Steel rail damage online intelligent identification and judgment method based on edge calculation

Abstract

The invention discloses a steel rail damage online intelligent identification and judgment method based on edge calculation, and the method comprises the following steps: S1, collecting data through an edge calculation platform, and carrying out the preprocessing; s2, carrying out real-time analysis at an edge computing node, and generating a preliminary damage result; s3, structural optimization is carried out in combination with geometric features, loads and historical data; s4, improving the generative adversarial network, and enhancing the repair precision of the damaged region; and S5, processing flaw positioning and classification in real time, and optimizing and feeding back an identification result. According to the method, a self-adaptive hierarchical multi-scale topological optimization algorithm and simulated annealing are adopted to optimize the steel rail structure, improvement is carried out based on the generative adversarial network, and real-time detection, intelligent optimization and accurate positioning of steel rail damage are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of rail damage detection, and in particular to an online intelligent identification method for rail damage based on edge computing. Background Art

[0002] In modern railway transportation systems, rail damage detection and maintenance management have always been key factors in ensuring safe railway operation. As a critical component that bears enormous pressure during railway transportation, rail damage not only affects railway operation safety but can also lead to major accidents. Therefore, early detection and accurate identification of rail damage are crucial to railway transportation safety. Traditional rail damage detection methods rely on manual inspections, mechanical testing, or static sensor monitoring. These methods often face many limitations, particularly in terms of real-time performance, accuracy, and adaptability.

[0003] Traditional manual inspection methods rely primarily on human vision and simple tools to detect cracks or damage on the rail surface. While this method achieved some success in early railway operations, with the increase in rail transportation volume and the continued expansion of inspection scope, manual inspections have gradually exposed problems such as low efficiency and high error rates. Manual inspections not only require significant manpower and material resources, but also fail to provide all-weather, real-time monitoring of rails. Furthermore, due to human factors, manual inspections often miss areas, especially in inclement weather or poor lighting conditions, leading to serious problems with missed inspections and misidentification of damaged areas.

[0004] Compared with manual inspection, mechanical inspection methods offer higher accuracy in some aspects. Mechanical inspection methods mainly utilize track inspection vehicles or automated inspection equipment to scan the rail surface with sensors and automatically identify cracks and other damage on the rail surface. However, this method usually has many limitations. First, mechanical inspection relies on large equipment, which is costly and not suitable for all tracks, especially those in complex and narrow track environments. Second, mechanical inspection equipment may not provide sufficient accuracy at high speeds, especially in the detection of tiny cracks. Mechanical equipment often has blind spots, resulting in some small or early damage not being discovered in time.

[0005] To improve the accuracy and efficiency of rail damage detection, sensor technology and data analysis methods have been widely used in recent years. The combination of static and dynamic sensors makes real-time monitoring of rails possible. However, the effectiveness of single sensor technology in complex working environments remains limited. For example, vibration sensors can be used to detect rail deformation, but when faced with more complex damage types, the sensors may have difficulty accurately distinguishing between cracks, indentations, and other types of damage. In addition, environmental factors such as temperature, humidity, rain, and snow can also interfere with the sensor's output signal, affecting the accuracy of the data. Therefore, data processing and analysis from a single sensor often cannot meet actual needs, especially for the precise location and classification of damage types and locations.

[0006] Currently, rail damage detection methods based on machine learning and deep learning are becoming a hot topic of research. By analyzing sensor data and combining it with image processing and pattern recognition techniques, more accurate rail damage identification can be achieved. However, existing AI-based rail damage detection methods still face several technical challenges. First, many existing methods rely on static data models, which makes them less adaptable to the dynamic and complex railway operating environment. For example, under adverse weather conditions, varying rail loads, and other environmental factors, model accuracy can drop significantly. Second, although deep learning and other methods have made significant progress in image recognition, image processing in complex backgrounds remains a significant challenge. Rail damage images often contain significant noise, and details of damaged areas are unclear, which limits the effectiveness of image recognition algorithms. Furthermore, while existing generative adversarial network (GAN) technology has demonstrated excellent performance in image restoration, effectively merging multimodal data and improving the model's detail representation and robustness remain unresolved challenges in rail damage image restoration and classification.

[0007] With the development of edge computing and cloud computing technologies, edge computing nodes can realize local real-time data processing, greatly improving the real-time and accuracy of rail damage detection. However, existing edge computing solutions mostly rely on a single data processing method and lack the adaptive ability to optimize the rail working environment and structure. Although edge computing nodes can reduce data transmission delays and improve real-time feedback capabilities, without flexible algorithm support, it is still difficult to adapt to different rail damage types and structural changes in a dynamic environment, resulting in insufficient damage identification accuracy. In addition, the synergy between real-time data analysis based on edge computing and cloud-based model training still has problems with data synchronization and model update delays, further affecting the accuracy and efficiency of the overall detection system.

