A vehicle bogie fatigue life estimation method, system, device and medium

Abstract

The application provides a vehicle bogie fatigue life estimation method, system, device and medium, comprising: collecting multi-source monitoring data of key bearing parts of the bogie, extracting corresponding stress signals based on the mapping relationship between sensors, components and stress / strain proxies. The low-fidelity fatigue physical model is used to calculate the physical branch estimation value, and the shared feature extraction model and related labels trained in the source domain are combined to obtain the transfer life. According to the labeled samples arranged by parts in the target domain, the residual correction amount is solved, and finally the physical branch estimation value, the transfer life and the residual correction amount are fused to output the fatigue life estimation result of the key parts of the bogie under the new working condition. By using the above method, the accuracy and working condition adaptability of the estimation result can be improved.

Description

Technical Field

[0001] This application relates to the field of rail vehicle technology, and more specifically, to a method, system, device, and medium for estimating the fatigue life of a vehicle bogie. Background Technology

[0002] Rail transit equipment has become the core carrier of passenger and freight transportation. As a crucial running, load-bearing, and vibration-damping component of rail vehicles, the bogie, with its key load-bearing areas such as weld hotspots, traction rod connection areas, axle box positioning connection areas, vibration dampers, and various mounting seats, continuously endures complex loads including alternating loads, wheel-rail impacts, random vibrations, and seasonal temperature changes during long-term operation. These areas are highly susceptible to gradual fatigue cracking and structural damage, and fatigue failure directly threatens vehicle operational safety. Therefore, routine online fatigue life estimation of key load-bearing components of the bogie, predicting component health status in advance, and supporting predictive maintenance are key tasks in the rail transit operation and maintenance field, and are also essential requirements for ensuring long-term safe service of vehicles and reasonably controlling operation and maintenance costs.

[0003] Currently, the industry generally adopts a purely data-driven approach for estimating bogie fatigue life. This method first deploys sensors on the vehicle to continuously collect multi-source operational monitoring data, including vibration, displacement, operating speed, mileage, ambient temperature, strain, and stress proxy signals. Then, relying on a large number of fatigue damage and life-related labeled samples accumulated from mature operating lines and existing vehicle models, a data-driven model with a neural network at its core is trained. After the model is trained, it is directly deployed and used. The multi-source monitoring data collected in real time is input into the model, and relying on the correlation of the data's own characteristics, the fatigue life and damage assessment results of the key load-bearing parts of the bogie are directly output. This approach can achieve relatively ideal assessment results under normal operating conditions with sufficient training samples.

[0004] This purely data-driven solution heavily relies on a sufficient number of effective labeled samples. However, under entirely new operating conditions such as the opening of new lines, the introduction of new train models, seasonal changes, and alterations in wheel-rail conditions, it is difficult to quickly collect sufficient labeled data on fatigue failure, damage levels, and lifespan on-site. This represents a typical small-sample application scenario. Under these conditions, the data-driven model is highly susceptible to overfitting, which not only significantly reduces the accuracy of fatigue life prediction but may also produce abnormal extrapolation results that violate the fundamental mechanical laws of metal fatigue. Consequently, it cannot reliably complete online evaluations under new operating conditions and is ill-suited to the actual operation and maintenance needs of rail transit under diverse operating conditions. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a method, system, device and medium for estimating the fatigue life of a vehicle bogie, which can improve the accuracy of the estimation results and the adaptability to operating conditions.

[0006] In a first aspect, embodiments of this application provide a method for estimating the fatigue life of a vehicle bogie, the method comprising: Collect multi-source monitoring data corresponding to key load-bearing components of the bogie during vehicle operation; Based on the mapping relationship between sensor installation location, key load-bearing parts of the bogie and stress / strain surrogate, stress-related signals corresponding to the key load-bearing parts of the bogie are extracted from the multi-source monitoring data. Based on the stress-related signals and the low-fidelity fatigue physical model used for online estimation, the physical branch estimates of the key load-bearing parts of the bogie are determined. Based on the multi-source monitoring data, structural attribute labels and load path labels of the key load-bearing parts of the bogie, and the shared feature extraction model trained by the source domain, the estimated migration life of the key load-bearing parts of the bogie is determined, wherein the source domain is the working condition domain of the existing samples. Based on the labeled samples organized according to the key load-bearing parts of the bogie in the target domain, the residual correction amount for the target working condition is determined, wherein the target domain is the data domain corresponding to the new working condition to be launched. Based on the physical branch estimate, the migration life estimate, and the residual correction, the target fatigue life estimate of the key load-bearing components of the bogie under the target working condition is generated.

[0007] Optionally, the key load-bearing parts of the bogie include at least one of the following: frame weld hot spot area, traction rod seat connection area, axle box positioning connection area, shock absorber mounting seat, air spring seat, brake hanger seat, motor hanger seat, lateral stop installation area and its adjacent transition area.

[0008] Optionally, the multi-source monitoring data includes at least three of the following: vibration, displacement or relative displacement, vehicle speed, mileage, temperature, strain signal, and stress proxy signal.

[0009] Optionally, the low-fidelity fatigue physical model includes at least one or more of the following: equivalent stress amplitude calculation, stress ratio correction, fatigue life relationship, and cumulative damage calculation.

[0010] Optionally, determining the physical branch estimate of the key load-bearing component of the bogie based on the stress-related signal and the low-fidelity fatigue physical model used for online estimation includes: Based on the stress-related signals, the equivalent stress amplitude of the key load-bearing parts of the bogie within the sliding window is determined; A stress ratio correction operation is performed on the equivalent stress amplitude to obtain the corrected equivalent stress amplitude, wherein the stress ratio correction operation adopts the Walker correction method; By introducing structural correction coefficients and load correction coefficients corresponding to the key load-bearing parts of the bogie, the corrected equivalent stress amplitude is modified to obtain the final equivalent stress amplitude. The structural correction coefficient is determined based on at least one of the connection form, weld detail level, local plate thickness, reinforcement form, and stiffness of openings or transition zones of the key load-bearing parts of the bogie; the load correction coefficient is determined based on the force path type of the key load-bearing parts of the bogie within the current operating window, and the force path type includes at least one of bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load, or abnormal wheel-rail excitation. Based on the final equivalent stress amplitude, the physical branch estimate of the key load-bearing parts of the bogie is calculated.

[0011] Optionally, before collecting multi-source monitoring data during vehicle operation, the method further includes: A mapping relationship is pre-established between the sensor installation location, the key load-bearing parts of the bogie, and the stress / strain surrogate quantity, and the structural attribute labels of the key load-bearing parts of the bogie are recorded; The mapping relationship is used to extract stress-related signals corresponding to the key load-bearing parts of the bogie from the multi-source monitoring data, and the structural attribute label is used to determine the structural correction coefficient in the low-fidelity fatigue physical model.

[0012] Optionally, the structural attribute labels include the part category, connection type, weld detail level, local plate thickness, reinforcement type, stiffness level of adjacent components, and component area number of the key load-bearing parts of the bogie; The structural correction coefficient is determined by reading at least one parameter value from the structural attribute label, including the connection type, weld detail level, local plate thickness, reinforcement type, and stiffness of the opening or transition zone of the key load-bearing parts of the bogie, and calculating or querying the structural correction coefficient based on the read parameter values.

[0013] Optionally, the source domain includes one or more of the following: test bench samples, historical route samples, simulation samples, or measured samples of existing vehicle models; the target domain includes data domains corresponding to new routes, new vehicle models, new seasons, new wheel and rail conditions, or new component installation conditions.

[0014] Optionally, the stress ratio correction operation uses the Walker stress ratio correction parameter corresponding to the key load-bearing part of the bogie to correct the equivalent stress amplitude; The structural correction coefficient is determined based on the structural attribute label and is used to characterize the local structural differences corresponding to the connection form, weld detail level, local plate thickness, reinforcement form, and stiffness of openings or transition zones of the key load-bearing parts of the bogie.

[0015] Optionally, the load path correction coefficient is used to characterize the impact of at least one of the following force path types on fatigue damage of the key load-bearing parts of the bogie within the current operating window: bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load, or abnormal wheel-rail excitation.

[0016] Optionally, when correcting the equivalent stress amplitude, one or more of an environmental correction factor and an installation condition correction factor are also introduced; The environmental correction factor is used to characterize the effects of temperature range, seasonal variation, or thermo-mechanical coupling. The installation status correction coefficient is used to characterize factors such as reinstallation after maintenance, changes in pre-tightening status, or replacement of accessories.

[0017] Optionally, the shared feature extraction model includes a first branch, a second branch, and a third branch; The first branch is used to extract time-series representations, the second branch is used to extract frequency domain features, and the third branch is used to process operating conditions and structural attribute features. The source domain shared representation vector is output after the features of each branch are fused. The input to the shared feature extraction model includes time-domain features, frequency-domain features, and operating condition labels extracted from the multi-source monitoring data, as well as structural attribute labels and load path labels for the key load-bearing parts of the bogie.

[0018] Optionally, generating the target fatigue life estimate of the key load-bearing components of the bogie based on the physical branch estimate, the migration life estimate, and the residual correction includes: The physical branch estimate, the migration lifetime estimate, and the residual correction are weighted and fused in the logarithmic lifetime domain to obtain the fusion result; Based on the fusion results, the target fatigue life estimation results of the key load-bearing parts of the bogie are generated.

[0019] Optionally, the method further includes: Determine whether the fusion result meets the preset physical consistency constraints, wherein the physical consistency constraints include that the fatigue life does not increase abnormally when the stress amplitude increases, the fatigue life change rate of adjacent sliding windows does not exceed a reasonable range, and the fatigue life level of key load-bearing parts of the same type of bogie remains orderly. When the fusion result does not meet the physical consistency constraint, at least one of the fusion weights corresponding to the physical branch estimate, the migration lifetime estimate, and the residual correction amount is adjusted, and the weighted fusion is re-executed based on the adjusted weights until the fusion result meets the physical consistency constraint.

[0020] Optionally, the method further includes: Calculate the working condition similarity between the target domain and the source domain, and determine the comprehensive migration score according to the key load-bearing parts of the bogie, wherein the comprehensive migration score is composed of one or more weighted factors of structural attribute similarity, load path similarity, sample coverage and path novelty; The freezing layer strategy is determined based on the comprehensive migration score and the target domain sample size of the key load-bearing parts of the bogie. According to the freezing layer strategy, the bottom layer of the shared feature extraction model is frozen and only the high-level residual branches are fine-tuned, or a conservative update is performed when the distribution offset is significant and the novelty of the key load-bearing parts of the bogie is too high.

[0021] Optionally, the method further includes: The confidence level of the target fatigue life estimation result is evaluated; the confidence level is composed of one or more of the following: sample coverage, structural attribute similarity, load path similarity, path novelty, physical consistency satisfaction, and sensor quality. When the confidence level is lower than a preset threshold, the target fatigue life estimation result is replaced with the smaller of the physical branch estimation value and the preset safe life threshold, and at least one of the following maintenance actions is triggered: resampling, increasing the retesting frequency, or outputting manual review suggestions.

[0022] Optionally, the method further includes: Obtain newly labeled samples for the target domain; Based on the newly labeled samples, the model branch used to determine the residual correction amount is updated online. The bottom shared feature layer of the shared feature extraction model is kept frozen, and only the residual correction head, interval estimation head, confidence evaluation head and some high-level fusion weights are updated. The updated residual correction amount is determined based on the updated model branch; Based on the physical branch estimate, the migration lifetime estimate, and the updated residual correction, an updated target fatigue lifetime estimate is generated.

