Railway vehicle abnormality early warning method and system based on infrared thermal imaging

Abstract

The application provides a railway vehicle abnormal early warning method and system based on infrared thermal imaging, first acquires an infrared thermal imaging data set of a railway vehicle, and constructs a thermal state benchmark feature set of each key component. Real-time collection of infrared thermal imaging data in operation is dynamically compared with the benchmark set to generate a thermal state deviation feature set. An infrared thermal state abnormality recognition model is called to associate and strengthen processing of the deviation feature set, and a strengthened and associated deviation feature vector is generated. The strengthened and associated deviation feature vector is matched and mapped with a preset abnormal thermal state feature library to determine an abnormal type and an associated component identifier. An early warning instruction containing component positioning information is generated based on the abnormal type and the associated component identifier, and is sent to a target monitoring terminal. The application can accurately and timely early warn thermal state abnormalities of a railway vehicle and ensure safe operation of the vehicle.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and more specifically, to a method and system for early warning of railway vehicle anomalies based on infrared thermal imaging. Background Technology

[0002] In the railway transportation sector, the safe and stable operation of railway vehicles is of paramount importance. Traditional railway vehicle fault detection methods mainly rely on regular manual inspections and monitoring methods based on signals such as mechanical vibration and sound. Manual inspections suffer from low efficiency and susceptibility to subjective factors, making it difficult to achieve real-time and comprehensive vehicle status monitoring. While monitoring methods based on mechanical vibration and sound signals can detect some mechanical faults, their effectiveness is limited for faults caused by thermal anomalies, such as bearing overheating or electrical component short circuits leading to abnormal temperature increases. Current technology lacks a method that can acquire real-time and accurate thermal information of key railway vehicle components and provide effective early warnings of thermal anomalies. Although some methods involve thermal imaging technology, they do not fully consider the complexity of the thermal characteristics of key components under different operating conditions of railway vehicles, as well as the mutual influence of thermal states between components. This results in insufficient accuracy and timeliness of anomaly warnings, failing to meet the high requirements of railway transportation for vehicle safety. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for early warning of railway vehicle anomalies based on infrared thermal imaging, the method comprising:

[0004] Acquire infrared thermal imaging data sets of railway vehicles and construct thermal reference feature sets of key components of railway vehicles;

[0005] Real-time infrared thermal imaging data of railway vehicles during operation is collected, and the real-time infrared thermal imaging data is dynamically compared with the thermal reference feature set to generate a thermal deviation feature set.

[0006] The infrared thermal anomaly identification model is invoked to perform correlation enhancement processing on the thermal deviation feature set, generating a strengthened correlation deviation feature vector;

[0007] The enhanced correlation deviation feature vector is matched and mapped with the preset abnormal hot state feature library to determine the abnormal type of railway vehicle and the identifier of the associated component.

[0008] Based on the anomaly type and associated component identifier, a railway vehicle anomaly warning command containing component location information is generated and sent to the target monitoring terminal.

[0009] In another aspect, embodiments of the present invention also provide a railway vehicle anomaly early warning system based on infrared thermal imaging, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0010] Based on the above, this invention constructs a thermal baseline feature set for each key component of a railway vehicle. This allows for the real-time acquisition and dynamic comparison of infrared thermal imaging data, enabling the rapid generation of a thermal deviation feature set and timely capture of thermal changes during vehicle operation. The infrared thermal anomaly identification model is invoked for correlation enhancement processing, fully considering the mutual influence of thermal states between components, generating a strengthened correlation deviation feature vector, thus improving the accuracy of anomaly identification. Matching and mapping with a preset anomaly thermal feature library accurately determines the anomaly type and associated component identifier. Finally, an early warning command containing component location information is generated and sent to the target monitoring terminal, achieving real-time and accurate early warning of thermal anomalies in railway vehicles. This effectively improves the safety and reliability of railway vehicle operation and ensures smooth railway transportation. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the execution flow of the railway vehicle anomaly early warning method based on infrared thermal imaging provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of exemplary hardware and software components of a railway vehicle anomaly early warning system based on infrared thermal imaging provided in an embodiment of the present invention. Detailed Implementation

[0013] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a railway vehicle anomaly early warning method based on infrared thermal imaging, according to an embodiment of the present invention. The following is a detailed description of this railway vehicle anomaly early warning method based on infrared thermal imaging.

[0014] Step S110: Obtain the infrared thermal imaging data set of railway vehicles and construct the thermal reference feature set of each key component of the railway vehicles.

[0015] In this embodiment, to construct a set of thermal baseline features for each key component of a railway vehicle, it is first necessary to acquire an infrared thermal imaging data set of the railway vehicle. Specifically, infrared thermal imaging equipment deployed at specific locations along the railway line is used to collect infrared thermal imaging data sets of railway vehicles under different operating conditions. These different operating conditions encompass the operating status of railway vehicles under different loads (e.g., empty, half-loaded, fully loaded), different operating speeds (e.g., low speed, medium speed, high speed), and different track gradients (e.g., flat slope, uphill, downhill). The infrared thermal imaging data set contains continuous thermal imaging frames of each key component (e.g., wheelsets, bearings, traction motors, braking devices, etc.) under normal operating conditions. These continuous thermal imaging frames are acquired at certain time intervals (e.g., every 0.1 seconds), forming a data set containing multiple frames of thermal imaging images.

[0016] Step S111: Collect infrared thermal imaging data sets of railway vehicles under different operating conditions. The infrared thermal imaging data sets include continuous thermal imaging frames of each key component under normal operating conditions.

[0017] Data is collected for various operating conditions of railway vehicles. For example, for the wheelset, a critical component, the friction between the wheelset and the track generates varying amounts of heat as the railway vehicle travels at different speeds. Infrared thermal imaging equipment captures thermal images of the wheelset at these different speeds. Similarly, for bearing components, the load on the bearing varies under different load conditions, resulting in different heat generation. Infrared thermal imaging equipment records thermal images of the bearing under these different load conditions. The collected infrared thermal imaging data is stored in a specific database.

[0018] Step S112: Perform key component region localization processing on continuous thermal imaging frames in the infrared thermal imaging dataset, extract the thermal imaging sub-regions corresponding to each key component, and maintain the correspondence between the thermal imaging sub-regions and the physical structural boundaries of the key components.

[0019] Image recognition algorithms are used to locate key component regions in continuous thermal imaging frames from an infrared thermal imaging dataset. Taking wheelsets as an example, wheelsets have specific physical structural shapes. The image recognition algorithm identifies the position of the wheelset in the thermal imaging frame and extracts the corresponding thermal imaging sub-region based on the physical structural boundaries of the wheelset. This thermal imaging sub-region accurately corresponds to the physical structure of the wheelset; for example, the hub, spokes, and rim of the wheelset are all included within the thermal imaging sub-region. For bearing components, the corresponding thermal imaging sub-region is also extracted based on their physical structural features, such as the outer ring, inner ring, and rolling elements of the bearing.