[0008] In summary, while existing rail damage detection technologies have improved detection efficiency and accuracy to a certain extent, they still have certain limitations and shortcomings. Traditional methods rely on manual inspections or mechanical equipment, which are subject to low efficiency, high cost, and large errors. Modern sensor-based and artificial intelligence-based detection methods, while highly automated, still face technical challenges in practical applications, such as data noise, environmental interference, and poor algorithm adaptability. Therefore, how to introduce more intelligent and adaptive algorithms into rail damage detection, improve the ability to integrate multimodal data, and combine edge computing and cloud optimization to achieve real-time and accurate damage identification has become a key direction of current technological development.

[0009] Therefore, how to provide an online intelligent identification method for rail damage based on edge computing is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0010] One objective of the present invention is to propose an online intelligent identification method for rail damage based on edge computing. This method fully utilizes edge computing, generative adversarial networks, topology optimization algorithms, and adaptive model updating techniques. It describes in detail a method for intelligently identifying and classifying rail damage by collecting and analyzing sensor data from rail surfaces in real time through an edge computing platform, combined with advanced adaptive optimization algorithms and deep learning techniques. This method enables real-time detection, precise location, and classification of rail damage, optimizes rail structural design, and continuously optimizes identification algorithms in dynamic environments. It possesses the advantages of high real-time performance, strong accuracy, wide adaptability, and robust system optimization capabilities.

[0011] The method for online intelligent identification of rail damage based on edge computing according to an embodiment of the present invention includes the following steps:

[0012] 1. An online intelligent identification method for rail damage based on edge computing is characterized by comprising the following steps:

[0013] S1. Deploy a rail damage detection system through an edge computing platform to collect real-time sensor data from the rail surface under different working environments. A filtering algorithm is used to remove environmental noise and perform data normalization. Local data preprocessing is achieved through time window division, feature extraction, and support vector machine preliminary recognition.

[0014] S2. On the edge computing node, the rail data collected by the sensor is analyzed in real time to identify whether there are obvious signs of damage on the rail surface and generate a preliminary damage judgment result;

[0015] S3. Based on the preliminary judgment results, edge computing nodes are used to collect the rail's geometric characteristics, load distribution, environmental data, and historical damage information. An initial finite element model is constructed and the initial stress and strain are calculated. The rail structure is optimized by combining an adaptive hierarchical multi-scale topology optimization algorithm to minimize total energy and maximize stiffness. Considering damage tolerance, a simulated annealing algorithm is introduced for global search. The structure is optimized by combining an adaptive optimization mechanism and a hierarchical strategy. By dynamically adjusting the material distribution and structural morphology, the design is finally optimized through finite element simulation analysis.

[0016] S4. Within the rail inspection area, improvements are made based on a generative adversarial network (GAN) by introducing a multimodal attention mechanism. The improved GAN simultaneously processes multimodal information from rail images and environmental data. The generator dynamically adjusts the attention weights of image regions, enhancing the repair accuracy, detail expression, and damage feature prominence of damaged areas.

[0017] S5. Optimize the rail structure based on an adaptive hierarchical multi-scale topology optimization algorithm and annealing algorithm. Combined with an improved generative adversarial network, the computing power of edge computing nodes is utilized to process damage location and classification in real time, optimize recognition results, and conduct data interaction and model optimization through a cloud platform to achieve accuracy improvement and operation and maintenance feedback for the rail damage detection system.

[0018] 2. The method for online intelligent identification of rail damage based on edge computing according to claim 1, wherein S1 specifically comprises:

[0019] S11. Deploy a rail damage detection system through edge computing nodes and collect rail surface data from different sensors in real time.

[0020] S12, using a filtering algorithm to remove noise from the collected sensor data;

[0021] S13. Normalize the denoised data to standardize data from different sources to the same scale

[0022] S14. Divide the data into time windows through the edge computing node, and process the data in each time period as a subset;

[0023] S15. Based on the working environment characteristics of the rail, use a statistical analysis method based on machine learning to extract features from each data subset;

[0024] S16. Combine the extracted features and use the support vector machine algorithm to perform preliminary damage detection on the rail surface;

[0025] S17. Implement local data preprocessing through the edge computing platform.

[0026] 3. The method for online intelligent identification of rail damage based on edge computing according to claim 1, wherein S3 specifically includes:

[0027] S31, collect the geometric characteristics, load distribution, working environment data and historical damage information of the rail in real time through the edge computing node, build the initial finite element model, and for each discrete unit x of the rail i Calculate the initial stress based on its geometry, load distribution and environmental factors and initial strain And use this as the initial condition:

[0028]

[0029] Where C is the constitutive matrix of the material, f i For unit x i external loads;

[0030] S32. Use the adaptive hierarchical multi-scale topology optimization algorithm to perform preliminary optimization of the rail structure. According to finite element analysis, the objective function of rail topology optimization is to minimize the total energy of the rail and maximize its stiffness, while also considering the damage tolerance of the rail. The objective function is:

[0031]

[0032] Among them, Y is the objective function, is the norm of the gradient of the stress field, ∈(x) is the strain field, α1, α2 are weighting coefficients, Ω is the calculation area, is the stress gradient, It is the process of minimizing the objective function;

