A method and device for evaluating a vehicle component, an electronic device and a storage medium

Abstract

This application provides a method, apparatus, electronic device, and storage medium for evaluating vehicle components. The method divides collected multi-source monitoring data into multiple monitoring windows according to a preset process sequence and constructs a historical sequence. The historical sequence is then temporally degraded to obtain a degradation coding vector. The multi-source monitoring data is converted into a multi-axis fatigue characterization sequence independent of local coordinate selection. Finally, based on the degradation coding vector and the multi-axis fatigue characterization sequence, the fatigue damage and remaining life of the vehicle component under physical constraints are evaluated. This application can effectively distinguish between accidental impacts and continuous degradation, eliminate the influence of sensor installation posture differences, ensure that the evaluation results conform to the laws of fatigue physics, and output damage increment, cumulative damage, and remaining life range, providing a reliable decision-making basis for predictive maintenance of key vehicle components.

Description

Technical Field

[0001] This application relates to the field of vehicle evaluation technology, and more specifically, to a method, apparatus, electronic device, and storage medium for evaluating vehicle components. Background Technology

[0002] During the long-term service of rail vehicles, especially high-speed trains, critical load-bearing components such as bogies are subjected to the coupled effects of wheel-rail excitation, track irregularities, and complex environmental loads, making them prone to high-cycle fatigue damage. Accurately assessing the fatigue damage state of vehicle components and predicting their remaining life is crucial for ensuring operational safety and developing predictive maintenance plans.

[0003] In existing technologies, various methods for assessing the fatigue life of vehicle components have been proposed. For example, some schemes extract features from dynamic response signals such as acceleration and strain within a single monitoring window to directly estimate the current damage level or remaining life. The fundamental problem is that fatigue damage is essentially a cumulative process related to load history, rather than a function of a single instantaneous response, leading to inaccurate assessment results. Another example is that some schemes directly input multi-source monitoring data into data-driven models such as neural networks to output life prediction values. However, such pure black-box models are prone to producing results that violate physical consistency in small sample sizes or scenarios involving shifting operating conditions. For instance, damage may decrease with increasing stress, and their output format is singular, making it difficult to support engineering maintenance decisions. Furthermore, existing schemes suffer from differences in sensor installation orientation and measurement point coordinate definitions, and the original signal components are easily affected by the measurement posture, resulting in decreased model generalization ability and difficulty in adapting to different vehicles or cross-line applications. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a method, apparatus, electronic device and storage medium for evaluating vehicle components to overcome the problems in the prior art.

[0005] In a first aspect, embodiments of this application provide a method for evaluating vehicle components, the method comprising: According to the preset process sequence, the multi-source monitoring data collected during the operation of the vehicle components to be evaluated are constructed into a historical sequence; The historical sequence is subjected to temporal degradation encoding to obtain a degradation encoding vector; The multi-source monitoring data is converted into a multi-axis fatigue characterization sequence that is independent of the selection of local coordinates. Based on the degradation encoding vector and the multi-axis fatigue characterization sequence, the fatigue damage and remaining life of the vehicle component to be evaluated are assessed under physical constraints.

[0006] In some technical solutions of this application, the above methods collect monitoring data in the following ways, including: According to the time sequence, mileage sequence, or state sequence of the vehicle components to be evaluated, at least two of the following types of data are collected: bogie frame acceleration, axle box acceleration, displacement or relative displacement, key hot spot strain or stress proxy signal, operating speed, track type, ambient temperature, and mileage information.

[0007] In some technical solutions of this application, the multi-source monitoring data collected during the operation of the vehicle device to be evaluated is constructed into a historical sequence according to a preset process order, including: The current monitoring window and its preceding multiple monitoring windows are arranged sequentially according to the process order to form a historical sequence that reflects the degradation trend from the early state to the current state.

[0008] In some technical solutions of this application, the above-mentioned temporal degradation encoding of the historical sequence to obtain a degradation encoding vector includes: The historical sequence is input into a temporal degradation encoder to extract degradation encoding vectors that characterize the direction of structural performance degradation, rate of change, and state evolution trend; the temporal degradation encoder is implemented using at least one of a one-dimensional convolutional network, a gated recurrent unit, a long short-term memory network, or a temporal transformer network. When using a Long Short-Term Memory (LSTM) network, a two-layer LSTM network is used. The historical sequence containing the current window and its previous windows is sequentially input into the first layer of the LSTM network according to the process order to obtain the hidden state sequence output by the first layer. The hidden state sequence is input into the second layer of the long short-term memory network to obtain the hidden state vector output by the second layer. The hidden state vector output by the second layer is used as the degenerate coding vector.

[0009] In some technical solutions of this application, the above-mentioned conversion of the multi-source monitoring data into a multi-axis fatigue characterization sequence independent of local coordinate selection includes: The stress or strain components in the multi-source monitoring data are extracted and converted into fatigue equivalent parameters that are invariant to coordinate rotation, so as to eliminate the projection influence of sensor installation posture differences or local coordinate differences of measuring points on the evaluation results; the fatigue equivalent parameters include at least one of equivalent stress, principal stress surrogate quantity, key plane candidate parameter or Ilyusin path parameter; When the fatigue equivalent parameters include equivalent stress, the equivalent stress time series is calculated according to the equivalent stress formula with coordinate rotation invariance, and the equivalent stress amplitude of each window is extracted as a unit of monitoring window, as part of the multiaxial fatigue characterization sequence.

[0010] In some technical solutions of this application, the above-mentioned evaluation of fatigue damage and remaining life of the vehicle component under physical constraints based on the degradation encoding vector and the multi-axis fatigue characterization sequence includes: Based on the equivalent stress amplitude and load path non-proportional parameters in the multi-axis fatigue characterization sequence, the degradation state quantities in the degradation encoding vector, and the operating speed, ambient temperature, and circuit type of the vehicle components to be evaluated, a hotspot-sensitive damage driving parameter is constructed by fusing it into a single-valued scalar according to the power-law relationship and nonlinear correction. This parameter reflects the structural differences of key parts, the degree of load non-proportionality, and the influence of historical degradation and operating conditions on damage evolution. The multi-axis fatigue characterization sequence, the hotspot-sensitive damage driving parameters, and the degradation encoding vector are jointly input into the physical constraint damage solution model. The equivalent cycle number is obtained by performing equivalent cycle counting on the multi-axis fatigue characterization sequence. The fatigue life is calculated according to the logarithmic power law relationship based on the hotspot-sensitive damage driving parameters. The damage increment and cumulative damage of the current monitoring window are solved using the equivalent cycle number and the fatigue life. Based on the statistical distribution of the cumulative damage in the current monitoring window and the damage increment in subsequent windows, the remaining lifetime interval from the time corresponding to the current monitoring window is calculated. The current monitoring window damage increment, cumulative damage, and remaining lifetime interval are used as the evaluation results for the vehicle device to be evaluated.

