065] Furthermore, the offset control module M500 is used to perform the following method: Based on the dynamic wear feature tensor, the wear anomaly region and wear depth evolution index are determined; according to the wear anomaly region and wear depth evolution index, the feasible region of the control parameters of the raising mechanism and the lowering mechanism is updated, and the feasible region is mapped to the parameter combination that minimizes the wear offset obtained by the search.

[0066] Furthermore, the offset control module M500 is also used to perform the following methods: An iterative search is performed within the updated feasible region, with the minimization of wear offset as the reward function; during the iterative search process, the smoothness of the control parameters of the raising mechanism and the smoothness of the control parameters of the lowering mechanism are used as penalty terms.

[0067] Furthermore, the simulation module M200 is also used to perform the following methods: Based on the control parameters of the raising and lowering mechanisms, a first-class control parameter combination matrix and a second-class control parameter combination matrix are set respectively; based on the first-class and second-class control parameter combination matrices, an orthogonal array is configured to obtain the multiple sets of control parameter combinations.

[0068] Furthermore, the simulation module M200 is also used to perform the following methods: Based on the control parameters of the bow lifting mechanism, a first parameter optimization space is constructed; the partial derivatives of the crack propagation rate with respect to the air pressure value and bow lifting speed in the control parameters of the bow lifting mechanism are calculated to determine the redundant control parameter combinations in the low sensitivity region; in the first parameter optimization space, the redundant control parameter combinations are used to perform redundancy elimination, and a control parameter combination matrix is ​​configured.

[0069] Furthermore, the simulation module M200 is also used to perform the following methods: Define the crack propagation rate function: The crack propagation rate The air pressure value in the control parameters of the bow lifting mechanism First-order partial derivatives The crack propagation rate The bow lifting speed in the control parameters of the bow lifting mechanism First-order partial derivatives in, Used to characterize crack length Used to characterize the number of cycles.

[0070] Furthermore, the simulation module M200 is also used to perform the following methods: The crack propagation rate The air pressure value in the control parameters of the bow lifting mechanism and draw speed Second-order partial derivatives: Based on the air pressure value and draw speed The second-order partial derivatives are used to capture the interaction effects under the second-order partial derivatives, and the boundary adaptive adjustment is performed on the low-sensitivity region.

[0071] Furthermore, the simulation module M200 is also used to perform the following methods: Based on the control parameters of the raising mechanism, a second parameter optimization space is constructed; the damping coefficient fluctuation range and the lowering angle variation range are introduced; in the first parameter optimization space, the damping coefficient fluctuation range and the lowering angle variation range are used for data filtering, and a combination matrix of two types of control parameters is configured.

[0072] Furthermore, the convolution processing module M300 is used to perform the following method: Gradient features of the propagation path are extracted along the time, space, and crack depth dimensions, and a three-dimensional convolution kernel is set. Based on the three-dimensional convolution kernel, the first fatigue crack propagation path set and the second fatigue crack propagation path set are decomposed in the frequency domain to extract multi-frequency band features, including high-frequency abrupt change features and low-frequency trend features. Based on the multi-frequency band features, spatiotemporal correlation features are determined using an attention gating mechanism.

[0073] In summary, any step can be stored as a computer instruction or program in an unrestricted computer memory and can be called and recognized by an unrestricted computer processor; no further restrictions are imposed here.

[0074] Furthermore, the above technical solutions only embody the preferred technical solutions of the embodiments of this application. Any changes that those skilled in the art may make to certain parts of these solutions embody the novel principles of the embodiments of this application. Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application.

Claims

1. A machine vision-based intelligent detection method for pantograph-catenary wear, characterized in that, The method includes: In the pantograph-catenary contact area between the contact wire and the sliding plate, obtain pantograph working status and surface image information; The control parameters of the lifting mechanism corresponding to the upper frame and the lower boom, and the control parameters of the lowering mechanism corresponding to the upper frame and the lower boom, were respectively subjected to finite element simulation inversion to obtain the first fatigue crack propagation path set and the second fatigue crack propagation path set under multiple sets of control parameter combinations. The first fatigue crack propagation path set and the second fatigue crack propagation path set are subjected to three-dimensional convolution processing to extract spatiotemporal correlation features; Based on the pantograph's working status and surface image information, environmental compensation data is added to correct perception bias, and multi-scale feature fusion is performed by combining the first fatigue crack propagation path set and the second fatigue crack propagation path set to generate a dynamic wear feature tensor under spatiotemporal correlation features. Wear offset control is performed based on the dynamic wear characteristic tensor and the control parameters of the bow raising mechanism and bow lowering mechanism.