[0033] S33. Combine the simulated annealing algorithm with topology optimization to perform global search and local optimization. Set the initial temperature of the simulated annealing to T0. The temperature update formula during the annealing process is:

[0034]

[0035] Among them, δE k is the energy difference between the current solution and the candidate solution, γ is the temperature attenuation coefficient, and the simulated annealing algorithm decides whether to accept the new solution through a probabilistic selection mechanism:

[0036]

[0037] Among them, δE k is the energy difference between the current solution and the candidate solution, and min is the process of minimizing the objective function;

[0038] S34. An adaptive hierarchical strategy is introduced to decompose the optimization problem into multiple optimization problems at different scales. At each scale, optimization is performed by refining the finite element mesh and adjusting local constraints. A regularization factor is introduced to maintain the stability and continuity of the structure. For each level of optimization problem, the objective function is:

[0039]

[0040] Among them, w i is the weight factor of the i-th level, α1 and α2 are the weight coefficients of the adjustment energy term, is the norm of the gradient of the stress distribution at the i-th level, ∈ i (x) is the strain distribution of the i-th level; σ i (x) is the stress distribution at the i-th level;

[0041] S35. To cope with complex working environments and dynamic changes, an adaptive optimization mechanism is set up to update the optimization strategy based on real-time data. The optimization process of the rail minimizes losses by dynamically adjusting the material distribution and structural morphology. The adaptive update rules are as follows:

[0042]

[0043] in, For the energy change after adaptive update, is the time rate of change of the stress field, α3, α4, α5 are weighting coefficients;

[0044] S36. Throughout the optimization process, a multi-scale approach is used for local and global collaborative optimization. For each level of optimization, the optimization process combines simulated annealing with an adaptive hierarchical strategy to conduct local exploration and global search, ultimately obtaining the global optimal solution. During each round of simulated annealing, the simulated annealing algorithm decides whether to accept a new solution based on the energy difference between the current rail structure optimization solution and the candidate solution.

[0045] S37, through the finite element simulation analysis of the optimized rail structure, calculate the optimized stress field σ opt (x) and the strain field ∈ opt (x), and output the optimized rail structure design. At the same time, the rails are monitored in real time through edge computing nodes, and the detection area is dynamically adjusted according to the optimization results.

[0046] 4. The method for online intelligent identification of rail damage based on edge computing according to claim 1, wherein S4 specifically includes:

[0047] S41. Based on the generative adversarial network framework, a generator network is used to enhance and repair images of damaged rail areas. The generator network combines a multimodal attention mechanism to optimize the repair accuracy of damaged area images, highlight the characteristics of damaged areas, and classify damage types through a multi-task learning architecture. The optimization objectives of the generator network are:

[0048] L gen =λ1L adv +λ2L rec +λ3L reg +λ4L smooth ;

[0049] Among them, L adv To combat the loss, L rec To reconstruct the loss, L reg is the regularization loss, L smooth is the smoothing loss, λ1,λ2,λ3,λ4 are weight factors;

[0050] S42. The generator network adopts a multimodal attention mechanism and uses environmental data as additional input to perform feature fusion, so that the generator focuses on the damaged area of the rail. The contribution of different modal information is weighted by the attention mechanism to optimize the repair effect of the generated image and obtain an image of the damaged area. The optimization goal is:

[0051]

[0052] Among them, α i is the attention weight of the i-th modal data, G(x,z i ) is the repaired image generated based on the rail image and environmental data, G′(x,z i ) is the real rail damage area image, z i For environmental data;

[0053] S43. While repairing the image, the generator network uses a multi-task learning architecture to simultaneously perform damage area repair and damage type classification tasks. The optimization goal of damage type classification is:

[0054]

[0055] Among them, C j is the real damage type label, C pred is the damage type label predicted by the classifier, and M is the number of categories;

[0056] S44. The regularization part of the generator loss function introduces multi-scale constraints, combines the dynamic working environment data of the rail to adjust the details of the repaired image, and optimizes the detail performance and global structure of the image at different scales. The loss function is:

[0057]

[0058] Among them, G k (x,z) is the generated image at the kth scale, G′ k (x,z) is the real rail area image at the kth scale, β k is the scale weight, K is the number of scales;

[0059] S45. Based on the above optimized generated images and classification results, a discriminator network of the adversarial network is generated to further evaluate the authenticity of the generated damage images. The generator and discriminator are jointly optimized through adversarial training to improve the repair accuracy and feature visibility of the rail damage area. The loss function of the discriminator is:

[0060] L disc =logD(x)+log(1-D(G(x,z)));

[0061] Where D(x) is the discriminator's prediction of the input image, and D(G(x,z)) is the discriminator's prediction of the image generated by the generator;

[0062] S46. After the generated rail damage image is repaired and enhanced, it is further subjected to feature extraction through a convolutional neural network, and the damaged area is classified in combination with a classifier to generate the final recognition result of the rail damage.