[0011] In some technical solutions of this application, the above method also includes: Determine the confidence level of the evaluation results; A maintenance trigger signal is generated based on the cumulative damage in the current monitoring window, the remaining lifespan, and the confidence level.

[0012] Secondly, embodiments of this application provide an apparatus for evaluating vehicle components, the apparatus comprising: The data acquisition module is used to construct a historical sequence from the multi-source monitoring data collected during the operation of the vehicle components to be evaluated, according to a preset process sequence. The degradation coding module is used to perform temporal degradation coding on the historical sequence to obtain a degradation coding vector; The conversion module is used to convert the multi-source monitoring data into a multi-axis fatigue characterization sequence that is independent of the selection of local coordinates; The evaluation module is used to evaluate the fatigue damage and remaining life of the vehicle component to be evaluated under physical constraints, based on the degradation encoding vector and the multi-axis fatigue characterization sequence.

[0013] Thirdly, embodiments of this application provide an electronic device, a processor, a memory, and a bus. The memory stores machine instructions executed by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine instructions are executed by the processor, the steps of the above-described vehicle device evaluation method are performed.

[0014] Fourthly, embodiments of this application provide a computer storage medium storing a computer program, which, when run by a processor, executes the steps of the above-described vehicle device evaluation method.

[0015] The technical solutions provided by the embodiments of this application may include the following beneficial effects: This application provides a method for evaluating vehicle components. The method includes: constructing a historical sequence from multi-source monitoring data collected during the operation of the vehicle component to be evaluated according to a preset process sequence; performing temporal degradation encoding on the historical sequence to obtain a degradation encoding vector; converting the multi-source monitoring data into a multi-axis fatigue characterization sequence that is independent of the selection of local coordinates; and evaluating the fatigue damage and remaining life of the vehicle component to be evaluated under physical constraints based on the degradation encoding vector and the multi-axis fatigue characterization sequence.

[0016] This application divides multi-source monitoring data into multiple monitoring windows and constructs a historical sequence according to a preset process order. This allows the evaluation process to effectively distinguish between incidental impact responses and continuous performance degradation by utilizing the state evolution trends reflected in the current window and previous windows, thus improving the accuracy of fatigue damage assessment. By converting multi-source monitoring data into a multi-axis fatigue characterization sequence independent of local coordinate selection, the projection influence of sensor installation posture differences and local coordinate differences of measurement points on the evaluation results is eliminated, enhancing the model's generalization ability under different vehicles, different measurement points, and cross-line application scenarios. By evaluating fatigue damage and remaining life based on degradation encoding vectors and multi-axis fatigue characterization sequences under physical constraints, the evaluation results satisfy the mechanical laws that damage increases monotonically with driving parameters and life decreases monotonically with driving parameters. This avoids the results that violate physical consistency produced by pure data-driven models when samples are scarce or operating conditions are extrapolated, enhancing the engineering credibility and interpretability of the evaluation results and providing a more comprehensive basis for judging the structural health status of vehicle components.

[0017] 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

[0018] 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.

[0019] Figure 1A flowchart illustrating a method for evaluating vehicle components provided in an embodiment of this application is shown. Figure 2 A schematic diagram of a vehicle device evaluation apparatus provided in an embodiment of this application is shown; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] 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. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0021] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically 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 to illustrate selected embodiments of the 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.

[0022] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0023] During the long-term service of rail vehicles, especially high-speed trains, critical load-bearing components such as bogies are subjected to the coupled effects of wheel-rail excitation, track irregularities, and complex environmental loads, making them prone to high-cycle fatigue damage. Accurately assessing the fatigue damage state of vehicle components and predicting their remaining life is crucial for ensuring operational safety and developing predictive maintenance plans.

[0024] In existing technologies, various methods for assessing the fatigue life of vehicle components have been proposed. For example, some schemes extract features from dynamic response signals such as acceleration and strain within a single monitoring window to directly estimate the current damage level or remaining life. The fundamental problem is that fatigue damage is essentially a cumulative process related to load history, rather than a function of a single instantaneous response, leading to inaccurate assessment results. Another example is that some schemes directly input multi-source monitoring data into data-driven models such as neural networks to output life prediction values. However, such pure black-box models are prone to producing results that violate physical consistency in small sample sizes or scenarios involving shifting operating conditions. For instance, damage may decrease with increasing stress, and their output format is singular, making it difficult to support engineering maintenance decisions. Furthermore, existing schemes suffer from differences in sensor installation orientation and measurement point coordinate definitions, and the original signal components are easily affected by the measurement posture, resulting in decreased model generalization ability and difficulty in adapting to different vehicles or cross-line applications.

[0025] Based on this, embodiments of this application provide a method, apparatus, electronic device, and storage medium for evaluating vehicle components, which are described below through embodiments.

[0026] Figure 1 The diagram illustrates a flowchart of a method for evaluating vehicle components according to an embodiment of this application, wherein the method includes steps S101-S104; specifically: S101. According to the preset process sequence, construct a historical sequence from the multi-source monitoring data collected during the operation of the vehicle device to be evaluated; S102. Perform temporal degradation encoding on the historical sequence to obtain a degradation encoding vector; S103. Convert the multi-source monitoring data into a multi-axis fatigue characterization sequence that is independent of the selection of local coordinates; S104. Based on the degradation encoding vector and the multi-axis fatigue characterization sequence, the fatigue damage and remaining life of the vehicle device to be evaluated are assessed under physical constraints.