[0023] Optionally, the method further includes: The continuous operation data is divided into discrete sample units according to time windows or mileage windows, and structural attribute labels and load path labels of the key load-bearing parts of the bogie are attached to each sample unit. The features extracted for each window include one or more of the following: root mean square, peak-to-peak value, kurtosis, dominant band energy, bandpass power spectral density amplitude, spectral moment, phase difference, temperature rise rate, mean velocity, and strain range. The extracted features are used as input to the shared feature extraction model.

[0024] Secondly, embodiments of this application provide a vehicle bogie fatigue life estimation device, the device comprising: The data acquisition module is used to collect multi-source monitoring data corresponding to key load-bearing components of the vehicle bogie; The signal extraction module is used to extract stress-related signals from the multi-source monitoring data based on the mapping relationship between sensor installation location, key load-bearing parts of the bogie and stress / strain surcharge. The physical estimation module is used to calculate the physical branch estimate based on the stress-related signal and the low-fidelity fatigue physical model. The migration prediction module is used to combine multi-source monitoring data, structural attribute labels of key load-bearing parts of the bogie, load path labels, and a shared feature extraction model trained in the source domain to determine the migration lifetime estimate, wherein the source domain is the operating condition domain of existing samples. The residual correction module is used to determine the residual correction amount corresponding to the target working condition based on the target domain annotation samples organized according to the key load-bearing parts of the bogie. The target domain is the data domain corresponding to the new working condition to be launched. The fusion output module is used to generate fatigue life estimation results for key load-bearing components of the bogie under target working conditions based on the physical branch estimate, migration life estimate, and residual correction.

[0025] Optionally, the key load-bearing parts of the bogie include at least one of the following: frame weld hot spot area, traction rod seat connection area, axle box positioning connection area, shock absorber mounting seat, air spring seat, brake hanger seat, motor hanger seat, lateral stop installation area and its adjacent transition area.

[0026] Optionally, the multi-source monitoring data includes at least three of the following: vibration, displacement or relative displacement, vehicle speed, mileage, temperature, strain signal, and stress proxy signal.

[0027] Optionally, the low-fidelity fatigue physical model includes at least one or more of the following: equivalent stress amplitude calculation, stress ratio correction, fatigue life relationship, and cumulative damage calculation.

[0028] Optionally, the physical estimation module is specifically used for: Based on the stress-related signals, the equivalent stress amplitude of the key load-bearing parts of the bogie within the sliding window is determined; A stress ratio correction operation is performed on the equivalent stress amplitude to obtain the corrected equivalent stress amplitude, wherein the stress ratio correction operation adopts the Walker correction method; By introducing structural correction coefficients and load correction coefficients corresponding to the key load-bearing parts of the bogie, the corrected equivalent stress amplitude is modified to obtain the final equivalent stress amplitude. The structural correction coefficient is determined based on at least one of the connection form, weld detail level, local plate thickness, reinforcement form, and stiffness of openings or transition zones of the key load-bearing parts of the bogie; the load correction coefficient is determined based on the force path type of the key load-bearing parts of the bogie within the current operating window, and the force path type includes at least one of bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load, or abnormal wheel-rail excitation. Based on the final equivalent stress amplitude, the physical branch estimate of the key load-bearing parts of the bogie is calculated.

[0029] Optionally, the device further includes a pre-configuration module, which is used to pre-establish the mapping relationship between the sensor installation position, the key load-bearing parts of the bogie and the stress / strain surcharge before collecting multi-source monitoring data during vehicle operation, and to record the structural attribute labels of the key load-bearing parts of the bogie. The mapping relationship is used to extract stress-related signals corresponding to the key load-bearing parts of the bogie from the multi-source monitoring data, and the structural attribute label is used to determine the structural correction coefficient in the low-fidelity fatigue physical model.

[0030] Optionally, the structural attribute labels include the part category, connection type, weld detail level, local plate thickness, reinforcement type, stiffness level of adjacent components, and component area number of the key load-bearing parts of the bogie; The structural correction coefficient is determined by reading at least one parameter value from the structural attribute label, including the connection type, weld detail level, local plate thickness, reinforcement type, and stiffness of the opening or transition zone of the key load-bearing parts of the bogie, and calculating or querying the structural correction coefficient based on the read parameter values.

[0031] Optionally, the source domain includes one or more of the following: test bench samples, historical route samples, simulation samples, or measured samples of existing vehicle models; the target domain includes data domains corresponding to new routes, new vehicle models, new seasons, new wheel and rail conditions, or new component installation conditions.

[0032] Optionally, the stress ratio correction operation uses the Walker stress ratio correction parameter corresponding to the key load-bearing part of the bogie to correct the equivalent stress amplitude; the structural correction coefficient is determined based on the structural attribute label and is used to characterize the connection form, weld detail level, local plate thickness, reinforcement form and the local structural differences corresponding to the stiffness of the opening or transition zone of the key load-bearing part of the bogie.

[0033] Optionally, the load path correction coefficient is used to characterize the impact of at least one of the following force path types on fatigue damage of the key load-bearing parts of the bogie within the current operating window: bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load, or abnormal wheel-rail excitation.

[0034] Optionally, when the physical estimation module corrects the equivalent stress amplitude, it also introduces one or more of an environmental correction coefficient and an installation state correction coefficient; the environmental correction coefficient is used to characterize temperature range, seasonal changes, or thermo-mechanical coupling effects; the installation state correction coefficient is used to characterize factors such as reinstallation after maintenance, changes in pre-tightening status, or replacement of accessories.

[0035] Optionally, the shared feature extraction model includes a first branch, a second branch, and a third branch; the first branch is used to extract time-series representations, the second branch is used to extract frequency-domain features, and the third branch is used to process operating condition and structural attribute features. The features from each branch are fused to output a source-domain shared representation vector. The input of the shared feature extraction model includes time-domain features, frequency-domain features, operating condition labels extracted from the multi-source monitoring data, as well as structural attribute labels and load path labels of the key load-bearing parts of the bogie.

[0036] Optionally, the fusion output module is specifically used for: The physical branch estimate, the migration lifetime estimate, and the residual correction are weighted and fused in the logarithmic lifetime domain to obtain the fusion result; Based on the fusion results, the target fatigue life estimation results of the key load-bearing parts of the bogie are generated.

[0037] Optionally, the device further includes a consistency constraint module for: Determine whether the fusion result meets the preset physical consistency constraints, wherein the physical consistency constraints include that the fatigue life does not increase abnormally when the stress amplitude increases, the fatigue life change rate of adjacent sliding windows does not exceed a reasonable range, and the fatigue life level of key load-bearing parts of the same type of bogie remains orderly. When the fusion result does not meet the physical consistency constraint, at least one of the fusion weights corresponding to the physical branch estimate, the migration lifetime estimate, and the residual correction amount is adjusted, and the fusion output module is triggered to re-execute the weighted fusion based on the adjusted weights until the fusion result meets the physical consistency constraint.

[0038] Optionally, the device further includes a freezing strategy module for: Calculate the working condition similarity between the target domain and the source domain, and determine the comprehensive migration score according to the key load-bearing parts of the bogie, wherein the comprehensive migration score is composed of one or more weighted factors of structural attribute similarity, load path similarity, sample coverage and path novelty; The freezing layer strategy is determined based on the comprehensive migration score and the target domain sample size of the key load-bearing parts of the bogie. According to the freezing layer strategy, the bottom layer of the shared feature extraction model is frozen and only the high-level residual branches are fine-tuned, or a conservative update is performed when the distribution offset is significant and the novelty of the key load-bearing parts of the bogie is too high.

[0039] Optionally, the device further includes a confidence assessment module for: The confidence level of the target fatigue life estimation result is evaluated; the confidence level is composed of one or more of the following: sample coverage, structural attribute similarity, load path similarity, path novelty, physical consistency satisfaction, and sensor quality. When the confidence level is lower than a preset threshold, the target fatigue life estimation result is replaced with the smaller of the physical branch estimation value and the preset safe life threshold, and at least one of the following maintenance actions is triggered: resampling, increasing the retesting frequency, or outputting manual review suggestions.

[0040] Optionally, the device further includes an online update module for: Obtain newly labeled samples for the target domain; Based on the newly labeled samples, the model branch used to determine the residual correction amount is updated online. The bottom shared feature layer of the shared feature extraction model is kept frozen, and only the residual correction head, interval estimation head, confidence evaluation head and some high-level fusion weights are updated. The updated residual correction amount is determined based on the updated model branch; The fusion output module is triggered to generate an updated target fatigue life estimate based on the physical branch estimate, the migration lifetime estimate, and the updated residual correction.

[0041] Optionally, the device further includes a feature extraction module for: The continuous operation data is divided into discrete sample units according to time windows or mileage windows, and structural attribute labels and load path labels of the key load-bearing parts of the bogie are attached to each sample unit. The features extracted for each window include one or more of the following: root mean square, peak-to-peak value, kurtosis, dominant band energy, bandpass power spectral density amplitude, spectral moment, phase difference, temperature rise rate, mean velocity, and strain range. The extracted features are used as input to the shared feature extraction model.

[0042] Thirdly, embodiments of this application provide a computer device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the vehicle bogie fatigue life estimation method described in any of the optional embodiments of the first aspect are performed.

[0043] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the vehicle bogie fatigue life estimation method described in any of the optional embodiments of the first aspect.

[0044] The technical solution provided in this application includes, but is not limited to, the following beneficial effects: This method collects multi-source monitoring data corresponding to various key load-bearing components of the bogie, which can comprehensively and completely reflect the actual operating load and working status of the bogie, providing a solid and effective original data foundation for subsequent fatigue life calculation.

[0045] This method relies on a pre-established mapping relationship between sensors, key load-bearing components, and stress / strain surrogate quantities to extract stress-related signals. It can achieve accurate conversion of multi-source monitoring data into core parameters for fatigue analysis, ensuring that the extracted stress signals match the actual stress state of the bogie and effectively improving the data accuracy of subsequent life assessment.

[0046] This method uses a low-fidelity fatigue physical model to calculate the physical branch estimate. The output results have clear physical basis and interpretability, thus establishing a stable and reliable basic assessment benchmark for overall life assessment.

[0047] This method uses a shared feature extraction model trained on existing sample operating conditions to solve the migration lifetime estimate. It can make full use of the effective data features accumulated in historical operating conditions, without the need for large-scale model training for new operating conditions, and can quickly complete the preliminary lifetime assessment under operating condition migration.

[0048] This method uses the labeled samples corresponding to the new working condition to calculate the residual correction amount, which can specifically make up for the evaluation deviation between the original model and the new working condition. Even when the number of labeled samples is limited, it can achieve targeted correction of the evaluation results and adapt to the application characteristics of small samples in the new working condition.

[0049] This method integrates physical branch estimates, migration life estimates, and residual corrections to generate the final fatigue life estimate. It combines physical calculation benchmarks, historical data patterns, and correction results for new operating conditions, with multiple dimensions complementing and constraining each other, effectively improving the accuracy and reliability of the final fatigue life estimate of the bogie.