[0063] The beneficial effects of the present invention are as follows: the online intelligent identification method of rail damage based on edge computing of the present invention has achieved significant technological breakthroughs and application innovations in the field of rail damage detection by combining edge computing, generative adversarial networks, multimodal attention mechanisms, topology optimization algorithms and adaptive optimization technologies. First, the use of the edge computing platform to collect and process rail surface sensor data in real time not only effectively improves the timeliness of data processing, but also avoids the delay problem in the traditional centralized computing mode. Through local data preprocessing, real-time analysis and transmission, it ensures that the detection system can quickly identify damage during rail operation, significantly improving the real-time performance and response speed of damage monitoring.

[0064] Secondly, the adaptive hierarchical multi-scale topology optimization algorithm employed in this invention optimizes the rail structure, minimizing damage tolerance and energy loss while simultaneously increasing stiffness, significantly improving the rail's durability and structural performance in complex operating environments. Combined with a simulated annealing algorithm for global search and local optimization, this algorithm ensures that the rail design remains optimal under constantly changing operating environments and loading conditions. This optimized design not only enhances the rail's structural stability but also provides more accurate structural data for subsequent damage detection, further improving detection accuracy.

[0065] In terms of damage identification, this invention uses a generative adversarial network (GAN) and a multimodal attention mechanism to improve image processing and information fusion methods. The generator can dynamically adjust the attention weight based on the multimodal information of the rail image and environmental data, thereby accurately repairing and enhancing the details of the damaged area image, highlighting the damage characteristics, and improving the accuracy of damage repair. This innovation not only optimizes the visualization of the damaged area, but also provides more reliable data support for the classification of rail damage types. Through a multi-task learning architecture, the generator can classify the damage type while repairing the image, making the recognition results more accurate and reducing the risk of misidentification and missed identification.

[0066] Furthermore, through real-time model updates and adaptive optimization mechanisms, the present invention continuously optimizes the damage identification algorithm based on the dynamic operating environment and real-time data of the rail. This capability enables the system to adapt to complex changes in diverse operating environments, improving its adaptability and long-term stability in practical applications. Through data exchange between edge computing nodes and the cloud platform, the rail damage detection system can continuously improve its identification accuracy and overall performance.

[0067] Ultimately, this invention not only enables real-time detection and precise location of rail damage, but also possesses dynamic optimization capabilities, automatically adjusting detection strategies to suit different working environments and task requirements. The application of this system will significantly improve the safety and efficiency of railway transportation, reduce accidents and maintenance costs caused by rail damage, and provide railway operators with more accurate and real-time decision-making support. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0069] Figure 1 This is a flow chart of the online intelligent identification method for rail damage based on edge computing proposed by the present invention;

[0070] Figure 2 The adaptive hierarchical multi-scale topology optimization algorithm of the edge computing-based online intelligent identification method for rail damage proposed in this invention combines finite element analysis and simulated annealing algorithm to perform structural optimization on the rail; DETAILED DESCRIPTION

[0071] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0072] refer to Figure 1 and Figure 2The online intelligent identification method of rail damage based on edge computing includes the following steps:

[0073] The method for online intelligent identification of rail damage based on edge computing according to an embodiment of the present invention includes the following steps:

[0074] 1. An online intelligent identification method for rail damage based on edge computing is characterized by comprising the following steps:

[0075] S1. Deploy a rail damage detection system through an edge computing platform to collect real-time sensor data from the rail surface under different working environments. A filtering algorithm is used to remove environmental noise and perform data normalization. Local data preprocessing is achieved through time window division, feature extraction, and support vector machine preliminary recognition.

[0076] S2. On the edge computing node, the rail data collected by the sensor is analyzed in real time to identify whether there are obvious signs of damage on the rail surface and generate a preliminary damage judgment result;

[0077] S3. Based on the preliminary judgment results, edge computing nodes are used to collect the rail's geometric characteristics, load distribution, environmental data, and historical damage information. An initial finite element model is constructed and the initial stress and strain are calculated. The rail structure is optimized by combining an adaptive hierarchical multi-scale topology optimization algorithm to minimize total energy and maximize stiffness. Considering damage tolerance, a simulated annealing algorithm is introduced for global search. The structure is optimized by combining an adaptive optimization mechanism and a hierarchical strategy. By dynamically adjusting the material distribution and structural morphology, the design is finally optimized through finite element simulation analysis.

[0078] S4. Within the rail inspection area, improvements are made based on a generative adversarial network (GAN) by introducing a multimodal attention mechanism. The improved GAN simultaneously processes multimodal information from rail images and environmental data. The generator dynamically adjusts the attention weights of image regions, enhancing the repair accuracy, detail expression, and damage feature prominence of damaged areas.

[0079] S5. Optimize the rail structure based on an adaptive hierarchical multi-scale topology optimization algorithm and annealing algorithm. Combined with an improved generative adversarial network, the computing power of edge computing nodes is utilized to process damage location and classification in real time, optimize recognition results, and conduct data interaction and model optimization through a cloud platform to achieve accuracy improvement and operation and maintenance feedback for the rail damage detection system.