[0027] This application divides multi-source monitoring data into multiple monitoring windows and constructs a historical sequence according to a preset process order. This allows the evaluation process to effectively distinguish between incidental impact responses and continuous performance degradation by utilizing the state evolution trends reflected in the current window and previous windows, thus improving the accuracy of fatigue damage assessment. By converting multi-source monitoring data into a multi-axis fatigue characterization sequence independent of local coordinate selection, the projection influence of sensor installation posture differences and local coordinate differences of measurement points on the evaluation results is eliminated, enhancing the model's generalization ability under different vehicles, different measurement points, and cross-line application scenarios. By evaluating fatigue damage and remaining life based on degradation encoding vectors and multi-axis fatigue characterization sequences under physical constraints, the evaluation results satisfy the mechanical laws that damage increases monotonically with driving parameters and life decreases monotonically with driving parameters. This avoids the results that violate physical consistency produced by pure data-driven models when samples are scarce or operating conditions are extrapolated, enhancing the engineering credibility and interpretability of the evaluation results and providing a more comprehensive basis for judging the structural health status of vehicle components.

[0028] The following describes some embodiments of this application in detail. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0029] This application provides a method for evaluating vehicle components, which constructs a historical sequence from multi-source monitoring data collected during the operation of the vehicle component to be evaluated, according to a preset process order.

[0030] The vehicle components to be evaluated in this application refer to mechanical parts or structural units in rail vehicles that bear dynamic loads and require structural health monitoring and fatigue life assessment. Broadly speaking, these vehicle components can encompass the key load-bearing structures of the running gear in various types of rail vehicles, including high-speed trains, intercity EMUs, metro vehicles, suburban trains, and trams. Specifically, in a typical application scenario of this application, the vehicle components to be evaluated are preferably key load-bearing parts of high-speed train bogies, such as the transition area at the end of the bogie frame, the transition area of ​​the crossbeam, the weld hotspot area, and high stress concentration areas near the connecting seat. These parts are continuously subjected to the coupling effects of wheel-rail excitation, suspension-transmitted loads, track irregularities, and complex environmental loads during the long-term service of the vehicle, exhibiting typical high-cycle fatigue and multi-axis fatigue characteristics. It should be noted that the technical solution of this application is also applicable to the fatigue damage assessment and remaining life prediction of similar components in other large mechanical structural equipment with cyclic load service characteristics, and is not strictly limited to the specific parts listed above.

[0031] The preset process order described in this application refers to a logical rule used to arrange the order of data acquisition or data organization. Broadly speaking, this process order can reflect a temporal sequence, a mileage accumulation sequence, or a sequence of the degree of degradation within the vehicle's structural components. These sequences collectively reflect the evolution trajectory of the vehicle's components from the start of service to the present moment. Specifically, the process order can be sorted according to a fixed time interval, such as every 10 seconds, or according to a certain number of kilometers traveled, such as every 100 kilometers, or according to the evolution process from low to high or from healthy to degraded structural health status assessment levels. In a preferred embodiment, this application uses a temporal sequence as the default process order, that is, arranging the acquired data sequentially according to their absolute or relative time of occurrence.

[0032] Building upon this foundation, this application first collects multi-source monitoring data on the vehicle components under evaluation during operation using various sensors installed at key locations within the vehicle. "Multi-source" refers to data sources encompassing at least two or more categories of information with different physical meanings or measurement principles. Functionally, these data can be broadly categorized into two types: one characterizing the structural dynamic response of vehicle components, and the other describing the vehicle's current operating conditions and environmental conditions. Structural dynamic response data directly reflects the mechanical behavior of components under external excitation and is the core basis for fatigue damage assessment; operating condition and environmental data are used to correct for differences in damage evolution under different operating conditions, improving the scenario adaptability of the assessment results.

[0033] Specifically, structural dynamic response data can include the acceleration of the bogie frame, the acceleration of the axle boxes, the displacement or relative displacement between components, and strain or stress proxy signals at key hot spots. Bogie frame acceleration and axle box acceleration are typically acquired by piezoelectric or MEMS accelerometers to capture wheel-rail excitation and structural vibration response; displacement or relative displacement can be obtained by laser displacement sensors, eddy current sensors, or wire displacement gauges to reflect the deformation of the suspension system or connecting parts; strain or stress proxy signals at key hot spots are measured by strain gauges or fiber optic sensors attached to high stress concentration areas near the frame end transition zone, crossbeam transition zone, weld hot spots, or connecting seats, and are key inputs directly driving fatigue damage assessment. Operating condition and environmental data can include operating speed, track type, ambient temperature, and mileage information. Operating speed can be read from the train control system or speed sensor; line category reflects the smoothness and grade of the line, which can be obtained through onboard GPS combined with electronic map or ground beacon; ambient temperature is measured by temperature sensor, and temperature changes will affect the mechanical properties of materials and the output characteristics of sensor; mileage information is obtained through odometer or train network and is used to align data according to running distance.

[0034] In one specific implementation, for the hot spot area at the end of the bogie frame of a certain type of high-speed train, eight types of data are simultaneously collected: frame acceleration, axle box acceleration, suspension displacement, dynamic strain signals output by strain gauges, train speed, current track type, ambient temperature, and cumulative mileage. In another implementation, to reduce sensor deployment costs and data transmission burden, only a combination of two or three types of data can be collected. For example, only the strain signals and speed of key hot spots can be collected, or only axle box acceleration, displacement, and track type can be collected. Those skilled in the art can flexibly select at least two types of data to combine according to specific evaluation accuracy requirements and on-site installation conditions, without departing from the protection scope defined by this application. Regardless of the data types selected, the collected data must undergo preprocessing operations such as time synchronization, outlier removal, detrending, bandpass filtering, and standardization to ensure the effectiveness of subsequent feature extraction and modeling.

[0035] After collecting the above data, the continuously collected data stream is divided into multiple monitoring windows of fixed length according to the preset process sequence. Each monitoring window represents a continuous running process, and its window length can be flexibly set according to the actual sampling frequency and computing resources. For example, at a sampling frequency of 200Hz to 2000Hz, a window of 5 to 30 seconds can be used. A certain overlap rate, such as 25% to 75%, can be set between adjacent windows to ensure the continuity of monitoring.

[0036] Subsequently, the multiple monitoring windows are combined sequentially according to the process order to construct a historical sequence. This historical sequence is not a simple stacking of independent windows, but rather an ordered set that organizes the current monitoring window together with several of its predecessor monitoring windows.