[0050] In summary, this method, by sequentially completing the entire process of data acquisition, signal extraction, physical calculation, migration assessment, residual correction, and result fusion, achieves reliable estimation of the fatigue life of key load-bearing components of rail vehicle bogies without relying on massive amounts of new working condition labeled samples. It effectively adapts to various new working conditions such as new lines and new vehicle models, and meets the fatigue assessment needs in actual operation and maintenance.

[0051] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 A flowchart of a method for estimating the fatigue life of a vehicle bogie provided in Embodiment 1 of this application is shown; Figure 2 This paper shows a schematic diagram of the overall architecture of a vehicle bogie fatigue life estimation method provided in Embodiment 1 of this application; Figure 3 A flowchart of a method for determining physical branch estimates provided in Embodiment 1 of this application is shown; Figure 4 A flowchart of a method for determining fatigue life estimation results provided in Embodiment 1 of this application is shown; Figure 5 This paper shows a schematic diagram of a multi-fidelity fusion core module structure provided in Embodiment 1 of this application; Figure 6 This paper illustrates a flowchart of a method for estimating the fatigue life of a vehicle bogie according to Embodiment 1 of this application. Figure 7 A flowchart of a method for determining fatigue life estimation results provided in Embodiment 1 of this application is shown; Figure 8 A flowchart of a fatigue life estimation result update method provided in Embodiment 1 of this application is shown; Figure 9 The flowchart of a method for determining the input of a shared feature extraction model provided in Embodiment 1 of this application is shown. Figure 10 This paper shows a schematic diagram of the structure of a vehicle bogie fatigue life estimation device provided in Embodiment 2 of this application; Figure 11 A schematic diagram of the structure of a computer device provided in Embodiment 3 of this application is shown. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0055] Example 1 This solution belongs to the technical fields of structural health management, fatigue life prediction, remaining life estimation, transfer learning-driven predictive maintenance, and multi-fidelity physical-data fusion modeling for rail vehicles. The core application of this solution is to the critical load-bearing components of high-speed train bogies, and its technical principles can be extended to various types of rail transit equipment. While primarily applied to the critical load-bearing components of high-speed train bogies, this solution can also be extended to metro vehicles, suburban trains, intercity EMUs, and trams. In addition to rail transit vehicles, this solution can also be applied to other electromechanical structures subjected to cyclic loads and random vibrations, thus covering a wide range of applicable scenarios.

[0056] This method employs a multi-fidelity overall architecture consisting of a low-fidelity physical branch, a source domain shared feature branch, and a target domain residual correction branch, specifically adapted to new operating conditions with small sample sizes for railway bogies. This method can be deployed as an onboard online monitoring system, a vehicle-to-ground collaborative health management system, and can also be applied to a joint correction system for test benches and in-service data. Furthermore, the technical solution can be extended to subway vehicles, urban rail trains, intercity EMUs, trams, and other electromechanical structures subjected to cyclic loads and random vibrations, forming the core execution logic of the entire technical solution.

[0057] To facilitate understanding of this application, the following is combined with... Figure 1The flowchart illustrating a method for estimating the fatigue life of a vehicle bogie provided in Embodiment 1 of this application describes Embodiment 1 in detail.

[0058] See Figure 1 As shown, Figure 1 A flowchart of a method for estimating the fatigue life of a vehicle bogie according to Embodiment 1 of this application is shown, wherein the method includes steps S101 to S105: S101: Collect multi-source monitoring data corresponding to key load-bearing parts of the bogie during vehicle operation.

[0059] Specifically, the parts listed above are all core load-bearing areas of the bogie that are subjected to alternating loads for a long time and are prone to fatigue damage.

[0060] Specifically, each type of monitoring data has its own dedicated sampling frequency: vibration and strain signals are sampled at 200 Hz-5000 Hz, displacement or relative displacement at 50 Hz-1000 Hz, and vehicle speed and ambient temperature at 1 Hz-100 Hz.

[0061] Specifically, after data collection is completed, all channel data will undergo unified preprocessing, including timestamp alignment, missing value imputation, outlier removal, digital filtering, trend removal, and unit conversion, ultimately forming a standardized continuous sample stream.

[0062] Specifically, in addition to basic monitoring data, the system can also synchronously access line identification, vehicle type identification, wheel and rail status labels, historical maintenance records, and auxiliary data generated by test benches or simulations, enriching the model input dimensions and improving the accuracy of operating condition recognition.

[0063] S102: Based on the mapping relationship between the sensor installation location, the key load-bearing parts of the bogie, and the stress / strain surrogate quantity, extract the stress-related signal corresponding to the key load-bearing parts of the bogie from the multi-source monitoring data.

[0064] Specifically, a unique correspondence between the three will be established for each key load-bearing component of the bogie. Based on this mapping rule, the original multi-source monitoring data will be accurately converted into effective stress and strain signals for each component, providing an accurate data foundation for subsequent fatigue parameter calculations.

[0065] S103: Based on the stress-related signals and the low-fidelity fatigue physical model used for online estimation, determine the physical branch estimates of the key load-bearing parts of the bogie.

[0066] Specifically, this low-fidelity fatigue physics model will be continuously used throughout the entire system operation cycle as a stable and reliable anchor point for physical calculations.

[0067] Specifically, the model integrates four core calculation modules, which sequentially complete the equivalent stress amplitude calculation, stress ratio correction, fatigue life calculation, and linear cumulative damage calculation, and finally output a conservative life baseline and low-fidelity damage values.

[0068] S104: Based on the multi-source monitoring data, structural attribute labels of key load-bearing parts of the bogie, load path labels, and a shared feature extraction model trained in the source domain, determine the estimated migration life of the key load-bearing parts of the bogie, wherein the source domain is the operating condition domain of existing samples.

[0069] Specifically, the source domain refers to mature operating scenarios that have accumulated massive amounts of data. Data sources include test bench test data, long-term line operation history data, simulation calculation data, and actual test data of existing vehicle models.

[0070] Specifically, the shared feature extraction model is trained on source domain data and can learn the common patterns between bogie structure and fatigue characteristics under different track, vehicle type and seasonal conditions, and then output migration life prediction results adapted to the current working conditions.

[0071] S105: Based on the labeled samples organized according to the key load-bearing parts of the bogie in the target domain, determine the residual correction amount for the target working condition, wherein the target domain is the data domain corresponding to the new working condition to be launched.

[0072] Specifically, the target domain mainly corresponds to new operating conditions such as new lines, new vehicle models, new seasons, changes in wheel and rail conditions, and reinstallation of components. The number of available labeled samples in these scenarios is extremely small.

[0073] Specifically, this scheme adopts a dedicated domain division rule under the constraint of key parts. Under the same line operating environment, different key load-bearing parts can be divided into different target domains; the same key load-bearing part will also be divided into different target domains under different speed levels, seasons, and wheel-rail conditions.

[0074] Specifically, during the sample processing stage, data will be categorized and organized separately according to the key load-bearing parts of different bogies, without mixing and calibrating the whole vehicle data, so as to accurately calculate the residual correction amount corresponding to each part.

[0075] Specifically, the target domain labeled samples come from five sources: a small number of measured strain and stress verification windows, working condition verification results of bench reproduction tests, retest results after maintenance, known crack or damage level labels, and fatigue life back calculation labels.

[0076] Specifically, the number of labeled samples for a single critical load-bearing component should be controlled between 5 and 100 labeled windows, with the total number of samples accounting for 1% to 10% of the total number of windows in the target domain for that component. In practical applications, high-risk critical components can be sampled first, while low-risk components can be sampled later, forming a calibration dataset with maintenance priorities.

[0077] S106: Based on the physical branch estimate, the migration life estimate, and the residual correction, generate the target fatigue life estimate of the key load-bearing parts of the bogie under the target working condition.

[0078] Specifically, the output results of the three branches are integrated using multi-fidelity fusion logic, and weighted calculation is completed by combining adaptive weights. By integrating the physical model baseline, source domain migration results and small sample correction, the final fatigue life estimate adapted to the new working conditions is obtained.

[0079] Specifically, the system's final output includes lifespan correction values, lifespan ranges, confidence levels, and rollback flags. It also provides executable maintenance suggestions such as alarms, retesting, conservative use, manual verification, maintenance priority, speed limiting and load reduction, and shutdown inspection.

[0080] Specifically, when the system evaluates multiple key bogie components in parallel, it prioritizes each component based on its lower lifespan, cumulative damage value, result confidence level, and conservative rollback status, and automatically generates a maintenance priority list.

[0081] See Figure 2 As shown, Figure 2 The diagram illustrates the overall architecture of a vehicle bogie fatigue life estimation method provided in Embodiment 1 of this application. The diagram shows the complete process from multi-source data input to final maintenance output. The system uses a multi-source input and preprocessing module as the data entry point, receiving vibration, displacement, velocity, temperature, strain, and stress proxy data. It also incorporates track, vehicle type, season labels, maintenance records, wheel-rail status, and bench and simulation auxiliary data, which are then processed to form a standardized data stream.

[0082] Standardized data are respectively connected to the target domain small sample access module, the source domain data and shared feature training module, and the low-fidelity fatigue physics branch. The target domain small sample access module is used to receive a small amount of labeled windows, strain verification and retest feedback data under new lines, new models, new seasons and new speed levels, and provides support for similarity assessment, frozen layer decision and residual correction. The shared feature and migration prediction branch is based on time domain features, frequency domain features and working condition features encoded, and selectively preserved in the bottom shared layer according to the freezing strategy, and finally outputs the source domain migration lifetime and damage prediction results.

[0083] The source domain data and shared feature training module relies on historical line data, bench test data and simulation data to extract common features across lines, vehicle types and seasons, form a source domain shared representation and output the corresponding migration prediction results. The low-fidelity fatigue physics branch obtains the conservative life baseline and low-fidelity damage value through equivalent stress amplitude calculation, Walker correction, Basquin life relationship and Miner linear cumulative damage calculation.

[0084] The output data from the target domain small sample access module is further input into the target domain residual correction branch. This branch updates the high-level correction head based on a small number of target domain samples and outputs the residual correction amount δ or δD. The results of the low-fidelity fatigue physics branch, the shared feature and migration prediction branch, and the target domain residual correction branch are all integrated into the multi-fidelity fusion and confidence decision module. This module fuses the low-fidelity output, migration prediction results, and residual correction amount to generate lifetime correction values ​​and damage estimates and complete interval estimation. At the same time, it combines sample coverage, working condition similarity, path novelty, and physical consistency to complete confidence assessment. When the confidence is insufficient, a conservative backoff strategy is executed, outputting the fused lifetime correction value and damage estimate, lifetime interval and confidence level, and backoff flag and conservative result, respectively.

[0085] The output of the multi-fidelity fusion and confidence decision module is connected to the online incremental update module. After acquiring new samples, this module only updates the residual header, interval header, confidence header and some high-level fusion weights. At the same time, based on the fusion results, it outputs suggestions for retesting and supplementary sampling, maintenance triggering and priority suggestions, and speed limit and load reduction and shutdown inspection suggestions, thus forming a complete closed-loop system of online monitoring, life estimation and operation and maintenance triggering.

[0086] In one optional implementation, the key load-bearing parts of the bogie include at least one of the following: frame weld hot spot area, traction rod seat connection area, axle box positioning connection area, shock absorber mounting seat, air spring seat, brake hanger seat, motor hanger seat, lateral stop installation area, and adjacent transition area.