[0020] Step S113: Extract the thermal distribution features of each thermal imaging sub-region. The thermal distribution features include the distribution pattern of thermal radiation intensity and the trend of thermal gradient change within the thermal imaging sub-region.

[0021] For example, in step S1131: perform pixel division processing on each thermal imaging sub-region to divide the thermal imaging sub-region into multiple uniform pixel blocks.

[0022] For example, for the thermal imaging sub-region of a wheelset, it is divided into multiple pixel blocks of the same size. The size of each pixel block is set according to the size of the thermal imaging sub-region and the requirements of subsequent processing, so as to ensure that each pixel block can accurately reflect the thermal radiation of the region.

[0023] Step S1132: Extract the thermal radiation intensity value of each pixel block, statistically analyze the distribution of thermal radiation intensity values ​​of all pixel blocks within the thermal imaging sub-region, and generate a thermal radiation intensity distribution histogram.

[0024] When extracting the thermal radiation intensity value of each pixel block, the grayscale value of the pixel block is converted into a corresponding thermal radiation intensity value according to the principle of infrared thermal imaging. Then, these thermal radiation intensity values ​​are statistically analyzed, for example, by counting the frequency of different intensity values, to generate a thermal radiation intensity distribution histogram. Taking the thermal imaging sub-region of a wheelset as an example, the thermal radiation intensity distribution histogram shows the distribution of thermal radiation intensity in different areas of the wheelset; for example, the thermal radiation intensity of the wheel hub may be relatively high, while the thermal radiation intensity of the wheel rim may be relatively low.

[0025] Step S1133: Analyze the concentration range and dispersion of thermal radiation intensity based on the thermal radiation intensity distribution histogram to determine the thermal radiation intensity distribution pattern within the thermal imaging sub-region.

[0026] By analyzing the histogram of thermal radiation intensity distribution, the concentrated range of thermal radiation intensity is determined. For example, the thermal radiation intensity of most pixel blocks is concentrated within a specific interval. Simultaneously, the dispersion of thermal radiation intensity is analyzed, i.e., whether the distribution of thermal radiation intensity values ​​is uniform. Taking the thermal imaging sub-region of a bearing as an example, if the bearing is operating normally, its thermal radiation intensity distribution is relatively concentrated and uniform; if the bearing malfunctions, the thermal radiation intensity distribution may become uneven, for example, the thermal radiation intensity in one area may be significantly higher than in other areas.

[0027] Step S1134: Calculate the difference in thermal radiation intensity between adjacent pixel blocks and generate a thermal gradient matrix.

[0028] Adjacent pixel blocks are compared in a certain direction (e.g., horizontal, vertical, or diagonal) to calculate the difference in thermal radiation intensity between them. For example, for two adjacent pixel blocks in the horizontal direction, the thermal radiation intensity value of the right pixel block is subtracted from the thermal radiation intensity value of the left pixel block to obtain a difference. These differences are then arranged according to the position of the pixel blocks to generate a thermal gradient matrix.

[0029] Step S1135: Perform directional analysis on the thermal gradient matrix to identify the main directions and magnitudes of change in the thermal gradient.

[0030] A matrix analysis method is used to calculate the variation of the thermal gradient matrix in different directions. For example, by calculating the eigenvectors of the thermal gradient matrix, the main direction of change of the thermal gradient is determined. Simultaneously, by calculating the absolute values ​​of the elements in the thermal gradient matrix, the magnitude of the thermal gradient variation is determined. Taking the thermal imaging sub-region of a traction motor as an example, the traction motor generates heat during operation, and the main direction of change of its thermal gradient may be related to the motor's heat dissipation channels, while the magnitude of the variation reflects the temperature differences between different areas of the motor.

[0031] Step S1136: Based on the main direction and magnitude of change of the thermal gradient, summarize the trend of thermal gradient change within the thermal imaging sub-region.

[0032] Based on the main direction and magnitude of change of the thermal gradient, summarize the trend of thermal gradient change within the thermal imaging sub-region. For example, if the main direction of change of the thermal gradient is from the motor windings to the outer casing, and the magnitude of change is large, it indicates that the temperature of the motor windings is higher, and heat is being transferred to the outer casing.

[0033] Step S1137: Combine the distribution pattern of thermal radiation intensity and the trend of thermal gradient change to form the thermal distribution characteristics of each thermal imaging sub-region.

[0034] By combining the distribution patterns of thermal radiation intensity and the trends in thermal gradient changes, the thermal distribution characteristics of each thermal imaging sub-region are obtained. These thermal distribution characteristics are stored in a structured data format, such as a vector containing histogram data of thermal radiation intensity distribution and thermal gradient trend data.

[0035] Step S114: Perform statistical aggregation on the thermal distribution characteristics of the same key component under different operating conditions to generate the thermal characteristic baseline of the corresponding key component. The thermal characteristic baseline reflects the thermal fluctuation of the key component within the normal operating range.

[0036] For each key component, its thermal distribution characteristics under different operating conditions are statistically aggregated. For example, for the wheelset, a key component, its thermal distribution characteristics under different loads, speeds, and track gradients are collected. Then, statistical analysis methods are used, such as calculating the mean and standard deviation of these thermal distribution characteristics, to generate a thermal characteristic baseline for the wheelset. The thermal characteristic baseline reflects the thermal fluctuations of the wheelset within its normal operating range, such as the fluctuation range of the wheelset's heat radiation intensity distribution and the range of changes in the thermal gradient trend.

[0037] Step S115: Summarize the thermal characteristic baselines of all key components and construct a set of thermal benchmark features for each key component of the railway vehicle that corresponds one-to-one with the component identifier and thermal characteristic baseline.

[0038] The thermal characteristic baselines of all key components (such as wheelsets, bearings, traction motors, and braking systems) are compiled to construct a thermal baseline feature set. This thermal baseline feature set is indexed by component identifiers, corresponding to the thermal characteristic baseline of each key component. For example, the component identifier of a wheelset is "Wheelset-001," and its corresponding thermal characteristic baseline contains statistical information on the thermal distribution characteristics of the wheelset under different operating conditions; the component identifier of a bearing is "Bearing-001," and its corresponding thermal characteristic baseline also contains statistical information on the thermal distribution characteristics of the bearing under different operating conditions.

[0039] Step S116: Perform feature dimension unification processing on each thermal feature baseline in the thermal reference feature set so that the thermal feature baselines of different key components have the same feature representation form.