[0037] For example, in one specific implementation, the k-th window is taken as the current window, and the preceding m-1 windows are combined with the k-th window to form an ordered historical sample X(k-m+1:k)={x(k-m+1),...,x(k)}, where m is usually an integer between 4 and 8. This constructed historical sequence can fully reflect the trend information of the gradual degradation of vehicle components from an early relatively healthy state to the current state. This allows subsequent evaluation models to utilize not only the response at the current moment but also the state evolution process at previous moments, effectively avoiding misjudging incidental impacts as continuous fatigue damage.

[0038] After obtaining the historical sequence reflecting the state evolution trend of vehicle components, this application further performs temporal degradation encoding on the historical sequence to extract a degradation encoding vector that can characterize the direction of performance degradation, the rate of change, and the state evolution trend. Temporal degradation encoding refers to using a computational model capable of processing sequential data to process each monitoring window in the historical sequence sequentially according to the process order, and accumulating and passing the information of previous windows to the current window through a state transfer mechanism, ultimately outputting a fixed-dimensional vector. This vector is no longer a direct mapping of the original physical signal, but a degradation feature representation that integrates the sequential information of multiple windows. Its core function is to condense the evolution process of "how the state degrades step by step from the early state to the current state" into a form that can be directly used by the subsequent damage solving module. For example, the temporal degradation encoder is implemented using at least one of a one-dimensional convolutional network, a gated recurrent unit, a long short-term memory network, or a time-series transformer network.

[0039] In a preferred implementation, this application employs a temporal neural network from deep learning as the degradation encoder. Taking the k-th window as the current window, the ordered historical samples X(k-m+1:k) formed by the preceding m-1 windows and these samples are input into the encoder. The encoder processes the multi-channel monitoring data within each window sequentially, and the feature vector of each window is non-linearly fused with the hidden state of the preceding window in the time dimension. After iterating through all windows, the encoder outputs a fixed-length degradation encoding vector z_k. This vector must reflect at least three types of information: first, the direction of performance degradation, i.e., the degradation mode to which the current state shifts compared to the initial healthy state; second, the rate of change, i.e., the magnitude of change in damage features between adjacent windows; and third, the state evolution trend, i.e., whether the current state is in the early, middle, or accelerated stage of degradation. To ensure the effectiveness of the encoding, the model parameters should be trained using a large amount of historical data in a supervised or semi-supervised manner, so that the encoding vectors corresponding to samples in similar degradation stages are close to each other in the feature space, while samples in different degradation stages are far apart.

[0040] Specifically, in one feasible approach, this application employs a two-layer Long Short-Term Memory (LSTM) network as the temporal degradation encoder. The historical sequence is set to include the current window and its three preceding windows (i.e., m=4). The monitoring data for each window includes eight channels such as acceleration, strain, and velocity, with 1024 sampling points. The first layer of the LSTM receives data from each window at each time step and outputs a sequence of hidden states. The second layer re-encodes the output of the first layer, taking the hidden state of the last time step as the temporal feature of that window. Then, the LSTM output features of the four windows are concatenated chronologically and passed through a fully connected layer containing 128 neurons, ultimately obtaining a 128-dimensional degradation encoding vector. In online applications, whenever a new monitoring window is added, the system slides and discards the oldest window and adds the new window, rerunning the encoder to update the degradation encoding vector in real time, providing a dynamically changing degradation context for subsequent physical constraint damage assessment. Through the above methods, this application successfully transforms the slow degradation information contained in the historical sequence into a numerical vector representation, effectively solving the technical problem that a single monitoring window cannot distinguish between occasional shocks and continuous degradation accumulation.

[0041] In this application, multi-source monitoring data is converted into a multi-axis fatigue characterization sequence independent of local coordinate selection. This refers to the mathematical transformation that recombines the original multi-channel stress or strain signals output by the sensors, which depend on the local coordinate direction, into a set of fatigue equivalent parameters that are independent of or weakly correlated with coordinate rotation. The core purpose is to eliminate the projection effects caused by differences in sensor installation posture, different definitions of local coordinates of measuring points, or changes in local force paths, so that the features learned by the subsequent model are closer to the real fatigue damage driving factors, rather than the differences in measurement posture.

[0042] The phrase "independence on local coordinate selection" does not require absolute invariance, but rather means that compared to directly using the original local components, the transformed parameters exhibit higher stability in different coordinate directions, thereby improving the model's generalization ability across measurement points, vehicles, and tracks. Typically, this can be achieved based on equivalent stress, principal stress, stress amplitude on the critical plane, or path parameters in Ilyushin space from multiaxial fatigue theory.

[0043] In an optional embodiment, this application calculates the equivalent stress as a coordinate-independent characterization based on the three normal stress components (σ_x, σ_y, σ_z) and three shear stress components (τ_xy, τ_yz, τ_zx) measured in a local coordinate system. The formula for calculating the equivalent stress is as follows: σ_eq(t)=sqrt(σ_x²+σ_y²+σ_z² σ_xσ_y σ_yσ_z σ_zσ_x+3τ_xy²+3τ_yz²+3τ_zx²) The equivalent stress σ_eq(t) is a scalar function that remains invariant under coordinate rotation transformations (i.e., it is independent of the choice of local coordinate system), thus directly eliminating projection changes caused by differences in sensor installation orientation. Alternatively, principal stress surrogates can be used, i.e., solving for the eigenvalues ​​of the stress tensor yields the three principal stresses σ1, σ2, and σ3, and equivalent parameters can be constructed based on multiaxial fatigue criteria. Furthermore, the critical plane method can be used to search for the plane that maximizes fatigue damage parameters, extracting the normal stress amplitude and shear stress amplitude on that plane as characterizations. Or, path parameters in Ilyushin space can be used to map stress or strain components to five-dimensional or six-dimensional space, calculating path length or curvature as invariants. All of the above methods fall within the scope of "coordinate-independent multiaxial fatigue characterization" covered by this application.