[0087] Specifically, the aforementioned parts are the core load-bearing areas of the bogie that bear concentrated alternating loads and have high stress levels. They are also key locations where fatigue cracks are prone to initiation and have a significant impact on the overall vehicle operation safety. Conducting life assessments on these parts can accurately focus on operation and maintenance risk points and ensure the engineering practical value of the assessment results.

[0088] In one alternative implementation, the multi-source monitoring data includes at least three of the following: vibration, displacement or relative displacement, vehicle speed, mileage, temperature, strain signal, and stress proxy signal.

[0089] Specifically, vibration and strain signals can directly reflect the dynamic stress and load amplitude of the structure, speed and mileage data can be correlated with operating conditions and support damage accumulation statistics, and temperature data can characterize the influence of environmental factors on material fatigue performance. Selecting three or more types of data can characterize the real operating state of the bogie from multiple dimensions and avoid the limitations of a single data source.

[0090] In an optional implementation, the low-fidelity fatigue physical model includes at least one or more of the following: equivalent stress amplitude calculation, stress ratio correction, fatigue life relationship, and cumulative damage calculation.

[0091] Specifically, the equivalent stress amplitude calculation can transform complex alternating loads into a unified fatigue assessment equivalent; the stress ratio correction can adapt to the differences in fatigue characteristics under different average stress conditions; the fatigue life relationship and cumulative damage calculation can quantify the degree of fatigue damage under single-cycle loads and long-term service; and the physical model composed of the above modules can output a conservative life assessment baseline with clear physical basis based on mature fatigue mechanics laws.

[0092] In an optional implementation, see Figure 3 As shown, Figure 3 A flowchart of a method for determining physical branch estimates provided in Embodiment 1 of this application is shown, wherein determining the physical branch estimates of the key load-bearing parts of the bogie based on the stress-related signal and a low-fidelity fatigue physical model for online estimation includes steps S301 to S304.

[0093] Specifically, this process fully implements the calculation logic of the low-fidelity fatigue physical model, and combines the bogie structural characteristics, load characteristics, and environmental conditions to complete multi-level stress correction and life solution. It is the core execution process of the physical branch output results.

[0094] S301: Determine the equivalent stress amplitude of the key load-bearing parts of the bogie within the sliding window based on the stress-related signal.

[0095] Specifically, a preset sliding window is used as the smallest calculation unit, and the foundation stress parameters are calculated based on the stress-related signals within the window. The corresponding calculation formula is as follows:

[0096]

[0097]

[0098] in: : No. Within the sliding window, the bogie... Stress amplitude at key locations; : No. Within the sliding window, the bogie... Maximum stress in key components; : No. Within the sliding window, the bogie... Minimum stress in key components; : No. Within the sliding window, the bogie... Average stress in key components; : No. Within the sliding window, the bogie... Stress ratio of key components; Key load-bearing component numbers of the bogie; : Data sliding window number.

[0099] S302: Perform a stress ratio correction operation on the equivalent stress amplitude to obtain the corrected equivalent stress amplitude, wherein the stress ratio correction operation adopts the Walker correction method.

[0100] Specifically, the Walker correction formula is used to complete the stress amplitude correction. The formula is as follows:

[0101] in: After Walker's correction, the first... The bogie within the sliding window Preliminary equivalent stress amplitude of key components; Bogie No. The Walker stress ratio correction factor for each key component is determined by a combination of component material, weld grade, and test data.

[0102] S303: Introduce structural correction coefficients and load correction coefficients corresponding to the key load-bearing parts of the bogie to correct the corrected equivalent stress amplitude and obtain the final equivalent stress amplitude; wherein, the structural correction coefficients are determined according to at least one of the connection form, weld detail level, local plate thickness, reinforcement form, opening or transition zone stiffness of the key load-bearing parts of the bogie; the load correction coefficients are determined according to the force path type of the key load-bearing parts of the bogie in the current operating window, and the force path type includes at least one of bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load or wheel-rail abnormal excitation.

[0103] Specifically, in addition to structural correction factors and load path correction factors, the system also simultaneously introduces environmental correction factors and installation status correction factors. These four types of factors work together to complete a secondary correction, and the final equivalent stress amplitude calculation formula is as follows:

[0104] in: After multiple revisions, the first The bogie within the sliding window The final equivalent stress amplitude of each key component; Bogie No. The structural detail correction coefficients for key parts characterize local structural differences such as welds, openings, and abrupt changes in plate thickness. : No. Within the sliding window, the bogie... The load path correction coefficients for key components are matched to different stress forms such as bending-torsional coupling and turnout impact. : No. Within the sliding window, the bogie... Environmental correction factors for key components characterize the effects of temperature, season, and thermo-mechanical coupling. : No. Within the sliding window, the bogie... The installation status correction coefficient for key components characterizes the impact of component reinstallation, changes in preload, and replacement of accessories.

[0105] S304: Based on the final equivalent stress amplitude, calculate the estimated physical branch value of the key load-bearing part of the bogie.

[0106] Specifically, the Basquin fatigue life formula is first used to calculate the single-window low-fidelity fatigue life. The formula is as follows:

[0107] in: : No. Within the sliding window, the bogie... The low-fidelity fatigue life of a key component is output by a low-fidelity physical model. Bogie No. The life constant of the material-structure for each key component is determined by material properties, weld grade, and test database. Bogie No. The fatigue index for each key component is a fixed calibration parameter.

[0108] Specifically, the next step is to calculate the single-window low-fidelity damage increment, using the following formula:

[0109] in: : No. Within the sliding window, the bogie... $ Low-fidelity damage increment in key areas; : No. Within the sliding window, the bogie... The load path weighting coefficient of each key component represents the additional influence of loads such as curves, turnouts, traction and braking. : No. Within the sliding window, the bogie... The equivalent number of cycles borne by each key component.

[0110] Specifically, the cumulative damage is finally calculated based on Miner's linear cumulative damage rule, using the following formula:

[0111] in: As of the The first sliding window, the bogie's first... The cumulative low-fidelity damage value of each key component, with a maximum value limited to 1; As of the The first sliding window, the bogie's first... Historical cumulative low-fidelity damage values ​​for key components; : Minimum value function to prevent cumulative damage from exceeding the engineering limit value of 1.

[0112] Specifically, the system ultimately outputs two types of physical branch estimation results simultaneously: low-fidelity fatigue life and cumulative low-fidelity damage.

[0113] In an optional implementation, before collecting multi-source monitoring data during vehicle operation, the method further includes: pre-establishing a mapping relationship between the sensor installation location, the key load-bearing parts of the bogie, and the stress / strain surrogate quantity, and recording the structural attribute labels of the key load-bearing parts of the bogie.

[0114] Specifically, this step is a preliminary preparation for data collection. The recorded structural attribute labels include part category, component connection type, weld detail level, local plate thickness, reinforcement structure type, stiffness level of adjacent components, component area number, etc., which provide support for subsequent model parameter configuration and feature extraction.

[0115] The mapping relationship is used to extract stress-related signals corresponding to the key load-bearing parts of the bogie from the multi-source monitoring data, and the structural attribute label is used to determine the structural correction coefficient in the low-fidelity fatigue physical model.

[0116] Specifically, this mapping relationship is the core link in transforming raw monitoring data into effective stress and strain signals.

[0117] Specifically, the collected and recorded structural attribute labels will directly correspond to the $K_{str,h}$ structural detail correction coefficients in the model, achieving a precise match between the model parameters and the bogie's physical structure.

[0118] In an optional implementation, the structural attribute label includes the part category, connection type, weld detail level, local plate thickness, reinforcement type, stiffness level of adjacent components, and component area number of the bogie's critical load-bearing parts.

[0119] Specifically, the aforementioned labels can provide a complete and standardized digital representation of the structural characteristics of each key load-bearing part of the bogie, clearly distinguishing the structural differences and mechanical properties of different parts. This provides a structured basis for subsequent physical model parameter calibration, stress amplitude correction, and working condition domain matching, ensuring that fatigue life calculations can be adapted to the specific structural characteristics of different parts, and improving the relevance and accuracy of the evaluation results.

[0120] The structural correction coefficient is determined by reading at least one parameter value from the structural attribute label, including the connection type, weld detail level, local plate thickness, reinforcement type, and stiffness of the opening or transition zone of the key load-bearing parts of the bogie, and calculating or querying the structural correction coefficient based on the read parameter values.

[0121] Specifically, by using parameter reading combined with calculation or table lookup to determine the structural correction coefficients, the model parameters of different structural parts can be quickly adapted without the need for additional special tests and modeling. At the same time, each parameter corresponds to the stress concentration effect, size effect, fatigue notch effect and other influencing factors of the local structure, which can accurately quantify the effect of structural detail differences on fatigue performance and ensure that the corrected physical model results are more consistent with the actual fatigue characteristics of the component.

[0122] In one optional implementation, the source domain includes one or more of test bench samples, historical track samples, simulation samples, or measured samples of existing vehicle models; the target domain includes data domains corresponding to new tracks, new vehicle models, new seasons, new wheel and rail conditions, or new component installation conditions.

[0123] Specifically, the source domain relies on bench testing, long-term line operation, simulation modeling, and in-service vehicle testing to accumulate sufficient samples, from which the model can learn stable fatigue operation patterns.

[0124] Specifically, the target domains are all newly launched operating conditions, which generally suffer from the problems of scarce labeled samples and data distribution deviation, and are also the small sample adaptation scenarios that this solution focuses on solving.

[0125] In an optional implementation, the stress ratio correction operation uses Walker stress ratio correction parameters corresponding to the critical load-bearing parts of the bogie to correct the equivalent stress amplitude.

[0126] Specifically, Walker stress ratio correction parameters The bogie is independently calibrated for each key load-bearing component. The values ​​are determined by taking into account the properties of the component matrix material, the weld process grade, bench fatigue test data, and industry standard database. This can effectively eliminate the interference of different stress ratios and average stress levels on the calculation of equivalent stress amplitude, so that the correction results accurately match the actual fatigue mechanical properties of the corresponding parts.

[0127] The structural correction coefficient is determined based on the structural attribute label and is used to characterize the local structural differences corresponding to the connection form, weld detail level, local plate thickness, reinforcement form, and stiffness of openings or transition zones of the key load-bearing parts of the bogie.

[0128] Specifically, this coefficient is As a structural detail correction factor, it can quantify the stress concentration effect and size effect brought about by features such as local welds, openings, abrupt changes in plate thickness, and structural transition zones. It reflects the influence of structural morphology differences on load transfer and fatigue strength, enabling low-fidelity physical models to be adapted to key load-bearing parts with different structural features, and improving the part adaptability and accuracy of life calculation results.

[0129] In an optional implementation, the load path correction factor is used to characterize the impact of at least one of the following force path types on fatigue damage of the bogie's critical load-bearing components within the current operating window: bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load, or abnormal wheel-rail excitation.

[0130] Specifically, under different stress paths, the stress composite state, load impact characteristics, and fatigue damage accumulation rate of key load-bearing components of the bogie vary significantly. For example, the instantaneous impact load caused by turnout passage or abnormal wheel-rail excitation contributes far more to fatigue damage than under stable straight-line operation conditions. The load path correction coefficient can differentiate the weighting of the dominant stress form in the current window, accurately reflecting the actual impact of different load paths on fatigue damage, correcting the deviation of single equivalent stress amplitude calculation, and making the fatigue life and damage assessment results more consistent with the actual load characteristics of the line operation.

[0131] In an optional implementation, when correcting the equivalent stress amplitude, one or more of an environmental correction factor and an installation condition correction factor are also introduced.