[0040] Since the thermal distribution characteristics of different key components may have different dimensions and representations, to facilitate subsequent comparison processing, the feature dimensions of each thermal feature baseline in the thermal baseline feature set are unified. For example, the thermal feature baseline of a wheelset contains 100 feature dimensions, while the thermal feature baseline of a bearing contains 80 feature dimensions. In this case, feature mapping or feature transformation methods are used to unify their feature dimensions to the same number (e.g., 100) and keep their feature representations (such as data type, numerical range, etc.) consistent.

[0041] Step S120: Collect real-time infrared thermal imaging data of railway vehicles in operation, dynamically compare the real-time infrared thermal imaging data with the thermal reference feature set, and generate a thermal deviation feature set.

[0042] During the operation of railway vehicles, real-time infrared thermal imaging data is collected. Real-time collection here refers to using the same infrared thermal imaging equipment as in step S110 to collect real-time infrared thermal imaging data of the railway vehicles at certain time intervals (e.g., every 0.05 seconds). The real-time infrared thermal imaging data includes a sequence of real-time thermal imaging frames of various key components during operation.

[0043] Step S121: Real-time infrared thermal imaging data of railway vehicles in operation is captured by infrared thermal imaging acquisition equipment deployed along the railway line. The real-time infrared thermal imaging data includes real-time thermal imaging frame sequences of key components in operation.

[0044] Infrared thermal imaging acquisition equipment deployed along railway lines continuously captures real-time infrared thermal imaging data of railway vehicles in operation. For example, when a railway vehicle reaches a specific monitoring point, the infrared thermal imaging equipment immediately captures a sequence of real-time thermal imaging frames of key components of the vehicle. These real-time thermal imaging frame sequences are transmitted to the data processing center in real time.

[0045] Step S122: Perform frame synchronization processing on the real-time thermal imaging frame sequence in the real-time infrared thermal imaging data to keep each real-time thermal imaging frame consistent in the time dimension.

[0046] Because railway vehicles are in motion, infrared thermal imaging devices at different locations may capture thermal images of the same vehicle at different times. Therefore, frame synchronization processing is performed on the real-time thermal imaging frame sequence. Frame synchronization processing calculates the timestamp corresponding to each thermal imaging frame based on the railway vehicle's speed and the location information of the infrared thermal imaging devices. Then, the real-time thermal imaging frame sequence is adjusted according to the timestamps to ensure that each real-time thermal imaging frame remains consistent in the time dimension. For example, assuming the railway vehicle's speed is v and the distance between infrared thermal imaging devices A and B is d, then the timestamp of the thermal imaging frame captured by device B is d / v later than the timestamp of the thermal imaging frame captured by device A. Frame synchronization is achieved by adjusting the timestamps in this way.

[0047] Step S123: For each real-time thermal imaging frame, extract the real-time thermal imaging sub-regions corresponding to each key component according to the positioning rules of the key components in the thermal reference feature set.

[0048] Using the same image recognition algorithm and key component localization rules as in step S112, real-time thermal imaging sub-regions corresponding to each key component are extracted for each real-time thermal imaging frame. For example, for the wheelset component in the real-time thermal imaging frame, its position in the thermal imaging frame is identified based on the physical structural characteristics of the wheelset, and the corresponding real-time thermal imaging sub-region is extracted. The size and shape of this real-time thermal imaging sub-region are consistent with the thermal imaging sub-region extracted in step S112.

[0049] Step S124: Extract the real-time thermal distribution features of each real-time thermal imaging sub-region. The feature dimension of the real-time thermal distribution features is consistent with the feature dimension of the corresponding thermal feature baseline in the thermal baseline feature set.

[0050] Using the same method as in step S113, the real-time thermal distribution features of each real-time thermal imaging sub-region are extracted. For example, for the real-time thermal imaging sub-region of the wheelset, it is first divided into pixels, then the thermal radiation intensity value of each pixel block is extracted, the thermal radiation intensity distribution is statistically analyzed, a thermal radiation intensity distribution histogram is generated, the concentration range and dispersion of thermal radiation intensity are analyzed, and the distribution law of thermal radiation intensity is determined. Next, the thermal radiation intensity difference between adjacent pixel blocks is calculated to generate a thermal gradient matrix. The thermal gradient matrix is ​​analyzed for direction to identify the main change direction and magnitude of the thermal gradient, and the trend of thermal gradient change is summarized. Finally, the distribution law of thermal radiation intensity and the trend of thermal gradient change are combined to form the real-time thermal distribution features of the wheelset. Furthermore, the feature dimension of the real-time thermal distribution features is consistent with the feature dimension of the corresponding thermal feature baseline in the thermal baseline feature set.

[0051] Step S125: Compare the real-time thermal distribution characteristics of each key component with the thermal characteristic baseline of the corresponding key component in the thermal benchmark feature set dimension by dimension to identify the feature dimensions that are not in the thermal fluctuation range.

[0052] For each key component, its real-time thermal distribution characteristics are compared dimension-by-dimensionally with the corresponding key component's thermal characteristic baseline in the thermal baseline feature set. For example, for the feature dimension of thermal radiation intensity distribution in the real-time thermal distribution characteristics of the wheelset, it is compared with the statistical range of thermal radiation intensity distribution in the thermal characteristic baseline. If the real-time thermal radiation intensity distribution exceeds the thermal fluctuation range specified in the thermal characteristic baseline, then this feature dimension is identified as a feature dimension that is not within the thermal fluctuation range. Similarly, a similar comparison is performed for the feature dimension of thermal gradient change trend.

[0053] Step S126: Summarize the feature dimensions of all key components that are not in the thermal fluctuation range, and generate a thermal deviation feature set containing component identifier, deviation feature dimension, and deviation direction.

[0054] The feature dimensions of all key components that are not within the thermal fluctuation range are summarized to generate a thermal deviation feature set. This set includes component identification, deviation feature dimension, and deviation direction. For example, a wheelset component is identified as "Wheelset-001," its deviation feature dimension is the distribution pattern of thermal radiation intensity, and its deviation direction is that the thermal radiation intensity exceeds the upper limit of the thermal fluctuation range; therefore, this information is recorded in the thermal deviation feature set. Similarly, a bearing component is identified as "Bearing-001," its deviation feature dimension is the trend of thermal gradient change, and its deviation direction is that the magnitude of the thermal gradient change exceeds the upper limit of the thermal fluctuation range; this information is also recorded in the thermal deviation feature set.

[0055] Step S130: Call the infrared thermal anomaly recognition model to perform correlation enhancement processing on the thermal deviation feature set, and generate the enhanced correlation deviation feature vector.

[0056] To enhance the association of the thermal deviation feature set, an infrared thermal anomaly recognition model is invoked. This model is a pre-trained artificial intelligence model that includes modules such as a feature association layer and a feature enhancement layer. The feature association layer is used to model the association of thermal deviation features within the component's associated feature group, while the feature enhancement layer is used to perform dimensional expansion and redundancy removal on the features.