[0044] In one specific implementation, for the hot spot area at the end of the high-speed train bogie frame, the strain gauges installed on-site output normal strain in three directions. These strains are converted into σ_x, σ_y, and τ_xy in the local coordinate system using elastic constitutive relations (τ_yz and τ_zx can be ignored due to the approximate plane stress state of the gauge surface). Substituting these components into the equivalent stress formula yields the σ_eq(t) sequence, which is a characterization sequence independent of the gauge direction. In practice, the stress or strain channels in the original multi-source monitoring data are first aligned by time, and the equivalent stress is calculated for each sampling point to form an equivalent stress time series. Then, the equivalent stress amplitude Δσ_eq,k is extracted for each monitoring window (e.g., through rainflow counting or peak-valley value extraction). This Δσ_eq,k serves as one of the core inputs for subsequent hot spot-sensitive damage driving parameters. In this way, even if the strain gauge bonding angle deviates by a few degrees in different vehicles or installation batches, the calculated equivalent stress amplitude remains consistent, thereby improving the robustness of the model. Furthermore, this application does not exclude the simultaneous use of multiple invariants (such as equivalent stress amplitude and critical plane parameters) to form a multi-channel characterization sequence, in order to more comprehensively describe fatigue damage under complex load paths. The final output multi-axis fatigue characterization sequence maintains the same window division and time alignment relationship as the historical sequence, which facilitates joint input with the degenerate encoding vector into the physical constraint damage solution model.

[0045] After constructing the degradation encoding vector and the multi-axis fatigue characterization sequence, this application evaluates the fatigue damage and remaining life of the vehicle components under physical constraints based on these two pieces of information. Physical constraints refer to the fact that the evaluation process must satisfy the fundamental laws of fatigue mechanics. For example, damage should monotonically increase with increasing stress amplitude or non-proportional loading, remaining life should monotonically decrease with increasing cumulative damage, and the relationship between life and damage driving parameters should follow a power law or logarithmic power law. These constraints are not simply used as subsequent verification rules, but are embedded in the structure or loss function of the evaluation model to ensure that the model's output does not violate the above physical laws under any input conditions, thereby avoiding anti-physical predictions from purely data-driven methods when extrapolating from small samples or operating conditions.

[0046] This evaluation process consists of four sequential steps: constructing hotspot-sensitive damage-driving parameters, solving for damage increment and cumulative damage using a joint input physical constraint damage solution model, solving for the remaining lifetime interval based on the distribution of cumulative damage and damage increment, and outputting the evaluation results.

[0047] First, regarding the construction of the hotspot-sensitive damage driving parameter. Based on a multiaxial fatigue characterization sequence and a degradation coding vector, this application constructs a scalar parameter, η_k, that comprehensively reflects the structural differences in key components, the non-proportionality of load paths, and the degree of historical degradation. This parameter means that information related to stress amplitude and load path from the multiaxial fatigue characterization sequence, and information related to state evolution trends from the degradation coding vector, are fused into a single-valued driving quantity through a certain functional relationship for subsequent lifetime mapping.

[0048] In its implementation, the driving parameter can be expressed as a nonlinear function of the equivalent stress amplitude multiplied by a non-proportional load correction term and a degradation state correction term, and then multiplied by a working condition correction function. Here, the equivalent stress amplitude reflects the load intensity within the current window, the non-proportional load correction term reflects the degree to which the load path deviates from proportional loading, the degradation state correction term reflects the impact of historical cumulative degradation on the current damage rate, and the working condition correction function adjusts the damage driving parameter according to operating conditions such as speed, temperature, and line type.

[0049] In one specific implementation, for the hot spot area at the end of the high-speed train bogie frame, the hot spot sensitive damage driving parameters are constructed using the following formula: η_k= β_h·(Δσ_eq,k)^{m_f}·(1+c_pP_k)·(1+c_hH_k)·g(v_k,T_k,r_k).

[0050] Wherein, Δσ_eq,k is the equivalent stress amplitude of the k-th window, provided by the multi-axis fatigue characterization sequence; P_k is the load path non-proportional parameter, also derived from the multi-axis fatigue characterization sequence; H_k is the degradation state quantity extracted from the degradation encoding vector; v_k, T_k, and r_k are the operating speed, ambient temperature, and line type, respectively; β_h is the hotspot sensitivity coefficient, related to the specific key component type; m_f is the fatigue sensitivity index; c_p and c_h are empirical coefficients; and g(·) is the operating condition correction function. This formula maps information with different physical meanings to damage-driving parameters in a unified manner.

[0051] As a preferred implementation, the degenerate state variable H_k is updated recursively: H_k=α· _k+(1 α)·H_(k 1), 0 < α ≤ 1 in, _k is the normalized degradation index extracted from the k-th window, and α is the forgetting factor. This recursive structure ensures that the current state estimate reflects both the current response and retains information about previous slow degradation.

[0052] Secondly, the multi-axis fatigue characterization sequence, hotspot-sensitive damage driving parameters, and degradation encoding vector are jointly input into the physical constraint damage solving model to solve for the damage increment and cumulative damage within the current monitoring window. The physical constraint damage solving model is a mathematical model embedding fatigue mechanics laws. Its core function is to calculate the amount of fatigue damage generated within the current window based on the input driving parameters and load information, and then add it to the historical total damage. The engine is characterized by the following: damage must monotonically increase with increasing stress amplitude, non-proportional degree, or degradation state; lifetime must monotonically decrease with increasing damage driving parameters; and the lifetime and driving parameters must follow a power law or logarithmic power law relationship.

[0053] In its specific implementation, the engine first performs equivalent cycle counting on the multi-axis fatigue characterization sequence to obtain the equivalent cycle number n_eq,k of the current window. Then, it calculates the corresponding fatigue life N_f,k based on the hotspot-sensitive damage driving parameter η_k, with the relationship satisfying logN_f,k=ab·log(η_k+ε), where a and b are parameters to be calibrated, and ε is a small positive number to prevent logarithmic singularity. Finally, the damage increment ΔD_k of the current window is n_eq,k / N_f,k, and the cumulative damage D_k=min(1,D_{k-1}+ΔD_k). To ensure physical consistency, monotonicity constraints need to be applied during engine training or calibration. D / Δσ_eq>0, D / P>0, D / H>0, N_f / η<0.

[0054] In one specific implementation, for the end of a high-speed train bogie frame, take a=25, b=4, ε=1e-6; use the rainflow counting method to process the equivalent stress time history, and obtain the equivalent cycle number n_eq,k=10 times within the window; if η_k corresponds to the fatigue life N_f,k=1.2e6 times at this time, then: ΔD_k = 8.33e-6; The cumulative damage D_k is updated from 0.023 in the previous window to 0.02300833. In this way, the engine follows the fatigue accumulation criterion and monotonicity constraint at each step, avoiding the anti-physics results that a purely data-driven model might produce.