[0132] Specifically, environmental correction factor Installation status correction factor It will be used in conjunction with structural correction factors and load correction factors to form a multi-dimensional equivalent stress amplitude correction system, further improving calculation accuracy.

[0133] The environmental correction factor is used to characterize temperature range, seasonal variation, or thermo-mechanical coupling effects.

[0134] Specifically, this coefficient This invention compensates for the thermo-mechanical coupling effect caused by seasonal changes and large fluctuations in ambient temperature, and solves the problem of deviations in fatigue load calculation under different temperature conditions.

[0135] The installation status correction coefficient is used to characterize factors such as reinstallation after maintenance, changes in pre-tightening status, or replacement of accessories.

[0136] Specifically, this coefficient It is designed to adapt to maintenance scenarios such as vehicle repair, component reassembly, changes in connection preload, and replacement of auxiliary parts, and to compensate for the impact of changes in assembly status on the stress state of key parts of the bogie.

[0137] In an optional implementation, the shared feature extraction model includes a first branch, a second branch, and a third branch.

[0138] Specifically, the model adopts a three-branch parallel structure design, with different branches handling different types of data features. This is the core structure for mining the common laws of fatigue across working conditions in the source domain.

[0139] The first branch is used to extract the temporal representation, the second branch is used to extract the frequency domain features, and the third branch is used to process the operating conditions and structural attribute features. The source domain shared representation vector is output after the features of each branch are fused.

[0140] Specifically, the first branch uses a temporal convolutional network, GRU (Gated Recurrent Unit), LSTM (Long Short-Term Memory) or a combination thereof to analyze the temporal change features in the monitoring data; the second branch extracts frequency domain features such as spectral peaks, spectral band energy, and spectral moments from the spectral data after FFT (Fast Fourier Transform).

[0141] Specifically, the third branch is dedicated to parsing the operating condition labels and component structural attribute labels. The features output from the upstream of the three branches are concatenated, weighted, or fused with attention to ultimately generate a source domain shared representation vector. .

[0142] Specifically, character definitions: For the first Within the sliding window, the bogie... The source domain shared representation vectors corresponding to each key component.

[0143] The input to the shared feature extraction model includes time-domain features, frequency-domain features, and operating condition labels extracted from the multi-source monitoring data, as well as structural attribute labels and load path labels for the key load-bearing parts of the bogie.

[0144] Specifically, the model input dimensions cover time-domain and frequency-domain features derived from multi-source data, while also forcibly incorporating operating condition labels, component structure labels, and load path labels to enhance the model's ability to learn about the bogie's specific structure, operating conditions, and stress patterns.

[0145] In an optional implementation, see Figure 4 As shown, Figure 4 The flowchart illustrates a fatigue life estimation result determination method provided in Embodiment 1 of this application, wherein the step of generating a target fatigue life estimation result for the key load-bearing component of the bogie based on the physical branch estimate, the migration life estimate, and the residual correction amount includes steps S401-S402: S401: The physical branch estimate, the migration lifetime estimate, and the residual correction are weighted and fused in the logarithmic lifetime domain to obtain the fusion result.

[0146] Specifically, the logarithmic lifetime domain fusion formula:

[0147] Weight constraints:

[0148] in: : No. Within the sliding window, the bogie... Estimated fatigue life of the target after fusion of key components; : No. Within a sliding window, the adaptive fusion weights corresponding to the low-fidelity physical branch results; : No. Within a sliding window, the adaptive fusion weights corresponding to the source domain migration branch results; : No. Within a sliding window, the adaptive fusion weights corresponding to the residual correction of the target domain; : No. Within a sliding window, the predicted migration lifetime output by the source domain shared feature branch; : No. Within the sliding window, the bogie... Lifetime residual correction for key components.

[0149] Specifically, it also supports damage domain fusion formulas to adapt to damage value calculation scenarios: in: : No. Within the sliding window, the bogie... Damage estimates of the target after fusion of key components; : No. Within a sliding window, the predicted migration damage output of the source domain shared feature branch; : No. Within the sliding window, the bogie... Damage residual correction for key components.

[0150] Specifically, the fusion weights are adaptive dynamic parameters that are adjusted in real time based on factors such as operating condition similarity, target domain sample coverage, sensor working quality, and consistency of physical results.

[0151] S402: Based on the fusion results, generate the target fatigue life estimation results for the key load-bearing parts of the bogie.

[0152] Specifically, the combined fatigue life value of each key component under the target working condition can be calculated by inverse logarithmic operation. This result will then be used in subsequent processes such as interval estimation, confidence assessment, and conservative backoff to form a complete output system.

[0153] See Figure 5As shown, Figure 5 A schematic diagram of a multi-fidelity fusion core module structure provided in Embodiment 1 of this application is shown. The diagram clearly illustrates three parallel branches, the fusion core processing flow, and the deployment and update strategy. The low-fidelity fatigue physics branch serves as the basic physical anchor point, sequentially performing the equivalent stress amplitude calculation, Walker stress ratio correction, and damage accumulation baseline calculation under the Miner and Basquin criteria, outputting a conservative life baseline and a low-fidelity damage estimate.

[0154] The shared feature and migration prediction branch, through sensor fusion encoding, operating condition encoding, cross-domain common pattern extraction, and source domain shared representation learning, outputs source domain migration lifetime prediction results and shared feature representations, providing the model with cross-operating condition migration capabilities. The target domain residual correction branch, based on few-sample calibration, bias correction, and uncertainty-sensitive correction learning, outputs residual correction δ and damage correction δD to eliminate distribution differences between the source and target domains.

[0155] The outputs of the three branches are input to the multi-fidelity fusion core module. This module takes the low-fidelity conservative baseline, migration prediction results and residual correction as inputs, performs weighted fusion in the logarithmic lifetime domain, and combines adaptive fusion weight allocation, confidence-aware interval generation and physical boundary consistency verification to obtain fusion lifetime estimates and interval estimates that can be directly applied in engineering and complete the physical consistency check.

[0156] The output of the fusion core is connected to the frozen bottom shared layer and the deployment and update module. This module keeps the bottom shared layer frozen during the actual deployment and iterative update phases, and only updates the residual head, interval head, and confidence head. It uses micro-batch samples from the target domain to achieve online updates without performing full retraining. The freezing layer strategy is jointly determined by the target domain sample size, working condition similarity, and path novelty.

[0157] During the deployment and operation phase, the module outputs online inference results after freezing the underlying shared layer, including updated residual correction, interval estimation and confidence assessment results, enabling rapid deployment and continuous accuracy correction in new working conditions and small sample scenarios. At the same time, the core output results are fed back to the deployment and update phase to form a closed-loop update mechanism for continuous optimization.

[0158] See Figure 6 As shown, Figure 6The diagram illustrates the steps of a fatigue life estimation method for a vehicle bogie provided in Embodiment 1 of this application. The diagram fully presents the execution logic of the entire process, from data processing to a closed-loop operation and maintenance system. First, multi-source monitoring data is collected, preprocessed, and spatiotemporally aligned. Simultaneously, the correspondence between sensor locations, hotspots, and stress-strain surrogate quantities is established, and structural attribute labels are added. Then, the source and target domains are divided according to the combination of key components and operating conditions. Finally, time-series samples are constructed using a sliding window, and component attributes and load path labels are added.

[0159] Based on the completed sample data, the low-fidelity fatigue life baseline is calculated by combining the part-level life parameters with the structural and load path correction. At the same time, the common fatigue laws of key parts are learned through the shared feature extraction module, and a shared feature representation applicable to cross-working conditions is constructed. Then, the residual correction branch is fine-tuned and optimized by using small samples of the target domain organized by key parts.

[0160] After completing basic calculations and feature learning, the system conducts similarity assessment and determines transfer learning strategies based on structural attribute similarity, load path similarity, and sample coverage. Subsequently, multi-fidelity fusion and correction processing is performed at the key part level to obtain the fused fatigue life and cumulative damage estimation results.

[0161] Furthermore, based on part-level rollback rules that match the maintenance window, lifespan estimation and confidence assessment are conducted. When the confidence level does not meet the requirements, a conservative rollback strategy is implemented, and the corresponding confidence level and lifespan results are output. Finally, based on the risk ranking of key parts and the maintenance priority determination results, an online incremental update process is initiated and an operation and maintenance decision output is generated, constructing a closed-loop system covering the entire process from data collection and lifespan estimation to operation and maintenance guidance.

[0162] In an optional implementation, the method further includes: determining whether the fusion result meets preset physical consistency constraints, wherein the physical consistency constraints include that fatigue life does not increase abnormally when stress amplitude increases, the fatigue life change rate of adjacent sliding windows does not exceed a reasonable range, and the fatigue life level of key load-bearing parts of similar bogies remains orderly.

[0163] Specifically, the core purpose of setting physical consistency constraints is to avoid extrapolation problems that violate the basic laws of metal fatigue in data-driven algorithms under small sample conditions, and to ensure that all calculation results conform to engineering common sense and fatigue mechanics principles.

[0164] When the fusion result does not meet the physical consistency constraint, at least one of the fusion weights corresponding to the physical branch estimate, the migration lifetime estimate, and the residual correction amount is adjusted, and the weighted fusion is re-executed based on the adjusted weights until the fusion result meets the physical consistency constraint.

[0165] Specifically, once a result is detected as violating physical constraints, the system will proactively increase the weight of the low-fidelity physical branch. Reduce the weight of the residual correction branch .

[0166] Specifically, after adjusting the weights, the weighted fusion calculation is carried out again, and the process is repeated until the result meets all physical constraints.

[0167] In an optional implementation, see Figure 7 As shown, Figure 7 A flowchart of a fatigue life estimation result determination method provided in Embodiment 1 of this application is shown, wherein the method further includes S701~S703: S701: Calculate the working condition similarity between the target domain and the source domain, and determine the comprehensive migration score according to the key load-bearing parts of the bogie, wherein the comprehensive migration score is composed of one or more weighted factors of structural attribute similarity, load path similarity, sample coverage and path novelty.

[0168] Specifically, the formula for calculating the comprehensive transfer score is as follows:

[0169] in: : No. Within the sliding window, the bogie... Comprehensive migration score for each key component; : Fixed weighting coefficients, corresponding to the weight percentages of structure, load, sample coverage, and novelty, respectively; : No. Within the sliding window, the bogie... Structural attribute similarity of key components; : No. Within the sliding window, the bogie... Load path similarity of key components; : No. Within the sliding window, the bogie... Target domain sample coverage of key parts; : No. Within the sliding window, the bogie... Path novelty (data distribution offset) of key parts.

[0170] Specifically, the calculation process is performed independently on a single bogie key load-bearing component, taking into account structural matching degree, load matching degree, sample coverage and data distribution offset, to quantify the adaptation relationship between the source domain and the target domain.

[0171] S702: Determine the freezing layer strategy based on the comprehensive migration score and the target domain sample size of the bogie's key load-bearing parts.

[0172] Specifically, the formula for calculating the decision quantity of the frozen layer is as follows:

[0173] in: : No. Within the sliding window, the bogie... The decision quantity of the frozen layer of each key part is used to determine the freezing / updating range of the model; : Fixed decision weighting coefficients, corresponding to the target domain sample size and the weight of the comprehensive transfer score, respectively; Bogie No. The total number of target domain labeled samples corresponding to each key part.