[0057] Step S131: Group the thermal deviation feature set according to the association relationship of key components to generate multiple component association feature groups. Each component association feature group contains the thermal deviation features of multiple key components that interact functionally.

[0058] Based on the mechanical structure principles of railway vehicles and the functional relationships between key components, the key components in the thermal deviation feature set are grouped to generate multiple component-related feature groups. For example, wheelsets, bearings, and braking devices interact functionally; the rotation of wheelsets requires the support of bearings, and the braking device brakes the rotation of wheelsets. Therefore, their thermal deviation features are grouped into the same component-related feature group. Similarly, there is a functional relationship between the traction motor and the wheelset; the traction motor drives the wheelset to rotate, so their thermal deviation features are also grouped into the same component-related feature group. Each component-related feature group contains the thermal deviation features of multiple functionally interacting key components.

[0059] Step S132: Input the associated feature groups of each component into the feature association layer of the infrared thermal anomaly identification model, and generate the thermal association matrix within the associated feature groups of the components by modeling the thermal influence relationship between components.

[0060] Step S1321: Perform feature encoding on the thermal deviation features of each key component in the component association feature group to generate a standardized component deviation feature vector.

[0061] The feature encoding method is used to convert the thermal deviation features of each key component in the component association feature group into a standardized component deviation feature vector. For example, for the thermal deviation features of wheelsets, the feature dimensions such as the distribution law of thermal radiation intensity and the trend of thermal gradient change are encoded and converted into a fixed-length vector. The value of this vector is standardized so that it is within a specific numerical range (e.g., between 0 and 1).

[0062] Step S1322: Based on the mechanical structure principle of railway vehicles, construct a prior knowledge graph of thermal influence between components. The prior knowledge graph includes the heat conduction path and the degree of thermal influence between key components.

[0063] Based on the mechanical structure principles of railway vehicles, this study analyzes the heat conduction paths and the degree of thermal impact between key components, constructing a priori knowledge graph of the thermal effects between components. For example, there is a heat conduction path between wheelsets and bearings; heat from the wheelsets is transferred to the bearings through contact. Therefore, the knowledge graph records the heat conduction path between the wheelsets and bearings, as well as the degree of impact of the thermal changes of the wheelsets on the thermal state of the bearings (e.g., how much the bearing temperature will rise if the wheelset temperature increases). Similarly, there is a thermal impact relationship between the traction motor and the wheelsets; heat from the traction motor is transferred to the wheelsets through the transmission system. These relationships are also recorded in the knowledge graph.

[0064] Step S1323: Input the standardized component deviation feature vector and the prior knowledge graph into the association modeling unit of the feature association layer, and learn the dynamic thermal association relationship between components through the graph neural network algorithm.

[0065] Standardized component deviation feature vectors and prior knowledge graphs are input into the association modeling unit of the feature association layer of the infrared thermal anomaly recognition model. The association modeling unit employs a graph neural network algorithm, based on the prior knowledge graph, to learn the dynamic thermal relationships between components. The graph neural network algorithm, based on the prior knowledge graph, continuously adjusts the model parameters according to changes in the standardized component deviation feature vectors and the heat conduction paths and thermal impact levels recorded in the prior knowledge graph, to accurately learn the dynamic thermal relationships between components. For example, when the thermal deviation feature vector of a wheelset changes, the graph neural network algorithm analyzes the impact of this change on the bearing's thermal state based on the heat conduction relationship between the wheelset and the bearing in the prior knowledge graph, and learns this dynamic relationship.

[0066] Step S1324: Calculate the thermal correlation coefficient of any two key components within each component association feature group. The thermal correlation coefficient reflects the degree of interaction between the thermal deviation features of the two key components.

[0067] Based on the dynamic thermal correlations between components learned by the graph neural network algorithm, the thermal correlation coefficient between any two key components within each component's correlation feature group is calculated. The calculation of the thermal correlation coefficient considers the changes in the thermal deviation features of the two key components, as well as the heat conduction path and the degree of thermal influence between them. For example, for the two key components, wheelsets and bearings, a thermal correlation coefficient is calculated based on the changes in their thermal deviation feature vectors and the heat conduction relationships recorded in the prior knowledge graph. The larger this coefficient, the stronger the interaction between the thermal deviation features of the wheelset and the bearing.

[0068] Step S1325: Construct a thermal correlation matrix within the component correlation feature group based on all thermal correlation coefficients. The rows and columns of the thermal correlation matrix correspond to the key components in the component correlation feature group, and the matrix elements are the corresponding thermal correlation coefficients.

[0069] All thermal correlation coefficients within each component association feature group are organized to construct a thermal correlation matrix. The rows and columns of the thermal correlation matrix correspond to the key components within the component association feature group, and each element in the matrix is ​​the thermal correlation coefficient between two corresponding key components. For example, if a component association feature group contains three key components: wheelset, bearing, and braking device, then the rows and columns of the thermal correlation matrix represent these three components. Matrix elements (wheelset, bearing) correspond to the thermal correlation coefficient between the wheelset and bearing, (wheelset, braking device) corresponds to the thermal correlation coefficient between the wheelset and braking device, and so on.

[0070] Step S133: Based on the thermal correlation matrix, perform weighted fusion processing on each thermal deviation feature in the component correlation feature group to enhance the collaborative attributes of the deviation features of interrelated components.

[0071] Based on the thermal correlation coefficients in the thermal correlation matrix, a weighted fusion process is performed on the thermal deviation features within each component correlation feature group. The weighted fusion process assigns a weight to the thermal deviation feature of each key component based on the magnitude of the thermal correlation coefficient. These weighted thermal deviation features are then fused to strengthen the synergistic properties of the deviation features of interconnected components. For example, in the component correlation feature group of wheelsets, bearings, and braking devices, the thermal correlation coefficient between wheelsets and bearings is relatively large, indicating strong synergy in their thermal deviation features. Therefore, during weighted fusion, higher weights are assigned to the thermal deviation features of wheelsets and bearings. These thermal deviation features are then fused with the thermal deviation features of the braking device to obtain a fused thermal deviation feature, which further strengthens the synergistic properties of the deviation features of wheelsets and bearings.

[0072] Step S134: Input the weighted fusion of the associated feature groups of each component into the feature enhancement layer of the infrared thermal anomaly recognition model, and expand the dimension of the features through nonlinear transformation to generate multi-dimensional associated features.

[0073] The weighted and fused feature sets of associated components are input into the feature enhancement layer of the infrared thermal anomaly recognition model. The feature enhancement layer expands the dimensionality of the input features through nonlinear transformations, generating multi-dimensional associated features. Nonlinear transformations can be performed in various ways, such as using a multilayer perceptron (MLP) structure to perform multiple nonlinear mappings on the input features, increasing the feature dimensionality. For example, if the feature dimension of the weighted and fused component-associated feature set is 100, after the nonlinear transformation by the feature enhancement layer, the feature dimension expands to 200. The generated multi-dimensional associated features contain more information, enabling more accurate anomaly identification.