[0055] Next, based on the statistical distribution of accumulated damage in the current monitoring window and damage increments in subsequent windows, the remaining lifetime interval from the corresponding time of the current monitoring window is calculated. The remaining lifetime interval is not a single value, but an interval estimate consisting of a lower limit and an upper limit, used to reflect the lifetime range under uncertain conditions. The remaining lifetime depends on the currently accumulated damage amount D_k and the magnitude of the damage increment that may occur per unit window in the future. Since the future load is random, this application uses the distribution of damage increments under similar working conditions in historical data to infer the possible range in the future. In a specific implementation, firstly, the high quantile value ΔD_high (e.g., 90th quantile) and low quantile value ΔD_low (e.g., 10th quantile) of the damage increment per unit window are statistically analyzed from a period of history before the current window (or historical windows under similar working conditions); then, the lower limit and upper limit of the remaining lifetime interval are calculated using the following formulas: RUL_low=(1-D_k) / ΔD_high, RUL_high=(1-D_k) / ΔD_low. RUL_low corresponds to a worse case (larger damage increment), while RUL_high corresponds to a better case (smaller damage increment).

[0056] In one specific implementation, if the current cumulative damage D_k = 0.00833, and the statistically obtained ΔD_high = 1.2e-5 and ΔD_low = 8.0e-6, then the lower limit of the remaining lifetime is approximately (1-0.00833) / 1.2e-5 ≈ 82639 windows, and the upper limit is approximately (1-0.00833) / 8.0e-6 ≈ 123958 windows. If each window corresponds to 10 seconds of operation time, then the remaining lifetime range is approximately 9.6 days to 14.4 days. This range provides more comprehensive risk information for maintenance decisions.

[0057] Finally, the current damage increment, cumulative damage, and remaining lifespan are combined as the evaluation results for the vehicle component being evaluated. Of these three outputs, the damage increment reflects the rate of fatigue deterioration over a recent period, the cumulative damage reflects the total service consumption to date, and the remaining lifespan provides the possible safe operating range in the future. The combination of these three provides both real-time status information and trend predictions and uncertainty boundaries, enabling maintenance personnel to develop more scientific maintenance plans. In a specific application scenario, when the cumulative damage of a critical part of a bogie exceeds a preset threshold (e.g., 0.9) or the lower limit of the remaining lifespan is lower than the time corresponding to the next planned maintenance mileage, the system can trigger a maintenance warning, achieving a closed loop from condition assessment to predictive maintenance. Through these four steps, this application completes the joint evaluation of fatigue damage and remaining lifespan of vehicle components under physical constraints, effectively solving the problems of lack of credibility and engineering operability in existing single numerical predictions.

[0058] In an optional implementation, after obtaining the damage increment, cumulative damage, and remaining lifespan interval of the current monitoring window as evaluation results, this application further determines the confidence level of the evaluation results and generates a maintenance trigger signal based on the cumulative damage, remaining lifespan interval, and confidence level. The purpose of introducing a confidence level is that, since the evaluation model relies on limited historical data and statistical distributions, the evaluation results may have significant uncertainty in cases of insufficient coverage of certain operating conditions, large model prediction variance, or input data deviating from the training distribution. In such cases, directly issuing maintenance or speed-limiting commands based on the evaluation results may lead to misjudgments. By quantifying the confidence level of the evaluation results and binding the confidence level to the maintenance triggering logic, the decision-making can be made more robust and executable.

[0059] A credibility level is a quantitative indicator or discrete grade reflecting the reliability of the current evaluation result. Its determination can be based on a comprehensive analysis of multiple factors, such as feature distribution distance (the degree of deviation between the current input features and the training sample distribution), model variance (the degree of dispersion of multiple predicted outputs under the same input), operating condition coverage (the frequency or similarity of the current operating condition in historical training data), and physical constraint satisfaction (whether the evaluation result fully satisfies the preset monotonicity and power-law constraints). These factors can be mapped to a continuous credibility value or several levels (such as high, medium, and low) through weighted scoring, fuzzy logic, or classifiers.

[0060] In one implementation, four sub-indicators are first calculated: the characteristic distribution distance is estimated using Mahalanobis distance or kernel density; the model variance is obtained through Monte Carlo dropout or model ensemble; the operating condition coverage is based on the matching frequency after discretization of speed, line, temperature, etc.; and the physical constraint satisfaction is determined based on whether the monotonicity between damage increment and stress amplitude holds. Then, the sub-indicators are normalized and weighted to obtain a comprehensive confidence score, which is then divided into three levels—"high confidence," "medium confidence," and "low confidence"—based on a preset threshold. High confidence indicates that the assessment results are reliable and can be directly used for maintenance decisions; medium confidence indicates that the results have some reference value but should be reviewed; and low confidence indicates that the results have high uncertainty and should not be used to trigger strong intervention actions.

[0061] In one specific implementation, the reliability level for the end of the high-speed train bogie frame is determined using the following method. First, the Mahalanobis distance between the input features of the current window and the features of all windows in the training set is calculated. The smaller the distance, the better the distribution match. A score of 1 is set when the distance is less than a threshold d1, and 0 when the distance is greater than d2, with linear interpolation in between. Second, by retaining the dropout layer in the model and performing 30 random forward passes during forward inference, the standard deviation σ_D of the cumulative damage in the evaluation results is calculated. A score of 1 is set when σ_D is less than 0.01, and 0 when it is greater than 0.05. Third, the speed, line type, and ambient temperature of the current window are matched with combinations in the historical database. If there is an exact match, the coverage score is 1; if only partially matched, the score is reduced proportionally. Fourth, it is checked whether the damage increment and equivalent stress amplitude of the current window satisfy a monotonically positive correlation. If they do, the constraint satisfaction score is 1; otherwise, it is 0. Finally, the weighted average of the four scores (with weights of 0.3, 0.2, 0.3, and 0.2 respectively) is taken to obtain the comprehensive reliability score. A score ≥ 0.8 is rated as high confidence; a score between 0.5 and 0.8 is rated as medium confidence; and a score < 0.5 is rated as low confidence.