[0174] S703: According to the freezing layer strategy, freeze the bottom layer of the shared feature extraction model and only fine-tune the high-level residual branches, or perform a conservative update when the distribution offset is significant and the novelty of the key load-bearing parts of the bogie is too high.

[0175] Specifically, when the number of samples in the target domain is small and the operating conditions of the source domain and the target domain are highly similar ( When the value is too low, freeze the underlying network of the shared feature extraction model and only fine-tune the high-level residual branches to retain the general features learned from the source domain.

[0176] Specifically, when the data distribution is severely skewed and the working conditions are highly novel, the system only performs conservative update operations and does not conduct full model retraining throughout the process, thus avoiding overfitting caused by small samples at the source.

[0177] In an optional implementation, the method further includes: evaluating the confidence level of the target fatigue life estimation result; the confidence level is composed of one or more of the following: sample coverage, structural attribute similarity, load path similarity, path novelty, physical consistency satisfaction, and sensor quality.

[0178] Specifically, the confidence level is calculated using a multivariate function, expressed as follows: in: : No. Within the sliding window, the bogie... The overall confidence level of the prediction results for each key component; : Multivariate nonlinear mapping function to complete the fusion calculation of multi-dimensional indicators; : No. Within the sliding window, the bogie... Physical consistency satisfaction of results for key components; : No. Within the sliding window, the bogie... The key components correspond to the sensor's operational quality indicators.

[0179] Specifically, the computational dimensions include sample coverage, inter-domain similarity between structure and load, data novelty, physical constraint satisfaction, and sensor operating status, comprehensively evaluating the reliability of the prediction results.

[0180] When the confidence level is lower than a preset threshold, the target fatigue life estimation result is replaced with the smaller of the physical branch estimation value and the preset safe life threshold, and at least one of the following maintenance actions is triggered: resampling, increasing the retesting frequency, or outputting manual review suggestions.

[0181] Specifically, when the confidence level is insufficient, a conservative backoff logic is initiated, calculated using the following formula:

[0182] in: : No. Within the sliding window, the bogie... The final fatigue life value of a key component after conservative regression; Bogie No. The preset safe lifespan thresholds for each key component are set in conjunction with the vehicle maintenance cycle.

[0183] Specifically, the operation and maintenance system will simultaneously perform corresponding actions, including supplementing the collection of strain / stress verification samples, increasing the frequency of component retesting, and pushing manual review reminders, thus forming a closed-loop management of data collection, monitoring, and review.

[0184] In an optional implementation, see Figure 8 As shown, Figure 8The flowchart illustrates a fatigue life estimation result update method provided in Embodiment 1 of this application, wherein the method further includes S801~S804: S801: Obtain newly labeled samples for the target domain.

[0185] Specifically, as the vehicle continues to operate under the target conditions, it will continuously generate new monitoring data, inspection and repair data, and newly labeled samples such as crack and damage level tags. This type of data will be connected to the online update module in real time.

[0186] S802: Based on the newly labeled samples, the model branch used to determine the residual correction amount is updated online, wherein the bottom shared feature layer of the shared feature extraction model is kept frozen, and only the residual correction head, interval estimation head, confidence evaluation head and some high-level fusion weights are updated.

[0187] Specifically, the update process strictly preserves the underlying network of the shared feature extraction model and does not modify the cross-condition general features learned from the source domain.

[0188] Specifically, the residual correction head, lifetime interval estimation head, confidence assessment head, and some high-level fusion weights are updated only to significantly reduce the amount of computation and adapt to the computing power conditions of vehicle-mounted equipment.

[0189] S803: Determine the updated residual correction amount based on the updated model branch.

[0190] Specifically, after iterative optimization of local modules, the residual correction branch will output higher accuracy. Lifetime residual correction amount The damage residual correction gradually reduces the data distribution deviation between the source and target domains as new samples accumulate.

[0191] S804: Based on the physical branch estimate, the migration lifetime estimate, and the updated residual correction, generate the updated target fatigue lifetime estimate.

[0192] Specifically, by combining the latest residual correction values, the system re-executes processes such as multi-fidelity weighted fusion, confidence assessment, and life interval calculation, and outputs the fatigue life results after iterative optimization, so as to realize the continuous adaptive correction of the system according to the operating conditions.

[0193] In an optional implementation, see Figure 9 As shown, Figure 9 The flowchart illustrates a method for determining the input of a shared feature extraction model according to Embodiment 1 of this application, wherein the method further includes steps S901-S903: S901: Divide the continuous running data into discrete sample units according to time windows or mileage windows, and attach structural attribute labels and load path labels to the key load-bearing parts of the bogie to each sample unit.

[0194] Specifically, a sliding window method is used to segment the data, with a window duration of [missing information]. Set to 2s~30s, sliding step size Set to 0.5s~10s to split the continuous data stream into independent discrete sample units.

[0195] Specifically, character definitions: The duration of the sliding window. This represents the step size of the sliding window.

[0196] Specifically, the standard sample unit expression for a single sliding window is: .

[0197] Specifically, the definitions of each variable within the sample unit: For the first The first window A set of vibration characteristics of key components; This is the set of displacement characteristics for the corresponding location; A set of speed and running tags; A collection of temperature and environmental labels; For strain and strain proxy feature set; For structural attribute tags of key parts; Set the category label for the dominant load path in the current window.

[0198] Specifically, each sample unit is bound to the structural attribute label of the corresponding part and the dominant load path label of the current window, so as to realize the one-to-one correspondence between data samples and structural characteristics and load conditions, improve the sample dimensional information, and enhance the part adaptability and working condition targeting of subsequent model inference.

[0199] S902: The features extracted for each window include one or more of the following: root mean square, peak-to-peak value, kurtosis, dominant band energy, bandpass power spectral density amplitude, spectral moment, phase difference, temperature rise rate, mean velocity, and strain range.

[0200] Specifically, for each sliding window, three major categories of basic features are uniformly extracted: time domain, frequency domain, and statistical domain. Among them, the time domain features reflect the amplitude level and fluctuation pattern of the load, the frequency domain features reflect the spectral distribution and energy proportion of the load, and the statistical domain features quantify the distribution pattern and impact characteristics of the data. The above features are all typical and effective indicators for bogie fatigue state analysis, which can comprehensively characterize the load and operating status within a single window.

[0201] Specifically, the system also extracts additional features for specific operating conditions, including curve / straight section labels, turnout passing labels, braking / traction status labels, bending-torsional coupling features, impact load features, and relative phase features of adjacent measuring points near hotspots. This further enriches the feature dimensions, characterizes the differentiated impact of different line operating conditions and load forms on fatigue damage, and enhances the feature's ability to represent fatigue conditions.

[0202] S903: Use the extracted features as input to the shared feature extraction model.

[0203] Specifically, after feature extraction and label binding are completed, the standardized window feature set is directly input into the three branches of the shared feature extraction model to carry out feature parsing and source domain shared representation. calculate.

[0204] This solution includes a lifetime interval estimation module, with the corresponding formula as follows:

[0205] in: : No. Within the sliding window, the bogie... Lower limit of fatigue life range for key components; : No. Within the sliding window, the bogie... Upper limit of fatigue life range for key components; : Corresponding confidence level quantile coefficients; : No. Within the sliding window, the bogie... Uncertainty estimates of prediction results for key components.

[0206] This innovative solution constructs a multi-fidelity fusion architecture integrating a low-fidelity physics branch, a source domain shared feature branch, and a target domain residual correction branch, overcoming the design limitations of existing technologies that employ a single model. The three functional branches have clear divisions of labor and work in concert, simultaneously considering fatigue physics constraints, historical operational experience, and rapid adaptation to new operating conditions at the system architecture level. The low-fidelity fatigue physics branch runs continuously online throughout the entire system lifecycle, always serving as a reliable and conservative physics anchor, effectively addressing the problems of pure data-driven and general transfer learning schemes lacking physical constraints and prone to non-conservative predictions. This physics branch is not abandoned after model deployment, fundamentally avoiding prediction results that violate the basic laws of metal fatigue, and comprehensively improving the safety and stability of system operation.

[0207] This solution includes a model freezing layer strategy that dynamically manages the model's freezing region and update range by considering the number of samples in the target domain and the similarity between the source and target domains. This strategy can be flexibly adjusted according to the degree of deviation under different operating conditions, adapting to complex application scenarios such as new routes, new seasons, and new vehicle models, and possessing strong scenario compatibility. The system only fine-tunes the high-level residual branches, avoiding full model retraining. This effectively avoids the risk of overfitting in small-sample scenarios and reduces overall computational overhead, enabling rapid deployment and adaptation to new operating conditions. The lightweight update method adapts to the limited computing power of in-vehicle edge devices, while significantly shortening the deployment cycle for new operating conditions and improving on-site application efficiency.

[0208] This solution deeply integrates railway-specific operation and maintenance rules, such as confidence assessment, conservative backoff, supplementary sampling, manual verification, and maintenance prioritization, into the entire system's computational process. It breaks away from the traditional algorithm's single-mode output of numerical results, bridging the application barriers between algorithmic technology and on-site operation and maintenance, ensuring that technological achievements align with the industry's actual management logic. This solution transforms from single-algorithm prediction to a complete, directly implementable operation and maintenance system, significantly enhancing the engineering application value of the technology. The system output covers multiple types of executable information, including fatigue life values, lifespan ranges, confidence levels, operation and maintenance recommendations, and maintenance priorities, perfectly matching the operation and maintenance management needs of railway sites.

[0209] This solution features a lightweight online incremental update mechanism, focusing iterative optimization only on the residual correction header, interval estimation header, confidence assessment header, and some high-level fusion weights. The incremental update process does not interrupt normal system operation, supporting uninterrupted online monitoring and analysis, ensuring monitoring business continuity. The system relies on continuously accumulating new samples in the target domain to gradually reduce the data distribution differences between the source and target domains, ensuring the model's adaptability and prediction accuracy over the long term. As field samples are continuously added, model performance can be autonomously iterated and optimized, allowing the system to maintain good evaluation results throughout long-term service.

[0210] Based on the above methods, model architecture, system composition and technical advantages, the following three typical application examples will provide a detailed explanation of the practical application of this solution in actual rail transit operation and maintenance scenarios.

[0211] Example 1: Rapid Deployment in the Initial Stage of New Line Operation. This example uses the weld hotspots of a certain type of high-speed train bogie frame as the monitoring object to complete the system application in a new line scenario. This scenario is a typical small-sample new operating condition. After the train enters the new line, it is impossible to accumulate a sufficient number of labeled samples in a short period of time, which places high demands on the system's small-sample adaptability. In this example, the source domain uses test bench data, historical line operation data, and simulation data, while the target domain is a small amount of monitoring data collected in the first two weeks after the train enters the new line. The source domain has sufficient data reserves to support the shared feature extractor to complete the learning of common laws across operating conditions; the target domain only relies on short-term monitoring data for correction, which is consistent with the actual operation scenario. The system first completes multi-source data preprocessing, outputs a conservative lifetime baseline from the low-fidelity physical branch, and then obtains the migration lifetime result through the source domain shared feature extractor. The physical branch outputs conservative results throughout as a safety baseline, while the migration branch completes preliminary predictions based on historical data. The combination of these two steps ensures the effectiveness of the basic prediction. Finally, the system uses a small number of labeled samples in the target domain to train the residual correction head and confidence head, outputting the final lifetime results with intervals and confidence levels, enabling rapid deployment of new lines without a large number of labeled samples. Only the high-level correction module is fine-tuned without changing the underlying feature model, resulting in fast deployment speed while avoiding overfitting problems caused by small samples.