[0074] Step S135: Perform feature redundancy removal processing on multi-dimensional related features, and retain the core feature dimensions related to anomaly identification.

[0075] Step S1351: Use a feature importance assessment algorithm to score the importance of each feature dimension of the multi-dimensional related features. The importance score reflects the impact of the corresponding feature dimension on the anomaly identification result.

[0076] Feature importance evaluation algorithms, such as the feature importance evaluation method in the random forest algorithm, are used to score the importance of each feature dimension in a multi-dimensional correlated feature set. For example, for each feature dimension in the multi-dimensional correlated feature set, its role in the anomaly identification process is analyzed, and its influence on the anomaly identification result is calculated to obtain an importance score. The higher the importance score, the greater the influence of that feature dimension on the anomaly identification result.

[0077] Step S1352: Set a feature importance threshold and select feature dimensions whose importance scores meet the requirements of the feature importance threshold as candidate core feature dimensions.

[0078] Based on the needs and experience of anomaly identification, a feature importance threshold is set. Then, feature dimensions with an importance score greater than or equal to this threshold are selected as candidate core feature dimensions. For example, if the feature importance threshold is set to 0.6, then feature dimensions with an importance score greater than or equal to 0.6 are selected as candidate core feature dimensions.

[0079] Step S1353: Calculate the pairwise correlations between candidate core feature dimensions and identify sets of feature dimensions that are mutually related.

[0080] Calculate the pairwise correlations between candidate core feature dimensions, for example, using the Pearson correlation coefficient method to calculate the correlation between every two candidate core feature dimensions. If the correlation between two feature dimensions is high (e.g., the correlation coefficient is greater than 0.8), it indicates that they are mutually related, and these two feature dimensions are grouped into one feature dimension set.

[0081] Step S1354: Perform feature fusion processing on each set of interrelated feature dimensions to generate a comprehensive feature dimension, which replaces the original feature dimension in the corresponding feature dimension set.

[0082] For each set of interrelated feature dimensions, feature fusion is performed. Principal Component Analysis (PCA) can be used to combine multiple feature dimensions into a single composite feature dimension. For example, for a set containing two interrelated feature dimensions, PCA can be used to merge them into a single composite feature dimension that contains the main information from both original feature dimensions.

[0083] Step S1355: Summarize the unfused feature dimensions and the generated comprehensive feature dimensions from the candidate core feature dimensions to form the final core feature dimensions.

[0084] The unfused feature dimensions and the generated composite feature dimensions from the candidate core feature dimensions are combined to form the final core feature dimensions. For example, if there are 10 candidate core feature dimensions, and 2 feature dimension sets are fused to generate 2 composite feature dimensions, then there are 10 - 2 × 2 = 6 unfused feature dimensions, and the final number of core feature dimensions is 6 + 2 = 8.

[0085] Step S1356: Project the multi-dimensional associated features onto the core feature dimension to obtain the core feature vector after removing redundancy.

[0086] The multi-dimensional associated features are projected according to the final core feature dimensions to obtain a core feature vector after redundancy removal. The projection process, based on the definition of core feature dimensions, maps each feature value in the multi-dimensional associated features to its corresponding core feature dimension, resulting in a new core feature vector. The dimension of this vector is the number of the final core feature dimensions.

[0087] Step S136: Concatenate the core feature dimensions corresponding to all component-associated feature groups in a preset order to generate the enhanced correlation deviation feature vector.

[0088] The core feature dimensions corresponding to all component-related feature groups are concatenated in a preset order (e.g., according to the importance of the component-related feature groups, or according to the functional order of key components) to generate a deviated feature vector with enhanced association. For example, if the core feature dimension of the first component-related feature group is vector A, and the core feature dimension of the second component-related feature group is vector B, with the preset order being A first and B last, then the concatenated deviated feature vector is a new vector formed by connecting vectors A and B in sequence. This vector contains the core feature information of all component-related feature groups and has undergone association enhancement processing.

[0089] Step S140: Match and map the enhanced correlation deviation feature vector with the preset abnormal hot state feature library to determine the abnormal type of railway vehicle and the identifier of the associated component.

[0090] The pre-built abnormal thermal feature library is a pre-built database that contains standard abnormal feature vectors and associated component identifier sets corresponding to various known abnormal types (such as bearing overheating, wheelset wear, traction motor failure, brake device jamming, etc.).

[0091] Step S141: Obtain a preset abnormal hot state feature library. The abnormal hot state feature library contains standard abnormal feature vectors corresponding to various known abnormal types and associated component identifier sets.

[0092] A pre-defined abnormal thermal feature library is retrieved from a specific database. This library is constructed through the analysis and summarization of numerous railway vehicle anomaly cases. For example, for the anomaly type of bearing overheating, thermal feature data from multiple bearing overheating cases are collected, processed, and a corresponding standard anomaly feature vector is generated. A set of component identifiers associated with this anomaly type (e.g., bearing component identifiers) is also determined. Similarly, for the anomaly type of wheelset uneven wear, a corresponding standard anomaly feature vector and an associated set of component identifiers are generated.

[0093] Step S142: Calculate the feature similarity between the deviated feature vector after enhanced association and each standard abnormal feature vector in the abnormal hot feature library.

[0094] Step S1421: Perform feature space alignment processing on the deviated feature vectors after enhanced association and each standard abnormal feature vector, so that the two are in the same feature space.

[0095] A feature space alignment method is used to transform the deviant feature vectors after enhanced association and each standard anomaly feature vector into the same feature space. Feature space alignment can be achieved through features normalization, feature standardization, etc. For example, the deviant feature vectors and standard anomaly feature vectors after enhanced association are standardized so that their mean is 0 and their standard deviation is 1, thus placing them in the same feature space.

[0096] Step S1422: Calculate the feature magnitude of the deviation feature vector after enhanced association and each standard abnormal feature vector.

[0097] Calculate the eigenvalues ​​of the deviated eigenvectors after enhanced association and each standard anomalous eigenvector. The eigenvalues ​​are calculated using the vector's dimension and the value of each dimension, following the formula for calculating vector magnitude. For example, the magnitude of the deviated eigenvector after enhanced association is L1, and the magnitude of a standard anomalous eigenvector is L2.

[0098] Step S1423: Calculate the dot product of the deviated feature vector after enhanced association with each standard abnormal feature vector.

[0099] Calculate the dot product between the enhanced deviation feature vector and each standard anomaly feature vector. The dot product is calculated using the vector dot product formula, based on the vector's dimensions and the value of each dimension. For example, the dot product of the enhanced deviation feature vector and a certain standard anomaly feature vector is P.