[0062] After determining the confidence level, this application generates maintenance trigger signals based on the cumulative damage, remaining lifespan, and confidence level of the current monitoring window. The types of maintenance trigger signals can include continued monitoring, encrypted monitoring, scheduling retesting, speed-limited operation, immediate repair, or generating a maintenance work order. The generation logic typically employs rule-based decision trees or threshold comparison methods. Triggering conditions should consider both damage severity and confidence level: when cumulative damage or remaining lifespan indicates high risk and high confidence, strong intervention actions (such as alarms or speed limits) can be directly triggered; when there is high risk but low confidence, retesting or data verification actions should be prioritized to avoid false alarms; when the risk is moderate but the lower limit of the remaining lifespan is close to the planned maintenance window, soft reminders such as adjusting maintenance priorities or shortening inspection intervals can be triggered.

[0063] In one specific implementation, this application sets the following rules: If the cumulative damage D_k ≥ 0.9 and the confidence level is high, a "Level 1 Alarm" is immediately generated and a speed-limited maintenance is recommended; if D_k ≥ 0.9 but the confidence level is medium or low, only a "Retest Recommendation" is generated and the monitoring frequency is increased (e.g., the window overlap rate is increased to 75%). If D_k is between 0.7 and 0.9 and the lower limit of the remaining lifespan interval is less than the time corresponding to the planned maintenance mileage, and the confidence level is high or medium, a "Maintenance Priority Increase" and a "Work Order Recommendation" are generated. If D_k < 0.7 but the lower limit of the remaining lifespan interval suddenly drops rapidly and the confidence level is high, an "Increased Monitoring" signal is generated. For cases with low confidence and D_k close to the threshold, the system prioritizes outputting "Data Verification" rather than any intervention action. Through the above methods, this application transforms the assessment results into engineering-executable maintenance decision instructions, effectively solving the problem that existing technologies only output a single lifespan value and are difficult to directly guide operation and maintenance actions, thus forming a complete closed loop from online monitoring to predictive maintenance.

[0064] Figure 2 This invention provides a schematic diagram of a vehicle device evaluation apparatus according to an embodiment of the present application. The apparatus includes: The data acquisition module is used to construct a historical sequence from the multi-source monitoring data collected during the operation of the vehicle components to be evaluated, according to a preset process sequence. The degradation coding module is used to perform temporal degradation coding on the historical sequence to obtain a degradation coding vector; The conversion module is used to convert the multi-source monitoring data into a multi-axis fatigue characterization sequence that is independent of the selection of local coordinates; The evaluation module is used to evaluate the fatigue damage and remaining life of the vehicle component to be evaluated under physical constraints, based on the degradation encoding vector and the multi-axis fatigue characterization sequence.

[0065] Monitoring data is collected through the following methods, including: According to the time sequence, mileage sequence, or state sequence of the vehicle components to be evaluated, at least two of the following types of data are collected: bogie frame acceleration, axle box acceleration, displacement or relative displacement, key hot spot strain or stress proxy signal, operating speed, track type, ambient temperature, and mileage information.

[0066] The process involves constructing a historical sequence from the multi-source monitoring data collected during the operation of the vehicle components to be evaluated, according to a preset process order. This includes: The current monitoring window and its preceding multiple monitoring windows are arranged sequentially according to the process order to form a historical sequence that reflects the degradation trend from the early state to the current state.

[0067] The step of performing temporal degradation encoding on the historical sequence to obtain a degradation encoding vector includes: The historical sequence is input into a temporal degradation encoder to extract degradation encoding vectors that characterize the direction of structural performance degradation, rate of change, and state evolution trend; the temporal degradation encoder is implemented using at least one of a one-dimensional convolutional network, a gated recurrent unit, a long short-term memory network, or a temporal transformer network. When using a Long Short-Term Memory (LSTM) network, a two-layer LSTM network is used. The historical sequence containing the current window and its previous windows is sequentially input into the first layer of the LSTM network according to the process order to obtain the hidden state sequence output by the first layer. The hidden state sequence is input into the second layer of the long short-term memory network to obtain the hidden state vector output by the second layer. The hidden state vector output by the second layer is used as the degenerate coding vector.

[0068] The step of converting the multi-source monitoring data into a multi-axis fatigue characterization sequence independent of local coordinate selection includes: The stress or strain components in the multi-source monitoring data are extracted and converted into fatigue equivalent parameters that are invariant to coordinate rotation, so as to eliminate the projection influence of sensor installation posture differences or local coordinate differences of measuring points on the evaluation results; the fatigue equivalent parameters include at least one of equivalent stress, principal stress surrogate quantity, key plane candidate parameter or Ilyusin path parameter; When the fatigue equivalent parameters include equivalent stress, the equivalent stress time series is calculated according to the equivalent stress formula with coordinate rotation invariance, and the equivalent stress amplitude of each window is extracted as a unit of monitoring window, as part of the multiaxial fatigue characterization sequence.

[0069] The evaluation of fatigue damage and remaining life of the vehicle component under physical constraints, based on the degraded encoding vector and the multi-axis fatigue characterization sequence, includes: Based on the equivalent stress amplitude and load path non-proportional parameters in the multi-axis fatigue characterization sequence, the degradation state quantities in the degradation encoding vector, and the operating speed, ambient temperature, and circuit type of the vehicle components to be evaluated, a hotspot-sensitive damage driving parameter is constructed by fusing it into a single-valued scalar according to the power-law relationship and nonlinear correction. This parameter reflects the structural differences of key parts, the degree of load non-proportionality, and the influence of historical degradation and operating conditions on damage evolution. The multi-axis fatigue characterization sequence, the hotspot-sensitive damage driving parameters, and the degradation encoding vector are jointly input into the physical constraint damage solution model. The equivalent cycle number is obtained by performing equivalent cycle counting on the multi-axis fatigue characterization sequence. The fatigue life is calculated according to the logarithmic power law relationship based on the hotspot-sensitive damage driving parameters. The damage increment and cumulative damage of the current monitoring window are solved using the equivalent cycle number and the fatigue life. Based on the statistical distribution of the cumulative damage in the current monitoring window and the damage increment in subsequent windows, the remaining lifetime interval from the time corresponding to the current monitoring window is calculated. The current monitoring window damage increment, cumulative damage, and remaining lifetime interval are used as the evaluation results for the vehicle device to be evaluated.

[0070] It also includes a maintenance module for determining the confidence level of the evaluation results; A maintenance trigger signal is generated based on the cumulative damage in the current monitoring window, the remaining lifespan, and the confidence level.

[0071] like Figure 3 As shown, this application provides an electronic device for performing the vehicle device evaluation method of this application. The device includes a memory, a processor, a bus, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the vehicle device evaluation method described above.