[0212] Example 2: Conservative Rollback under New Seasonal Operating Conditions. The vehicle operating environment enters a new season, with ambient temperature ranges significantly deviating from the source domain training range, and the number of effective samples in the target domain is insufficient. Seasonal changes bring multiple operating condition shifts, including temperature, thermo-mechanical coupling, and wheel-rail contact status; the scarcity of samples further reduces the reliability of prediction results. The system, after confidence assessment, determines that the current prediction results are unreliable and automatically selects a conservative value from the low-fidelity physical lifetime and the fused lifetime as the final output. The conservative rollback strategy prioritizes equipment operational safety, abandoning aggressive correction results and strictly adhering to fatigue physics. Simultaneously, the system pushes supplementary sampling and encrypted retesting suggestions to the maintenance team to ensure equipment operational safety under extreme conditions. Deeply linking prediction results with maintenance actions forms a closed loop of "monitoring-judgment-response," meeting the requirements of railway safety production management.

[0213] Example 3: Parallel Sorting and Tiered Maintenance of Multiple Fatigue Hotspots. Life assessments are conducted simultaneously on multiple fatigue hotspot areas of the same train bogie, enabling unified management of multiple parts. The bogie has several critical fatigue-prone parts; multi-target parallel assessment can comprehensively understand the health status of the entire bogie. The system independently calculates the lifespan value, upper and lower limits, confidence level, and regression status for each hotspot. Each critical part is calculated independently without data mixing, ensuring the accuracy of the assessment for each part. If the lower limit of a part's lifespan is shorter than the planned maintenance cycle and the confidence level is high, it is listed as a first-priority maintenance target. For high-risk, high-confidence assessment results, on-site maintenance is prioritized to avoid equipment failure risks. If a part has a high risk but a low confidence level, it is included in intensive monitoring and key retesting targets. For questionable assessment results, maintenance is not blindly scheduled; instead, data is supplemented through increased sampling and retesting to further verify the equipment status. The system ultimately outputs a complete maintenance priority list to guide on-site operation and maintenance work. The standardized priority list can be used directly by operations and maintenance personnel, reducing the difficulty of on-site judgment and improving operations and maintenance efficiency.

[0214] Example 2 See Figure 10 As shown, Figure 10 A schematic diagram of a vehicle bogie fatigue life estimation device provided in Embodiment 2 of this application is shown, wherein the device includes: Data acquisition module 1001 is used to collect multi-source monitoring data corresponding to key load-bearing parts of the vehicle bogie; The signal extraction module 1002 is used to extract stress-related signals from the multi-source monitoring data based on the mapping relationship between the sensor installation location, the key load-bearing parts of the bogie and the stress / strain surcharge. The physical estimation module 1003 is used to calculate the physical branch estimate based on the stress-related signal and the low-fidelity fatigue physical model. The migration prediction module 1004 is used to combine multi-source monitoring data, structural attribute labels of key load-bearing parts of the bogie, load path labels, and a shared feature extraction model trained in the source domain to determine the migration life estimate, wherein the source domain is the working condition domain of existing samples. The residual correction module 1005 is used to determine the residual correction amount corresponding to the target working condition based on the target domain annotation samples organized according to the key load-bearing parts of the bogie. The target domain is the data domain corresponding to the new working condition to be launched. The fusion output module 1006 is used to generate fatigue life estimation results of key load-bearing parts of the bogie under target working conditions based on the physical branch estimate, migration life estimate and residual correction amount.

[0215] In one optional implementation, the key load-bearing parts of the bogie include at least one of the following: frame weld hot spot area, traction rod seat connection area, axle box positioning connection area, shock absorber mounting seat, air spring seat, brake hanger seat, motor hanger seat, lateral stop installation area, and adjacent transition area.

[0216] In one alternative implementation, the multi-source monitoring data includes at least three of the following: vibration, displacement or relative displacement, vehicle speed, mileage, temperature, strain signal, and stress proxy signal.

[0217] In an optional implementation, the low-fidelity fatigue physical model includes at least one or more of the following: equivalent stress amplitude calculation, stress ratio correction, fatigue life relationship, and cumulative damage calculation.

[0218] In an optional implementation, the physical estimation module is specifically used for: Based on the stress-related signals, the equivalent stress amplitude of the key load-bearing parts of the bogie within the sliding window is determined; A stress ratio correction operation is performed on the equivalent stress amplitude to obtain the corrected equivalent stress amplitude, wherein the stress ratio correction operation adopts the Walker correction method; By introducing structural correction coefficients and load correction coefficients corresponding to the key load-bearing parts of the bogie, the corrected equivalent stress amplitude is modified to obtain the final equivalent stress amplitude; The structural correction coefficient is determined based on at least one of the connection form, weld detail level, local plate thickness, reinforcement form, and stiffness of openings or transition zones of the key load-bearing parts of the bogie; the load correction coefficient is determined based on the force path type of the key load-bearing parts of the bogie within the current operating window, and the force path type includes at least one of bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load, or abnormal wheel-rail excitation. Based on the final equivalent stress amplitude, the physical branch estimate of the key load-bearing parts of the bogie is calculated.

[0219] In an optional implementation, the device further includes a pre-configuration module for pre-establishing a mapping relationship between the sensor installation location, the key load-bearing parts of the bogie, and the stress / strain surcharge before collecting multi-source monitoring data during vehicle operation, and recording the structural attribute labels of the key load-bearing parts of the bogie. The mapping relationship is used to extract stress-related signals corresponding to the key load-bearing parts of the bogie from the multi-source monitoring data, and the structural attribute label is used to determine the structural correction coefficient in the low-fidelity fatigue physical model.

[0220] In an optional implementation, the structural attribute label includes the part category, connection type, weld detail level, local plate thickness, reinforcement type, stiffness level of adjacent components, and component area number of the bogie's critical load-bearing parts; The structural correction coefficient is determined by reading at least one parameter value from the structural attribute label, including the connection type, weld detail level, local plate thickness, reinforcement type, and stiffness of the opening or transition zone of the key load-bearing parts of the bogie, and calculating or querying the structural correction coefficient based on the read parameter values.

[0221] In one optional implementation, the source domain includes one or more of test bench samples, historical track samples, simulation samples, or measured samples of existing vehicle models; the target domain includes data domains corresponding to new tracks, new vehicle models, new seasons, new wheel and rail conditions, or new component installation conditions.

[0222] In an optional implementation, the stress ratio correction operation uses the Walker stress ratio correction parameter corresponding to the key load-bearing part of the bogie to correct the equivalent stress amplitude; the structural correction coefficient is determined based on the structural attribute label and is used to characterize the connection form, weld detail level, local plate thickness, reinforcement form, and local structural differences corresponding to the stiffness of openings or transition zones of the key load-bearing part of the bogie.

[0223] In an optional implementation, the load path correction factor is used to characterize the impact of at least one of the following force path types on fatigue damage of the bogie's critical load-bearing components within the current operating window: bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load, or abnormal wheel-rail excitation.

[0224] In an optional implementation, when the physical estimation module corrects the equivalent stress amplitude, it also introduces one or more of an environmental correction coefficient and an installation state correction coefficient; the environmental correction coefficient is used to characterize temperature range, seasonal variation, or thermo-mechanical coupling effects; the installation state correction coefficient is used to characterize factors such as reinstallation after maintenance, changes in pre-tightening status, or accessory replacement.

[0225] In an optional implementation, the shared feature extraction model includes a first branch, a second branch, and a third branch; the first branch is used to extract time-series representations, the second branch is used to extract frequency-domain features, and the third branch is used to process operating condition and structural attribute features. The features from each branch are fused to output a source-domain shared representation vector. The input to the shared feature extraction model includes time-domain features, frequency-domain features, operating condition labels extracted from the multi-source monitoring data, as well as structural attribute labels and load path labels for the key load-bearing parts of the bogie.

[0226] In an optional implementation, the fusion output module is specifically used for: The physical branch estimate, the migration lifetime estimate, and the residual correction are weighted and fused in the logarithmic lifetime domain to obtain the fusion result; Based on the fusion results, the target fatigue life estimation results of the key load-bearing parts of the bogie are generated.

[0227] In an optional implementation, the apparatus further includes a consistency constraint module for: Determine whether the fusion result meets the preset physical consistency constraints, wherein the physical consistency constraints include that the fatigue life does not increase abnormally when the stress amplitude increases, the fatigue life change rate of adjacent sliding windows does not exceed a reasonable range, and the fatigue life level of key load-bearing parts of the same type of bogie remains orderly. When the fusion result does not meet the physical consistency constraint, at least one of the fusion weights corresponding to the physical branch estimate, the migration lifetime estimate, and the residual correction amount is adjusted, and the fusion output module is triggered to re-execute the weighted fusion based on the adjusted weights until the fusion result meets the physical consistency constraint.

[0228] In an optional implementation, the apparatus further includes a freezing strategy module for: Calculate the working condition similarity between the target domain and the source domain, and determine the comprehensive migration score according to the key load-bearing parts of the bogie, wherein the comprehensive migration score is composed of one or more weighted factors of structural attribute similarity, load path similarity, sample coverage and path novelty; The freezing layer strategy is determined based on the comprehensive migration score and the target domain sample size of the key load-bearing parts of the bogie. According to the freezing layer strategy, the bottom layer of the shared feature extraction model is frozen and only the high-level residual branches are fine-tuned, or a conservative update is performed when the distribution offset is significant and the novelty of the key load-bearing parts of the bogie is too high.

[0229] In an optional implementation, the apparatus further includes a confidence assessment module for: The confidence level of the target fatigue life estimation result is evaluated; the confidence level is composed of one or more of the following: sample coverage, structural attribute similarity, load path similarity, path novelty, physical consistency satisfaction, and sensor quality. When the confidence level is lower than a preset threshold, the target fatigue life estimation result is replaced with the smaller of the physical branch estimation value and the preset safe life threshold, and at least one of the following maintenance actions is triggered: resampling, increasing the retesting frequency, or outputting manual review suggestions.

[0230] In an optional implementation, the device further includes an online update module for: Obtain newly labeled samples for the target domain; Based on the newly labeled samples, the model branch used to determine the residual correction amount is updated online. The bottom shared feature layer of the shared feature extraction model is kept frozen, and only the residual correction head, interval estimation head, confidence evaluation head and some high-level fusion weights are updated. The updated residual correction amount is determined based on the updated model branch; The fusion output module is triggered to generate an updated target fatigue life estimate based on the physical branch estimate, the migration lifetime estimate, and the updated residual correction.

[0231] In an optional implementation, the apparatus further includes a feature extraction module for: The continuous operation data is divided into discrete sample units according to time windows or mileage windows, and structural attribute labels and load path labels of the key load-bearing parts of the bogie are attached to each sample unit. The features extracted for each window include one or more of the following: root mean square, peak-to-peak value, kurtosis, dominant band energy, bandpass power spectral density amplitude, spectral moment, phase difference, temperature rise rate, mean velocity, and strain range. The extracted features are used as input to the shared feature extraction model.