[0100] Step S1424: Calculate the cosine similarity value between the two based on the feature magnitude and the dot product result.

[0101] Based on the feature magnitude and the dot product result, the cosine similarity value between the enhanced correlation deviation feature vector and each standard anomalous feature vector is calculated. The formula for calculating the cosine similarity value is: Cosine similarity value = Dot product result / (Feature magnitude L1 × Feature magnitude L2). For example, the cosine similarity value between the enhanced correlation deviation feature vector and a certain standard anomalous feature vector is cosθ = P / (L1 × L2).

[0102] Step S1425: Normalize the cosine similarity values ​​to obtain feature similarity values ​​within a fixed interval.

[0103] The cosine similarity values ​​are normalized to ensure that the feature similarity falls within a fixed range (e.g., between 0 and 1). Normalization can be achieved using min-max normalization, converting the cosine similarity values ​​into values ​​between 0 and 1. For example, assuming the cosine similarity value range is [-1, 1], min-max normalization converts it to values ​​between [0, 1], yielding the feature similarity.

[0104] Step S1426: Record the feature similarity corresponding to each standard anomaly feature vector to form a feature similarity list.

[0105] The feature similarity scores for each standard anomaly feature vector are recorded to form a feature similarity list. This list includes the identifier of each standard anomaly feature vector and its corresponding feature similarity score.

[0106] Step S143: Select standard abnormal feature vectors whose feature similarity meets the preset conditions as candidate abnormal feature vectors.

[0107] Set a preset condition (e.g., feature similarity greater than or equal to 0.7), and then filter standard anomalous feature vectors whose feature similarity meets the preset condition from the feature similarity list as candidate anomalous feature vectors. For example, if there are 10 standard anomalous feature vectors in the feature similarity list, and 5 of them have a feature similarity greater than or equal to 0.7, then these 5 standard anomalous feature vectors are selected as candidate anomalous feature vectors.

[0108] Step S144: If there are multiple candidate abnormal feature vectors, sort the candidate abnormal feature vectors based on the numerical value of feature similarity, and select the candidate abnormal feature vector whose feature similarity meets the preset priority condition as the target abnormal feature vector.

[0109] If multiple candidate anomaly feature vectors exist, they are sorted according to their feature similarity scores, with higher similarity scores ranking higher. Then, a preset priority condition is set (e.g., selecting the top 3 candidate anomaly feature vectors, or selecting the candidate anomaly feature vector with the highest feature similarity). The candidate anomaly feature vector that meets this preset priority condition is selected as the target anomaly feature vector. For example, if the preset priority condition is to select the candidate anomaly feature vector with the highest feature similarity, then the one with the highest feature similarity score among the multiple candidate anomaly feature vectors is selected as the target anomaly feature vector.

[0110] Step S145: Extract the anomaly type and associated component identifier set corresponding to the target anomaly feature vector.

[0111] Based on the records of the target anomaly feature vector in the anomaly thermal feature database, the corresponding anomaly type and the associated set of component identifiers are extracted. For example, if the anomaly type corresponding to the target anomaly feature vector is bearing overheating, the associated set of component identifiers is the bearing component identifiers.

[0112] Step S146: Combining the deviation degree of each key component in the thermal deviation feature set, determine the core associated component identifier from the associated component identifier set, and finally obtain the abnormal type and associated component identifier of the railway vehicle.

[0113] By combining the deviation degrees of each key component in the thermal deviation feature set, the component identifiers in the associated component identifier set are analyzed. The greater the deviation degree of a key component, the more likely it is to be a core associated component. For example, in the associated component identifier set, if the bearing's deviation degree (the degree to which thermal radiation intensity exceeds the upper limit of the thermal fluctuation range, the degree to which the thermal gradient change amplitude exceeds the upper limit of the thermal fluctuation range, etc.) is greater than that of other components, then the bearing's component identifier is determined as a core associated component identifier. Finally, the anomaly type and the core associated component identifier are combined to obtain the anomaly type and associated component identifiers of the railway vehicle.

[0114] Step S150: Generate a railway vehicle anomaly warning instruction containing component location information based on the anomaly type and associated component identifier, and send the railway vehicle anomaly warning instruction to the target monitoring terminal.

[0115] After determining the type of anomaly in the railway vehicle and the identification of the associated components, an early warning command for the railway vehicle anomaly containing component location information is generated and sent to the target monitoring terminal so that relevant personnel can take timely measures.

[0116] Step S151: Analyze the warning level rules corresponding to the anomaly type and determine the warning level corresponding to the anomaly type.

[0117] A pre-defined rule connects anomaly types with warning levels. For example, bearing overheating corresponds to a Level 1 warning (the highest level), while wheelset uneven wear corresponds to a Level 2 warning. Once an anomaly type is determined, the corresponding warning level is calculated based on this rule. For instance, if the anomaly type is bearing overheating, the corresponding warning level is Level 1.

[0118] Step S152: Query the preset railway vehicle component positioning database according to the associated component identifier to obtain the component positioning information corresponding to the associated component identifier. The component positioning information includes a description of the specific location of the component on the railway vehicle and its positioning coordinates.

[0119] A pre-defined railway vehicle component location database is established. This database stores the component location information corresponding to each component identifier, including a description of the component's specific location on the railway vehicle (e.g., wheelsets are located at the bottom of the railway vehicle, near both ends of the carriage; bearings are mounted on the axle journals of the wheelsets) and location coordinates (e.g., the location coordinates of the wheelset are (x1, y1, z1), and the location coordinates of the bearing are (x2, y2, z2)). Once the associated component identifier is obtained, this database is queried to retrieve the corresponding component location information.

[0120] Step S153: Extract keyframe images of the real-time thermal imaging sub-regions corresponding to the associated components from the real-time infrared thermal imaging data.

[0121] Keyframe images of the corresponding real-time thermal imaging sub-regions of the associated components are extracted from real-time infrared thermal imaging data. The extraction of keyframe images is based on the time point of the anomaly occurrence and the acquisition time of the thermal imaging frames, selecting the thermal imaging frame that best reflects the anomaly as the keyframe image. For example, after determining that the anomaly is bearing overheating, the thermal imaging frames of the corresponding real-time thermal imaging sub-region of the bearing near the time of the anomaly occurrence are found in the real-time infrared thermal imaging data, and the keyframe image within them is extracted. This keyframe image clearly shows the thermal anomaly of the bearing.

[0122] Step S154: Encapsulate and process the anomaly type, warning level, associated component identifier, component location information, and key frame image to generate a structured warning data unit.