[0072] Specifically, the aforementioned memory and processor can be general-purpose memory and processor, without any specific limitations. When the processor runs the computer program stored in the memory, it can execute the aforementioned method for evaluating vehicle components.

[0073] Corresponding to the vehicle component evaluation method in this application, this application embodiment also provides a computer storage medium storing a computer program, which is executed by a processor to perform the steps of the above-described vehicle component evaluation method.

[0074] Specifically, the storage medium can be a general-purpose storage medium, such as a removable disk or hard disk, and when the computer program on the storage medium is run, it can execute the above-mentioned vehicle device evaluation method.

[0075] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0076] 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; 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.

[0077] 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.

[0078] 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.

[0079] 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.

[0080] 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 evaluating vehicle components, characterized in that, The method includes: According to the preset process sequence, the multi-source monitoring data collected during the operation of the vehicle components to be evaluated are constructed into a historical sequence; The historical sequence is subjected to temporal degradation encoding to obtain a degradation encoding vector; The multi-source monitoring data is converted into a multi-axis fatigue characterization sequence that is independent of the selection of local coordinates. Based on the degradation encoding vector and the multi-axis fatigue characterization sequence, the fatigue damage and remaining life of the vehicle component to be evaluated are assessed under physical constraints.

2. The method according to claim 1, characterized in that, Monitoring data is collected through the following methods, including: According to the time sequence, mileage sequence, or state sequence of the vehicle components to be evaluated, at least two of the following types of data are collected: bogie frame acceleration, axle box acceleration, displacement or relative displacement, key hot spot strain or stress proxy signal, operating speed, track type, ambient temperature, and mileage information.

3. The method according to claim 1, characterized in that, The process involves constructing a historical sequence from the multi-source monitoring data collected during the operation of the vehicle components to be evaluated, according to a preset process order. This includes: According to the preset process sequence, the multi-source monitoring data collected during the operation of the vehicle components to be evaluated are divided into multiple monitoring windows; The current monitoring window and its preceding multiple monitoring windows are arranged sequentially according to the process order to form a historical sequence that reflects the degradation trend from the early state to the current state.

4. The method according to claim 3, characterized in that, The step of performing temporal degradation encoding on the historical sequence to obtain a degradation encoding vector includes: The historical sequence is input into a temporal degradation encoder to extract degradation encoding vectors that characterize the direction of structural performance degradation, rate of change, and state evolution trend; the temporal degradation encoder is implemented using at least one of a one-dimensional convolutional network, a gated recurrent unit, a long short-term memory network, or a temporal transformer network. When using a Long Short-Term Memory (LSTM) network, a two-layer LSTM network is used. The historical sequence containing the current window and its previous windows is sequentially input into the first layer of the LSTM network according to the process order to obtain the hidden state sequence output by the first layer. The hidden state sequence is input into the second layer of the long short-term memory network to obtain the hidden state vector output by the second layer. The hidden state vector output by the second layer is used as the degenerate coding vector.

5. The method according to claim 1, characterized in that, The step of converting the multi-source monitoring data into a multi-axis fatigue characterization sequence independent of local coordinate selection includes: The stress or strain components in the multi-source monitoring data are extracted and converted into fatigue equivalent parameters that are invariant to coordinate rotation, so as to eliminate the projection influence of sensor installation posture differences or local coordinate differences of measuring points on the evaluation results; the fatigue equivalent parameters include at least one of equivalent stress, principal stress surrogate quantity, key plane candidate parameter or Ilyusin path parameter; When the fatigue equivalent parameters include equivalent stress, the equivalent stress time series is calculated according to the equivalent stress formula with coordinate rotation invariance, and the equivalent stress amplitude of each window is extracted as a unit of monitoring window, as part of the multiaxial fatigue characterization sequence.

6. The method according to claim 1, characterized in that, The evaluation of fatigue damage and remaining life of the vehicle component under physical constraints, based on the degraded encoding vector and the multi-axis fatigue characterization sequence, includes: Based on the equivalent stress amplitude and load path nonproportional parameters in the multiaxial fatigue characterization sequence, the degradation state quantity in the degradation encoding vector, and the operating speed, ambient temperature, and line type of the vehicle components to be evaluated, a hotspot-sensitive damage driving parameter is constructed by fusing it into a single-valued scalar according to the power law relationship and nonlinear correction. This parameter reflects the structural differences of key parts, the degree of load nonproportion, and the impact of historical degradation and operating conditions on damage evolution. The multi-axis fatigue characterization sequence, the hotspot-sensitive damage driving parameters, and the degradation encoding vector are jointly input into the physical constraint damage solution model. The equivalent cycle number is obtained by performing equivalent cycle counting on the multi-axis fatigue characterization sequence. The fatigue life is calculated according to the logarithmic power law relationship based on the hotspot-sensitive damage driving parameters. The damage increment and cumulative damage of the current monitoring window are solved using the equivalent cycle number and the fatigue life. Based on the statistical distribution of the cumulative damage in the current monitoring window and the damage increment in subsequent windows, the remaining lifetime interval from the time corresponding to the current monitoring window is calculated. The current monitoring window damage increment, cumulative damage, and remaining lifetime interval are used as the evaluation results for the vehicle device to be evaluated.

7. The method according to claim 6, characterized in that, The method further includes: Determine the confidence level of the evaluation results; A maintenance trigger signal is generated based on the cumulative damage in the current monitoring window, the remaining lifespan, and the confidence level.

8. An apparatus for evaluating vehicle components, characterized in that, The device includes: The data acquisition module is used to construct a historical sequence from the multi-source monitoring data collected during the operation of the vehicle components to be evaluated, according to a preset process order. The degradation coding module is used to perform temporal degradation coding on the historical sequence to obtain a degradation coding vector; The conversion module is used to convert the multi-source monitoring data into a multi-axis fatigue characterization sequence that is independent of the selection of local coordinates; The evaluation module is used to evaluate the fatigue damage and remaining life of the vehicle component to be evaluated under physical constraints, based on the degradation encoding vector and the multi-axis fatigue characterization sequence.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine instructions that the processor executes. When the electronic device is running, the processor communicates with the memory via the bus. When the machine instructions are executed by the processor, they perform the steps of the method for evaluating vehicle devices as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium stores a computer program that, when executed by a processor, performs the steps of the method for evaluating vehicle components as described in any one of claims 1 to 7.

Images