[0232] Example 3 Based on the same application concept, see [link / reference] Figure 11 As shown, Figure 11 This illustration shows a structural schematic diagram of a computer device provided in Embodiment 3 of this application, wherein, as shown... Figure 11 As shown, the computer device 1100 provided in Embodiment 3 of this application includes: The computer device 1100 includes a processor 1101, a memory 1102, and a bus 1103. The memory 1102 stores machine-readable instructions that can be executed by the processor 1101. When the computer device 1100 is running, the processor 1101 and the memory 1102 communicate via the bus 1103. When the machine-readable instructions are executed by the processor 1101, the steps of the vehicle bogie fatigue life estimation method shown in Embodiment 1 above are executed.

[0233] Example 4 Based on the same concept, this application also provides a computer-readable storage medium storing a computer program, which, when run by a processor, performs the steps of the vehicle bogie fatigue life estimation method described in any of the above embodiments.

[0234] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0235] The computer program product for estimating the fatigue life of a vehicle bogie provided in this application includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0236] The vehicle bogie fatigue life estimation device provided in this application embodiment can be specific hardware on the device or software or firmware installed on the device. The device provided in this application embodiment has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0237] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0238] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0239] In addition, the functional units in the embodiments provided in this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0240] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0241] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0242] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method for estimating the fatigue life of a vehicle bogie, characterized in that, The method includes: Collect multi-source monitoring data corresponding to key load-bearing components of the bogie during vehicle operation; Based on the mapping relationship between sensor installation location, key load-bearing parts of the bogie and stress / strain surrogate, stress-related signals corresponding to the key load-bearing parts of the bogie are extracted from the multi-source monitoring data. Based on the stress-related signals and the low-fidelity fatigue physical model used for online estimation, the physical branch estimates of the key load-bearing parts of the bogie are determined. Based on the multi-source monitoring data, structural attribute labels and load path labels of the key load-bearing parts of the bogie, and the shared feature extraction model trained by the source domain, the estimated migration life of the key load-bearing parts of the bogie is determined, wherein the source domain is the working condition domain of the existing samples. Based on the labeled samples organized according to the key load-bearing parts of the bogie in the target domain, the residual correction amount for the target working condition is determined, wherein the target domain is the data domain corresponding to the new working condition to be launched. Based on the physical branch estimate, the migration life estimate, and the residual correction, the target fatigue life estimate of the key load-bearing components of the bogie under the target working condition is generated.

2. The method according to claim 1, characterized in that, The key load-bearing parts of the bogie include at least one of the following: frame weld hot spot area, traction rod seat connection area, axle box positioning connection area, shock absorber mounting seat, air spring seat, brake hanger seat, motor hanger seat, lateral stop installation area and its adjacent transition area.

3. The method according to claim 1, characterized in that, The multi-source monitoring data includes at least three of the following: vibration, displacement or relative displacement, vehicle speed, mileage, temperature, strain signal, and stress proxy signal.

4. The method according to claim 1, characterized in that, The low-fidelity fatigue physical model includes at least one or more of the following: equivalent stress amplitude calculation, stress ratio correction, fatigue life relationship, and cumulative damage calculation.

5. The method according to claim 1, characterized in that, The determination of the physical branch estimates of the key load-bearing components of the bogie based on the stress-related signals and the low-fidelity fatigue physical model used for online estimation includes: Based on the stress-related signals, the equivalent stress amplitude of the key load-bearing parts of the bogie within the sliding window is determined; A stress ratio correction operation is performed on the equivalent stress amplitude to obtain the corrected equivalent stress amplitude, wherein the stress ratio correction operation adopts the Walker correction method; By introducing structural correction coefficients and load correction coefficients corresponding to the key load-bearing parts of the bogie, the corrected equivalent stress amplitude is modified to obtain the final equivalent stress amplitude; The structural correction coefficient is determined based on at least one of the connection form, weld detail level, local plate thickness, reinforcement form, and stiffness of openings or transition zones of the key load-bearing parts of the bogie; the load correction coefficient is determined based on the force path type of the key load-bearing parts of the bogie within the current operating window, and the force path type includes at least one of bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load, or abnormal wheel-rail excitation. Based on the final equivalent stress amplitude, the physical branch estimate of the key load-bearing parts of the bogie is calculated.

6. The method according to claim 1, characterized in that, Before collecting multi-source monitoring data during vehicle operation, the method further includes: A mapping relationship is pre-established between the sensor installation location, the key load-bearing parts of the bogie, and the stress / strain surrogate quantity, and the structural attribute labels of the key load-bearing parts of the bogie are recorded; The mapping relationship is used to extract stress-related signals corresponding to the key load-bearing parts of the bogie from the multi-source monitoring data, and the structural attribute label is used to determine the structural correction coefficient in the low-fidelity fatigue physical model.

7. The method according to claim 6, characterized in that, The structural attribute labels include the part category, connection type, weld detail level, local plate thickness, reinforcement type, stiffness level of adjacent components, and component area number of the key load-bearing parts of the bogie. The structural correction coefficient is determined by reading at least one parameter value from the structural attribute label, including the connection type, weld detail level, local plate thickness, reinforcement type, and stiffness of the opening or transition zone of the key load-bearing parts of the bogie, and calculating or querying the structural correction coefficient based on the read parameter values.

8. The method according to claim 1, characterized in that, The source domain includes one or more of the following: test bench samples, historical route samples, simulation samples, or measured samples of existing vehicle models; the target domain includes the data domain corresponding to new routes, new vehicle models, new seasons, new wheel and rail conditions, or new component installation conditions.

9. The method according to claim 5, characterized in that, The stress ratio correction operation uses the Walker stress ratio correction parameter corresponding to the key load-bearing part of the bogie to correct the equivalent stress amplitude; The structural correction coefficient is determined based on the structural attribute label and is used to characterize the local structural differences corresponding to the connection form, weld detail level, local plate thickness, reinforcement form, and stiffness of openings or transition zones of the key load-bearing parts of the bogie.

10. The method according to claim 5, characterized in that, The load path correction coefficient is used to characterize the impact of at least one of the following force path types on fatigue damage in the current operating window: bending-torsional coupling, curve passage, turnout passage, traction and braking, impact load, or abnormal wheel-rail excitation on the key load-bearing parts of the bogie.

11. The method according to claim 5, characterized in that, When correcting the equivalent stress amplitude, one or more of the environmental correction factor and the installation state correction factor are also introduced; The environmental correction factor is used to characterize the effects of temperature range, seasonal variation, or thermo-mechanical coupling. The installation status correction coefficient is used to characterize factors such as reinstallation after maintenance, changes in pre-tightening status, or replacement of accessories.

12. The method according to claim 1, characterized in that, The shared feature extraction model includes a first branch, a second branch, and a third branch; The first branch is used to extract time-series representations, the second branch is used to extract frequency domain features, and the third branch is used to process operating conditions and structural attribute features. The source domain shared representation vector is output after the features of each branch are fused. The input to the shared feature extraction model includes time-domain features, frequency-domain features, and operating condition labels extracted from the multi-source monitoring data, as well as structural attribute labels and load path labels for the key load-bearing parts of the bogie.

13. The method according to claim 1, characterized in that, The process of generating target fatigue life estimates for the key load-bearing components of the bogie based on the physical branch estimate, the migration life estimate, and the residual correction includes: The physical branch estimate, the migration lifetime estimate, and the residual correction are weighted and fused in the logarithmic lifetime domain to obtain the fusion result; Based on the fusion results, the target fatigue life estimation results of the key load-bearing parts of the bogie are generated.

14. The method according to claim 13, characterized in that, The method further includes: Determine whether the fusion result meets the preset physical consistency constraints, wherein the physical consistency constraints include that the fatigue life does not increase abnormally when the stress amplitude increases, the fatigue life change rate of adjacent sliding windows does not exceed a reasonable range, and the fatigue life level of key load-bearing parts of the same type of bogie remains orderly. When the fusion result does not meet the physical consistency constraint, at least one of the fusion weights corresponding to the physical branch estimate, the migration lifetime estimate, and the residual correction amount is adjusted, and the weighted fusion is re-executed based on the adjusted weights until the fusion result meets the physical consistency constraint.

15. The method according to claim 1, characterized in that, The method further includes: Calculate the working condition similarity between the target domain and the source domain, and determine the comprehensive migration score according to the key load-bearing parts of the bogie, wherein the comprehensive migration score is composed of one or more weighted factors of structural attribute similarity, load path similarity, sample coverage and path novelty; The freezing layer strategy is determined based on the comprehensive migration score and the target domain sample size of the key load-bearing parts of the bogie. According to the freezing layer strategy, the bottom layer of the shared feature extraction model is frozen and only the high-level residual branches are fine-tuned, or a conservative update is performed when the distribution offset is significant and the novelty of the key load-bearing parts of the bogie is too high.

16. The method according to claim 1, characterized in that, The method further includes: The confidence level of the target fatigue life estimation result is evaluated; the confidence level is composed of one or more of the following: sample coverage, structural attribute similarity, load path similarity, path novelty, physical consistency satisfaction, and sensor quality. When the confidence level is lower than a preset threshold, the target fatigue life estimation result is replaced with the smaller of the physical branch estimation value and the preset safe life threshold, and at least one of the following maintenance actions is triggered: resampling, increasing the retesting frequency, or outputting manual review suggestions.

17. The method according to claim 1, characterized in that, The method further includes: Obtain newly labeled samples for the target domain; Based on the newly labeled samples, the model branch used to determine the residual correction amount is updated online. The bottom shared feature layer of the shared feature extraction model is kept frozen, and only the residual correction head, interval estimation head, confidence evaluation head and some high-level fusion weights are updated. The updated residual correction amount is determined based on the updated model branch; Based on the physical branch estimate, the migration lifetime estimate, and the updated residual correction, an updated target fatigue lifetime estimate is generated.

18. The method according to claim 1, characterized in that, The method further includes: The continuous operation data is divided into discrete sample units according to time windows or mileage windows, and structural attribute labels and load path labels of the key load-bearing parts of the bogie are attached to each sample unit. The features extracted for each window include one or more of the following: root mean square, peak-to-peak value, kurtosis, dominant band energy, bandpass power spectral density amplitude, spectral moment, phase difference, temperature rise rate, mean velocity, and strain range. The extracted features are used as input to the shared feature extraction model.

19. A device for estimating the fatigue life of a vehicle bogie, characterized in that, The device includes: The data acquisition module is used to collect multi-source monitoring data corresponding to key load-bearing components of the vehicle bogie; The signal extraction module is used to extract stress-related signals from the multi-source monitoring data based on the mapping relationship between sensor installation location, key load-bearing parts of the bogie and stress / strain surcharge. The physical estimation module is used to calculate the physical branch estimate based on the stress-related signal and the low-fidelity fatigue physical model. The migration prediction module is used to combine multi-source monitoring data, structural attribute labels of key load-bearing parts of the bogie, load path labels, and a shared feature extraction model trained in the source domain to determine the migration lifetime estimate, wherein the source domain is the operating condition domain of existing samples. The residual correction module is used to determine the residual correction amount corresponding to the target working condition based on the target domain annotation samples organized according to the key load-bearing parts of the bogie. The target domain is the data domain corresponding to the new working condition to be launched. The fusion output module is used to generate fatigue life estimation results for key load-bearing components of the bogie under target working conditions based on the physical branch estimate, migration life estimate, and residual correction.

20. A computer device, characterized in that, include: The computer device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the vehicle bogie fatigue life estimation method as described in any one of claims 1 to 18.

21. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the vehicle bogie fatigue life estimation method as described in any one of claims 1 to 18.

Images