[0080] 2. The method for online intelligent identification of rail damage based on edge computing according to claim 1, wherein S1 specifically comprises:

[0081] S11. Deploy a rail damage detection system through edge computing nodes and collect rail surface data from different sensors in real time.

[0082] S12, using a filtering algorithm to remove noise from the collected sensor data;

[0083] S13. Normalize the denoised data to standardize data from different sources to the same scale

[0084] S14. Divide the data into time windows through the edge computing node, and process the data in each time period as a subset;

[0085] S15. Based on the working environment characteristics of the rail, use a statistical analysis method based on machine learning to extract features from each data subset;

[0086] S16. Combine the extracted features and use the support vector machine algorithm to perform preliminary damage detection on the rail surface;

[0087] S17. Implement local data preprocessing through the edge computing platform.

[0088] 3. The method for online intelligent identification of rail damage based on edge computing according to claim 1, wherein S3 specifically includes:

[0089] S31, collect the geometric characteristics, load distribution, working environment data and historical damage information of the rail in real time through the edge computing node, build the initial finite element model, and for each discrete unit x of the rail i Calculate the initial stress based on its geometry, load distribution and environmental factors and initial strain And use this as the initial condition:

[0090]

[0091] Where C is the constitutive matrix of the material, f i For unit x i external loads;

[0092] S32. Use the adaptive hierarchical multi-scale topology optimization algorithm to perform preliminary optimization of the rail structure. According to finite element analysis, the objective function of rail topology optimization is to minimize the total energy of the rail and maximize its stiffness, while also considering the damage tolerance of the rail. The objective function is:

[0093]

[0094] Among them, Y is the objective function, is the norm of the gradient of the stress field, ∈(x) is the strain field, α1, α2 are weighting coefficients, Ω is the calculation area, is the stress gradient, It is the process of minimizing the objective function;

[0095] S33. Combine the simulated annealing algorithm with topology optimization to perform global search and local optimization. Set the initial temperature of the simulated annealing to T0. The temperature update formula during the annealing process is:

[0096]

[0097] Among them, δE k is the energy difference between the current solution and the candidate solution, γ is the temperature attenuation coefficient, and the simulated annealing algorithm decides whether to accept the new solution through a probabilistic selection mechanism:

[0098]

[0099] Among them, δE k is the energy difference between the current solution and the candidate solution, and min is the process of minimizing the objective function;

[0100] S34. An adaptive hierarchical strategy is introduced to decompose the optimization problem into multiple optimization problems at different scales. At each scale, optimization is performed by refining the finite element mesh and adjusting local constraints. A regularization factor is introduced to maintain the stability and continuity of the structure. For each level of optimization problem, the objective function is:

[0101]

[0102] Among them, w i is the weight factor of the i-th level, α1 and α2 are the weight coefficients of the adjustment energy term, is the norm of the gradient of the stress distribution at the i-th level, ∈ i (x) is the strain distribution of the i-th level; σ i (x) is the stress distribution at the i-th level;

[0103] S35. To cope with complex working environments and dynamic changes, an adaptive optimization mechanism is set up to update the optimization strategy based on real-time data. The optimization process of the rail minimizes losses by dynamically adjusting the material distribution and structural morphology. The adaptive update rules are as follows:

[0104]

[0105] in, For the energy change after adaptive update, is the time rate of change of the stress field, α3, α4, α5 are weighting coefficients;

[0106] S36. Throughout the optimization process, a multi-scale approach is used for local and global collaborative optimization. For each level of optimization, the optimization process combines simulated annealing with an adaptive hierarchical strategy to conduct local exploration and global search, ultimately obtaining the global optimal solution. During each round of simulated annealing, the simulated annealing algorithm decides whether to accept a new solution based on the energy difference between the current rail structure optimization solution and the candidate solution.

[0107] S37, through the finite element simulation analysis of the optimized rail structure, calculate the optimized stress field σ opt (x) and the strain field ∈ opt (x), and output the optimized rail structure design. At the same time, the rails are monitored in real time through edge computing nodes, and the detection area is dynamically adjusted according to the optimization results.

[0108] 4. The method for online intelligent identification of rail damage based on edge computing according to claim 1, wherein S4 specifically includes:

[0109] S41. Based on the generative adversarial network framework, a generator network is used to enhance and repair images of damaged rail areas. The generator network combines a multimodal attention mechanism to optimize the repair accuracy of damaged area images, highlight the characteristics of damaged areas, and classify damage types through a multi-task learning architecture. The optimization objectives of the generator network are:

[0110] L gen =λ1L adv +λ2L rec +λ3L reg +λ4L smooth ;

[0111] Among them, L adv To combat the loss, L rec To reconstruct the loss, L reg is the regularization loss, L smooth is the smoothing loss, λ1,λ2,λ3,λ4 are weight factors;