2. The intelligent detection method for bow and fire protection cable wear based on machine vision as described in claim 1, characterized in that, Based on the dynamic wear characteristic tensor, wear offset control is performed in conjunction with the control parameters of the raising and lowering mechanisms, including: Based on the dynamic wear feature tensor, the wear anomaly region and wear depth evolution index are determined; Based on the abnormal wear region and wear depth evolution index, the feasible region of the control parameters of the raising mechanism and the lowering mechanism is updated. The feasible region is mapped to the combination of control parameters that minimizes the wear offset obtained from the search.

3. The intelligent detection method for bow and catenary wear based on machine vision as described in claim 2, characterized in that, Perform an iterative search within the updated feasible region, using the minimization of wear offset as the reward function; During the iterative search process, the smoothness of the control parameters of the raising mechanism and the smoothness of the control parameters of the lowering mechanism are used as penalty terms.

4. The intelligent detection method for bow and catenary wear based on machine vision as described in claim 1, characterized in that, Finite element simulation inversions were performed to obtain the first fatigue crack propagation path set under multiple combinations of control parameters, the second fatigue crack propagation path set under multiple combinations of control parameters, and also including: Based on the control parameters of the raising and lowering mechanisms, a first-class control parameter combination matrix and a second-class control parameter combination matrix are respectively set. Based on the first type of control parameter combination matrix and the second type of control parameter combination matrix, an orthogonal table is configured to obtain the multiple sets of control parameter combinations.

5. The intelligent detection method for bow and catenary wear based on machine vision as described in claim 4, characterized in that, Based on the control parameters of the bow lifting mechanism, a first parameter optimization space is constructed; Calculate the partial derivatives of the crack propagation rate with respect to the air pressure value and the lifting speed in the control parameters of the lifting mechanism, and determine the combination of redundant control parameters in the low-sensitivity region; In the first parameter optimization space, the redundant control parameter combination is used to eliminate redundancy and configure a control parameter combination matrix.

6. The intelligent detection method for bow and catenary wear based on machine vision as described in claim 5, characterized in that, Define the crack propagation rate function: ; The crack propagation rate The air pressure value in the control parameters of the bow lifting mechanism First-order partial derivatives ; The crack propagation rate The bow lifting speed in the control parameters of the bow lifting mechanism First-order partial derivatives ,in, Used to characterize crack length Used to characterize the number of cycles.

7. The intelligent detection method for pantograph-catenary wear based on machine vision as described in claim 6, characterized in that, The crack propagation rate The air pressure value in the control parameters of the bow lifting mechanism and draw speed Second-order partial derivatives: ; Based on the air pressure value and draw speed The second-order partial derivatives are used to capture the interaction effects under the second-order partial derivatives, and the boundary adaptive adjustment is performed on the low-sensitivity region.

8. The intelligent detection method for bow and catenary wear based on machine vision as described in claim 4, characterized in that, Based on the control parameters of the bow lowering mechanism, a second parameter optimization space is constructed; Introduce the damping coefficient fluctuation range and the bow angle variation range; In the first parameter optimization space, the damping coefficient fluctuation range and the bow angle change range are used for data filtering, and a combination matrix of two types of control parameters is configured.

9. The intelligent detection method for bow and catenary wear based on machine vision as described in claim 1, characterized in that, The first fatigue crack propagation path set and the second fatigue crack propagation path set are subjected to three-dimensional convolution processing to extract spatiotemporal correlation features, including: Gradient features of the propagation path are extracted along the time, space, and crack depth dimensions, and a three-dimensional convolution kernel is set. Based on the three-dimensional convolution kernel, the first fatigue crack propagation path set and the second fatigue crack propagation path set are decomposed in the frequency domain to extract multi-frequency band features, including high-frequency abrupt change features and low-frequency trend features. Based on the multi-band features, spatiotemporal correlation features are determined using an attention gating mechanism.

10. A machine vision-based intelligent detection system for pantograph-catenary wear, characterized in that, For implementing the machine vision-based intelligent detection method for bow and cable rebar wear according to any one of claims 1-9, the system comprises: The information acquisition module is used to acquire pantograph working status and surface image information in the pantograph-catenary contact area between the contact wire and the sliding plate; The simulation module is used to perform finite element simulation inversion on the control parameters of the lifting mechanism corresponding to the upper frame and the lower boom, and the control parameters of the lowering mechanism corresponding to the upper frame and the lower boom, respectively, to obtain the first fatigue crack propagation path set and the second fatigue crack propagation path set under multiple sets of control parameter combinations. The convolution processing module is used to perform three-dimensional convolution processing on the first fatigue crack propagation path set and the second fatigue crack propagation path set to extract spatiotemporal correlation features. The feature fusion module is used to add environmental compensation data to correct perception deviations based on the pantograph's working state and surface image information, and to perform multi-scale feature fusion by combining the first fatigue crack propagation path set and the second fatigue crack propagation path set to generate a dynamic wear feature tensor under spatiotemporal correlation features. The offset control module is used to perform wear offset control based on the dynamic wear feature tensor and in combination with the control parameters of the bow lifting mechanism and the bow lowering mechanism.

Images