[0123] The anomaly type, warning level, associated component identifier, component location information, and keyframe image are encapsulated in a specific format to generate a structured warning data unit. This warning data unit contains a structured representation of this information, such as in JSON format. The anomaly type field stores the name of the anomaly type, the warning level field stores the numerical value of the warning level, the associated component identifier field stores the identifier of the associated component, the component location information field stores the specific location description and coordinates of the component, and the keyframe image field stores the binary data or storage path of the keyframe image.

[0124] Step S155: Convert the early warning data unit into a standard format railway vehicle anomaly early warning command according to the preset early warning command format specification.

[0125] According to the preset warning instruction format specification, the warning data unit is converted into a standard format railway vehicle anomaly warning instruction. The preset warning instruction format specification stipulates the requirements for the fields, data types, and formats of the warning instruction. For example, the warning instruction includes an instruction header, instruction body, and instruction tail. The instruction body contains information such as the anomaly type, warning level, associated component identifier, component location information, and keyframe image, and the format of this information conforms to specific requirements.

[0126] Step S156: Query the preset target monitoring terminal address list to determine the target monitoring terminal that needs to receive the corresponding railway vehicle abnormality warning command.

[0127] A pre-defined list of target monitoring terminal addresses is used, storing the address information (e.g., IP address, port number) of each target monitoring terminal. When a railway vehicle anomaly warning command is generated, this address list is queried to determine the target monitoring terminals that need to receive the warning command. For example, for the anomaly type of bearing overheating, the target monitoring terminal address list might include monitoring terminals from the depot and train drivers.

[0128] Step S157: Transmit railway vehicle anomaly warning commands to each target monitoring terminal through a secure communication protocol to maintain data integrity and transmission security during the command transmission process.

[0129] Secure communication protocols (such as SSL / TLS) are used to transmit railway vehicle anomaly warning commands to each target monitoring terminal. During transmission, the warning commands are encrypted to ensure data security; simultaneously, data verification methods (such as CRC checksum) are used to ensure data integrity during command transmission and prevent data from being tampered with or lost during transmission.

[0130] Figure 2 The illustration shows exemplary hardware and software components of an infrared thermal imaging-based railway vehicle anomaly early warning system 100 that can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the infrared thermal imaging-based railway vehicle anomaly early warning system 100 and to perform the functions in this application.

[0131] The railway vehicle anomaly early warning system 100 based on infrared thermal imaging can be a general-purpose server or a special-purpose server; both can be used to implement the railway vehicle anomaly early warning method based on infrared thermal imaging of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0132] For example, the infrared thermal imaging-based railway vehicle anomaly early warning system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the infrared thermal imaging-based railway vehicle anomaly early warning system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The infrared thermal imaging-based railway vehicle anomaly early warning system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0133] For ease of explanation, only one processor is described in the infrared thermal imaging-based railway vehicle anomaly early warning system 100. However, it should be noted that the infrared thermal imaging-based railway vehicle anomaly early warning system 100 of this application may also include multiple processors. Therefore, the steps performed by one processor described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the infrared thermal imaging-based railway vehicle anomaly early warning system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0134] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned railway vehicle anomaly early warning method based on infrared thermal imaging is implemented.

[0135] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for early warning of railway vehicle anomalies based on infrared thermal imaging, characterized in that, The method includes: Acquire infrared thermal imaging data sets of railway vehicles and construct thermal reference feature sets of key components of railway vehicles; Real-time infrared thermal imaging data of railway vehicles during operation is collected, and the real-time infrared thermal imaging data is dynamically compared with the thermal reference feature set to generate a thermal deviation feature set. The infrared thermal anomaly identification model is invoked to perform correlation enhancement processing on the thermal deviation feature set, generating a strengthened correlation deviation feature vector; The enhanced correlation deviation feature vector is matched and mapped with the preset abnormal hot state feature library to determine the abnormal type of railway vehicle and the identifier of the associated component. Based on the anomaly type and associated component identifier, a railway vehicle anomaly warning instruction containing component location information is generated and sent to the target monitoring terminal. The step of calling the infrared thermal anomaly recognition model to perform correlation enhancement processing on the thermal deviation feature set, generating an enhanced correlation deviation feature vector, includes: The thermal deviation feature set is grouped according to the relationship between key components to generate multiple component association feature groups. Each component association feature group contains the thermal deviation features of multiple key components that interact functionally. The associated feature groups of each component are input into the feature association layer of the infrared thermal anomaly identification model. By modeling the thermal influence relationship between components, a thermal association matrix is ​​generated within the associated feature groups of the components. The step of inputting the component association feature groups into the feature association layer of the infrared thermal anomaly identification model, and generating a thermal association matrix within the component association feature groups by modeling the thermal influence relationship between components, includes: The thermal deviation features of each key component in the component association feature group are processed by feature encoding to generate a standardized component deviation feature vector; Based on the mechanical structure principle of railway vehicles, a prior knowledge graph of thermal influence between components is constructed. The prior knowledge graph includes the heat conduction path and the degree of thermal influence between key components. The standardized component deviation feature vectors and prior knowledge graphs are input into the association modeling unit of the feature association layer, and the dynamic thermal association relationship between components is learned through graph neural network algorithm; Calculate the thermal correlation coefficient between any two key components within each component association feature group. The thermal correlation coefficient reflects the degree of interaction between the thermal deviation features of the two key components. A thermal correlation matrix is ​​constructed within the component correlation feature group based on all thermal correlation coefficients. The rows and columns of the thermal correlation matrix correspond to the key components in the component correlation feature group, and the matrix elements are the corresponding thermal correlation coefficients.

2. The railway vehicle anomaly early warning method based on infrared thermal imaging according to claim 1, characterized in that, The acquisition of infrared thermal imaging data sets of railway vehicles, and the construction of thermal reference feature sets for key components of railway vehicles, includes: The infrared thermal imaging data set of railway vehicles under different operating conditions is collected. The infrared thermal imaging data set contains continuous thermal imaging frames of each key component under normal operating conditions. The key component region localization process is performed on continuous thermal imaging frames in the infrared thermal imaging dataset, and the corresponding thermal imaging sub-regions of each key component are extracted. The thermal imaging sub-regions maintain a correspondence with the physical structural boundaries of the key components. Extract the thermal distribution features of each thermal imaging sub-region. The thermal distribution features include the distribution pattern of thermal radiation intensity and the trend of thermal gradient change within the thermal imaging sub-region. The thermal distribution characteristics of the same key component under different operating conditions are statistically aggregated to generate the thermal characteristic baseline of the corresponding key component. The thermal characteristic baseline reflects the thermal fluctuation of the key component within the normal operating range. Summarize the thermal characteristic baselines of all key components and construct a set of thermal benchmark features for each key component of a railway vehicle that includes a one-to-one correspondence between component identifiers and thermal characteristic baselines. Each thermal feature baseline in the thermal baseline feature set is processed to unify the feature dimensions, so that the thermal feature baselines of different key components have the same feature representation.