[0112] S42. The generator network adopts a multimodal attention mechanism and uses environmental data as additional input to perform feature fusion, so that the generator focuses on the damaged area of the rail. The contribution of different modal information is weighted by the attention mechanism to optimize the repair effect of the generated image and obtain an image of the damaged area. The optimization goal is:

[0113]

[0114] Among them, α i is the attention weight of the i-th modal data, G(x,z i ) is the repaired image generated based on the rail image and environmental data, G′(x,zi ) is the real rail damage area image, z i For environmental data;

[0115] S43. While repairing the image, the generator network uses a multi-task learning architecture to simultaneously perform damage area repair and damage type classification tasks. The optimization goal of damage type classification is:

[0116]

[0117] Among them, C j is the real damage type label, C pred is the damage type label predicted by the classifier, and M is the number of categories;

[0118] S44. The regularization part of the generator loss function introduces multi-scale constraints, combines the dynamic working environment data of the rail to adjust the details of the repaired image, and optimizes the detail performance and global structure of the image at different scales. The loss function is:

[0119]

[0120] Among them, G k (x,z) is the generated image at the kth scale, G′ k (x,z) is the real rail area image at the kth scale, β k is the scale weight, K is the number of scales;

[0121] S45. Based on the above optimized generated images and classification results, a discriminator network of the adversarial network is generated to further evaluate the authenticity of the generated damage images. The generator and discriminator are jointly optimized through adversarial training to improve the repair accuracy and feature visibility of the rail damage area. The loss function of the discriminator is:

[0122] L disc =logD(x)+log(1-D(G(x,z)));

[0123] Where D(x) is the discriminator's prediction of the input image, and D(G(x,z)) is the discriminator's prediction of the image generated by the generator;

[0124] S46. After the generated rail damage image is repaired and enhanced, it is further subjected to feature extraction through a convolutional neural network, and the damaged area is classified in combination with a classifier to generate the final recognition result of the rail damage.

[0125] Example 1:

[0126] To verify the feasibility of the present invention, it was applied to a railway transportation company to detect and optimize rail damage. The company's rails often suffer varying degrees of damage on long-term railway operations, leading to safety hazards and increasing maintenance costs year by year. Rail damage has always been a major issue facing railway transportation companies, especially in complex working environments. Traditional detection methods have difficulty identifying and accurately locating minor rail damage in real time, and the maintenance process cannot be accurately optimized, resulting in a significant waste of resources.

[0127] During the implementation process, a rail damage detection system was first deployed on the edge computing platform to collect real-time sensor data from the rail surface under different operating conditions. The sensors collected real-time information such as rail surface vibration, temperature, and pressure. A filtering algorithm was then used to remove environmental noise and perform data normalization, completing local data preprocessing. This step ensured the accuracy and reliability of subsequent data analysis.

[0128] The edge computing node then begins real-time analysis of the collected rail data, identifying any obvious signs of damage on the rail surface. During the initial system operation, a support vector machine algorithm performs a preliminary analysis of the data, identifying several areas of suspected damage. The system then automatically generates a preliminary assessment of the damage and promptly feeds it back to the monitoring platform. This process significantly improves identification speed and accuracy, saving significant time and labor costs compared to traditional manual inspection methods.

[0129] Next, on the edge computing node, an initial finite element model was constructed based on the rail's geometric characteristics, load distribution, environmental data, and historical damage information, and the rail's initial stress and strain were calculated. These initial conditions provided the foundational data for subsequent rail structural optimization. At this point, the adaptive hierarchical multi-scale topology optimization algorithm played a crucial role. This algorithm, incorporating the rail's actual operating environment and considering different loading conditions and damage tolerances, performed a preliminary structural optimization.

[0130] During the actual optimization process, a global search using a simulated annealing algorithm gradually optimized the rail's material distribution and structural morphology, minimizing stress concentration in damaged areas and maximizing rail stiffness. The simulated annealing algorithm helped the system escape local optimal solutions, improving global optimization performance. Ultimately, after multiple optimization attempts, the rail structure achieved the desired optimization results in finite element simulations.

[0131] To verify the reliability and feasibility of the optimized rail structure, the system uses real-time monitoring data, combined with information from the dynamic operating environment, to adaptively update the model and adjust the optimization strategy. Whenever new rail damage is detected, the system reanalyzes the data and conducts optimization, ensuring the stability and safety of the rails in long-term operation.

[0132] During implementation, the optimized rail damage identification results were uploaded to a cloud platform for further integration optimization and model training, improving the overall accuracy and adaptability of the rail damage detection system. Continuous updates to the optimized rail structural design ensured the system's adaptability to diverse operating environments and provided accurate maintenance recommendations to railway operators. This intelligent detection and optimization mechanism enables railway operators to achieve more precise maintenance, reduce maintenance costs, and improve overall safety.

[0133] To verify the effectiveness of this method, we conducted detailed data collection before and after implementation. Taking a railway line as an example, we conducted a comparative analysis of traditional manual inspection methods and the proposed method for online intelligent rail damage identification based on edge computing.

[0134] Table 1: Comparison of rail damage detection and optimization effects

[0135]

[0136]

[0137] The data in the table demonstrates the significant advantages of edge computing and the adaptive hierarchical multi-scale topology optimization algorithm for rail damage detection and structural optimization. First, comparing data acquisition to optimization results before and after optimization demonstrates that using an edge computing platform for data collection and processing effectively detects damage on the rail surface and monitors its condition in real time.

[0138] In terms of damage identification, the table shows that the accuracy of damage detection has increased by nearly 20% after optimization, from the original 85% to 95%. This result fully demonstrates the effectiveness of introducing a generative adversarial network and a multimodal attention mechanism. The generative adversarial network can perform a fusion analysis based on environmental data and rail images, enhancing the repair accuracy and detailed representation of damaged areas, making the damage characteristics more prominent. In addition, by introducing a structural optimization method that combines a simulated annealing algorithm with topology optimization, the table shows that the stability of the optimized rail structure under complex working conditions has increased by approximately 15%. This optimization not only plays a key role in the stress distribution of the rail, but also effectively reduces the strain energy of the structure and increases the service life of the rail.

[0139] For real-time monitoring and dynamic adjustment of optimization results, a table shows how the rail damage area and optimized structural design change over the entire monitoring cycle. In practice, by continuously collecting and analyzing real-time data, edge computing nodes can effectively adjust the rail detection area, thereby improving the accuracy and performance of damage identification. This also demonstrates the high adaptability of this invention in real-world environments and its ability to dynamically respond to complex working conditions and environmental changes.

[0140] The overall data shows that the optimization strategy of this invention has achieved significant results in improving detection accuracy, enhancing rail structural stability, and reducing maintenance time. In particular, the invention demonstrates significant advantages over traditional technologies in damage identification and optimization in high-stress areas. These improvements not only enhance rail transportation safety but also provide more scientific and precise support for future railway maintenance and optimization.

[0141] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. The online intelligent identification method of rail damage based on edge computing is characterized by: The steps include: S1. Deploy a rail damage detection system through an edge computing platform to collect real-time sensor data from the rail surface under different working environments. A filtering algorithm is used to remove environmental noise and perform data normalization. Local data preprocessing is achieved through time window division, feature extraction, and support vector machine preliminary recognition. S2. On the edge computing node, the rail data collected by the sensor is analyzed in real time to identify whether there are obvious signs of damage on the rail surface and generate a preliminary damage judgment result; S3. Based on the preliminary judgment results, edge computing nodes are used to collect the rail's geometric characteristics, load distribution, environmental data, and historical damage information. An initial finite element model is constructed and the initial stress and strain are calculated. The rail structure is optimized by combining an adaptive hierarchical multi-scale topology optimization algorithm to minimize total energy and maximize stiffness. Considering damage tolerance, a simulated annealing algorithm is introduced for global search. The structure is optimized by combining an adaptive optimization mechanism and a hierarchical strategy. By dynamically adjusting the material distribution and structural morphology, the design is finally optimized through finite element simulation analysis. S4. Within the rail inspection area, improvements are made based on a generative adversarial network (GAN) by introducing a multimodal attention mechanism. The improved GAN simultaneously processes multimodal information from rail images and environmental data. The generator dynamically adjusts the attention weights of image regions, enhancing the repair accuracy, detail expression, and damage feature prominence of damaged areas. S5. Optimize the rail structure based on an adaptive hierarchical multi-scale topology optimization algorithm and annealing algorithm. Combined with an improved generative adversarial network, the computing power of edge computing nodes is utilized to process damage location and classification in real time, optimize recognition results, and conduct data interaction and model optimization through a cloud platform to achieve accuracy improvement and operation and maintenance feedback for the rail damage detection system.

2. The method for online intelligent identification of rail damage based on edge computing according to claim 1 is characterized in that: Said S1 specifically includes: S11. Deploy a rail damage detection system through edge computing nodes and collect rail surface data from different sensors in real time. S12, using a filtering algorithm to remove noise from the collected sensor data; S13. Normalize the denoised data to standardize data from different sources to the same scale S14. Divide the data into time windows through the edge computing node, and process the data in each time period as a subset; S15. Based on the working environment characteristics of the rail, use a statistical analysis method based on machine learning to extract features from each data subset; S16. Combine the extracted features and use the support vector machine algorithm to perform preliminary damage detection on the rail surface; S17. Implement local data preprocessing through the edge computing platform.