3. The railway vehicle anomaly early warning method based on infrared thermal imaging according to claim 1, characterized in that, The real-time acquisition of infrared thermal imaging data during railway vehicle operation involves dynamically comparing the real-time infrared thermal imaging data with a thermal baseline feature set to generate a thermal deviation feature set, including: Infrared thermal imaging acquisition equipment deployed along the railway line captures real-time infrared thermal imaging data of railway vehicles in operation. The real-time infrared thermal imaging data includes real-time thermal imaging frame sequences of key components in operation. Frame synchronization processing is performed on the real-time thermal imaging frame sequence in the real-time infrared thermal imaging data to keep each real-time thermal imaging frame consistent in the time dimension. For each real-time thermal imaging frame, the real-time thermal imaging sub-regions corresponding to each key component are extracted according to the positioning rules of key components in the thermal reference feature set. Extract the real-time thermal distribution features of each real-time thermal imaging sub-region. The feature dimension of the real-time thermal distribution features is consistent with the feature dimension of the corresponding thermal feature baseline in the thermal baseline feature set. The real-time thermal distribution characteristics of each key component are compared dimension by dimension with the thermal characteristic baseline of the corresponding key component in the thermal benchmark feature set to identify the feature dimensions that are not in the thermal fluctuation range. Summarize the feature dimensions of all key components that are not in the thermal fluctuation range, and generate a thermal deviation feature set that includes component identification, deviation feature dimension, and deviation direction.

4. The railway vehicle anomaly early warning method based on infrared thermal imaging according to claim 1, characterized in that, The step of calling the infrared thermal anomaly identification model to perform correlation enhancement processing on the thermal deviation feature set, generating an enhanced correlation deviation feature vector, includes: Based on the thermal correlation matrix, the thermal deviation features within the component correlation feature group are weighted and fused to enhance the collaborative attributes of the deviation features of interrelated components. The weighted and fused feature groups of each component are input into the feature enhancement layer of the infrared thermal anomaly recognition model. The features are expanded in dimension through nonlinear transformation to generate multi-dimensional associated features. Multi-dimensional related features are processed to remove feature redundancy, retaining the core feature dimensions related to anomaly identification; All core feature dimensions corresponding to the associated feature groups of all components are concatenated in a preset order to generate a deviated feature vector with enhanced association.

5. The railway vehicle anomaly early warning method based on infrared thermal imaging according to claim 1, characterized in that, The step of matching and mapping the enhanced correlation deviation feature vector with a preset abnormal hot state feature library to determine the abnormal type and associated component identifier of the railway vehicle includes: Obtain a pre-defined abnormal hot state feature library, which contains standard abnormal feature vectors corresponding to various known abnormal types and associated component identifier sets; Calculate the feature similarity between the deviated feature vector after enhanced association and each standard anomaly feature vector in the abnormal hot state feature library; Standard abnormal feature vectors whose feature similarity meets the preset conditions are selected as candidate abnormal feature vectors; If there are multiple candidate abnormal feature vectors, the candidate abnormal feature vectors are sorted based on the numerical value of feature similarity, and the candidate abnormal feature vector that meets the preset priority condition is selected as the target abnormal feature vector. Extract the anomaly type and associated component identifier set corresponding to the target anomaly feature vector; By combining the deviation degree of each key component in the thermal deviation feature set, the core associated component identifier is determined from the associated component identifier set, and finally the abnormal type and associated component identifier of the railway vehicle are obtained.

6. The railway vehicle anomaly early warning method based on infrared thermal imaging according to claim 5, characterized in that, The calculation of the feature similarity between the deviated feature vector after enhanced correlation and each standard anomalous feature vector in the anomalous hot-state feature library includes: The deviation feature vectors after enhanced correlation and each standard abnormal feature vector are aligned in the feature space so that they are in the same feature space; Calculate the feature magnitude of the deviation feature vector after enhanced association and each standard anomaly feature vector; Calculate the dot product of the deviated feature vector after enhanced association with each standard anomaly feature vector; Based on the feature magnitude and the dot product result, the cosine similarity value between the two is calculated; Normalize the cosine similarity values ​​to obtain feature similarity values ​​within a fixed interval; Record the feature similarity corresponding to each standard anomaly feature vector to form a feature similarity list.

7. The railway vehicle anomaly early warning method based on infrared thermal imaging according to claim 1, characterized in that, The process of generating a railway vehicle anomaly warning command containing component location information based on the anomaly type and associated component identifier, and sending the railway vehicle anomaly warning command to the target monitoring terminal, includes: Analyze the warning level rules corresponding to the anomaly type and determine the warning level corresponding to the anomaly type; The system queries the pre-set railway vehicle component positioning database based on the associated component identifier to obtain the component positioning information corresponding to the associated component identifier. The component positioning information includes a description of the specific location of the component on the railway vehicle and its positioning coordinates. Extract keyframe images of the real-time thermal imaging sub-regions corresponding to the associated components from the real-time infrared thermal imaging data; The abnormality type, warning level, associated component identification, component location information and key frame image are encapsulated and processed to generate structured warning data units; According to the preset warning instruction format specifications, the warning data units are converted into standard format railway vehicle abnormality warning instructions; Query the preset list of target monitoring terminal addresses to determine the target monitoring terminal that needs to receive the corresponding railway vehicle abnormality warning command; The system transmits abnormal early warning commands for railway vehicles to each target monitoring terminal via a secure communication protocol, maintaining data integrity and transmission security during the command transmission process.

8. The railway vehicle anomaly early warning method based on infrared thermal imaging according to claim 4, characterized in that, The process of removing feature redundancy from multi-dimensional correlated features, retaining the core feature dimensions relevant to anomaly identification, includes: A feature importance evaluation algorithm is used to score the importance of each feature dimension of the multi-dimensional related features. The importance score reflects the impact of the corresponding feature dimension on the anomaly identification result. Set a feature importance threshold and select feature dimensions whose importance scores meet the threshold requirements as candidate core feature dimensions; Calculate the pairwise correlations between candidate core feature dimensions and identify sets of feature dimensions that are mutually related. For each set of interrelated feature dimensions, feature fusion processing is performed to generate a comprehensive feature dimension, which replaces the original feature dimension in the corresponding feature dimension set. The unfused feature dimensions and the generated comprehensive feature dimensions from the candidate core feature dimensions are combined to form the final core feature dimensions. The multi-dimensional related features are projected onto the core feature dimension to obtain the core feature vector after removing redundancy.

9. A railway vehicle anomaly early warning system based on infrared thermal imaging, characterized in that, The railway vehicle anomaly early warning system based on infrared thermal imaging includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the railway vehicle anomaly early warning method based on infrared thermal imaging as described in any one of claims 1-8.

Images