3. The method for online intelligent identification of rail damage based on edge computing according to claim 1 is characterized in that: The S3 specifically includes: S31, collect the geometric characteristics, load distribution, working environment data and historical damage information of the rail in real time through the edge computing node, build the initial finite element model, and for each discrete unit x of the rail i Calculate the initial stress based on its geometry, load distribution and environmental factors and initial strain And use this as the initial condition: Where C is the constitutive matrix of the material, f i For unit x i external loads; S32. Use the adaptive hierarchical multi-scale topology optimization algorithm to perform preliminary optimization of the rail structure. According to finite element analysis, the objective function of rail topology optimization is to minimize the total energy of the rail and maximize its stiffness, while also considering the damage tolerance of the rail. The objective function is: Among them, Y is the objective function, is the norm of the gradient of the stress field, ∈(x) is the strain field, α1, α2 are weighting coefficients, Ω is the calculation area, is the stress gradient, It is the process of minimizing the objective function; S33. Combine the simulated annealing algorithm with topology optimization to perform global search and local optimization. Set the initial temperature of the simulated annealing to T0. The temperature update formula during the annealing process is: Among them, δE k is the energy difference between the current solution and the candidate solution, γ is the temperature attenuation coefficient, and the simulated annealing algorithm decides whether to accept the new solution through a probabilistic selection mechanism: Among them, δE k is the energy difference between the current solution and the candidate solution, and min is the process of minimizing the objective function; S34. An adaptive hierarchical strategy is introduced to decompose the optimization problem into multiple optimization problems at different scales. At each scale, optimization is performed by refining the finite element mesh and adjusting local constraints. A regularization factor is introduced to maintain the stability and continuity of the structure. For each level of optimization problem, the objective function is: Among them, w i is the weight factor of the i-th level, α1 and α2 are the weight coefficients of the adjustment energy term, is the norm of the gradient of the stress distribution at the i-th level, ∈ i (x) is the strain distribution of the i-th level; σ i (x) is the stress distribution at the i-th level; S35. To cope with complex working environments and dynamic changes, an adaptive optimization mechanism is set up to update the optimization strategy based on real-time data. The optimization process of the rail minimizes losses by dynamically adjusting the material distribution and structural morphology. The adaptive update rules are as follows: in, For the energy change after adaptive update, is the time rate of change of the stress field, α3, α4, α5 are weighting coefficients; S36. Throughout the optimization process, a multi-scale approach is used for local and global collaborative optimization. For each level of optimization, the optimization process combines simulated annealing with an adaptive hierarchical strategy to conduct local exploration and global search, ultimately obtaining the global optimal solution. During each round of simulated annealing, the simulated annealing algorithm decides whether to accept a new solution based on the energy difference between the current rail structure optimization solution and the candidate solution. S37, through the finite element simulation analysis of the optimized rail structure, calculate the optimized stress field σ opt (x) and the strain field ∈ opt (x), and output the optimized rail structure design. At the same time, the rails are monitored in real time through edge computing nodes, and the detection area is dynamically adjusted according to the optimization results.

4. The method for online intelligent identification of rail damage based on edge computing according to claim 1 is characterized in that: The S4 specifically includes: S41. Based on the generative adversarial network framework, a generator network is used to enhance and repair images of damaged rail areas. The generator network combines a multimodal attention mechanism to optimize the repair accuracy of damaged area images, highlight the characteristics of damaged areas, and classify damage types through a multi-task learning architecture. The optimization objectives of the generator network are: L gen =λ1L adv +λ2L rec +λ3L reg +λ4L smooth ; Among them, L adv To combat the loss, L rec To reconstruct the loss, L reg is the regularization loss, L smooth is the smoothing loss, λ1,λ2,λ3,λ4 are weight factors; S42. The generator network adopts a multimodal attention mechanism and uses environmental data as additional input to perform feature fusion, so that the generator focuses on the damaged area of the rail. The contribution of different modal information is weighted by the attention mechanism to optimize the repair effect of the generated image and obtain an image of the damaged area. The optimization goal is: Among them, α i is the attention weight of the i-th modal data, G(x,z i ) is the repaired image generated based on the rail image and environmental data, G′(x,z i ) is the real rail damage area image, z i For environmental data; S43. While repairing the image, the generator network uses a multi-task learning architecture to simultaneously perform damage area repair and damage type classification tasks. The optimization goal of damage type classification is: Among them, C j is the real damage type label, C pred is the damage type label predicted by the classifier, and M is the number of categories; S44. The regularization part of the generator loss function introduces multi-scale constraints, combines the dynamic working environment data of the rail to adjust the details of the repaired image, and optimizes the detail performance and global structure of the image at different scales. The loss function is: Among them, G k (x,z) is the generated image at the kth scale, G′ k (x,z) is the real rail area image at the kth scale, β k is the scale weight, K is the number of scales; S45. Based on the above optimized generated images and classification results, a discriminator network of the adversarial network is generated to further evaluate the authenticity of the generated damage images. The generator and discriminator are jointly optimized through adversarial training to improve the repair accuracy and feature visibility of the rail damage area. The loss function of the discriminator is: L disc =logD(x)+log(1-D(G(x,z))); Where D(x) is the discriminator's prediction of the input image, and D(G(x,z)) is the discriminator's prediction of the image generated by the generator; S46. After the generated rail damage image is repaired and enhanced, it is further subjected to feature extraction through a convolutional neural network, and the damaged area is classified in combination with a classifier to generate the final recognition result of the rail damage.

Images