Method for evaluating fatigue life of wedge brake based on machine learning

Abstract

The application discloses a wedge brake fatigue life evaluation method based on machine learning and belongs to the technical field of brake life evaluation. The method comprises the following steps: collecting wedge structure dynamic data and stress field data of a wedge brake; obtaining original simulation data sets through preprocessing; extracting brake dynamic characteristics; obtaining a sensitive node set at the root of a brake wedge groove; obtaining a physically constrained graph neural network with completed training according to the sensitive node set at the root of the brake wedge groove and the brake dynamic characteristics; obtaining real-time working condition parameters, predicting a stress change curve according to the physically constrained graph neural network, statistically counting a cumulative damage value and self-locking correcting the cumulative damage value; and analyzing a fatigue life evaluation value of the wedge brake. Through embedding physical constraints into a comprehensive loss function of the physically constrained graph neural network, the application realizes deep coupling between a machine learning model and physical laws, and solves the problem that a current pure data-driven model deviates from actual physical behaviors under extreme working conditions.

Description

Technical Field

[0001] This invention relates to the field of brake life assessment technology, and in particular to a machine learning-based method for assessing the fatigue life of wedge brakes. Background Technology

[0002] As a core component of mechanical braking systems, the fatigue life assessment of wedge brakes directly affects the operational safety and reliability of vehicles, construction machinery, and other equipment. With the increasing speed and heavy-duty nature of industrial equipment, the nonlinear mechanical behavior of brakes under complex operating conditions has a significant impact on fatigue failure.

[0003] Current fatigue assessment technologies for wedge brakes are mainly divided into two categories: pure physics engine simulation and pure data-driven models. For example, the dynamic equations are solved using software such as ANSYS and ABAQUS, while simulation data is directly fitted using networks such as LSTM and CNN.

[0004] In existing technologies, when the brake is under extreme conditions such as self-locking or impact, the model cannot capture the stress surge caused by the sudden change in contact force, resulting in distortion of fatigue damage calculation. There is a problem that the pure data-driven model's prediction deviates from the actual physical behavior under extreme conditions. Summary of the Invention

[0005] This application provides a machine learning-based method for assessing the fatigue life of wedge brakes, which solves the problem in the prior art where purely data-driven models predict deviations from actual physical behavior under extreme conditions, and achieves deep coupling between machine learning models and physical laws.

[0006] This application provides a machine learning-based method for evaluating the fatigue life of a wedge brake, including: Step 1, collecting dynamic data and stress field data of the wedge structure during multibody dynamics simulation based on the CAD model and working parameters of the wedge brake;

[0007] Step 2: Preprocess the dynamic data of the wedge structure to obtain the original simulation dataset;

[0008] Step 3: Extract the dynamic features of the brakes from the original simulation dataset;

[0009] Step 4: Obtain the stress gradient field of each node at the root of the brake wedge groove from the stress field data, and extract the sensitive nodes based on the stress gradient field of each node to obtain the set of sensitive nodes at the root of the brake wedge groove.

[0010] Step 5: Based on the set of sensitive nodes at the root of the brake wedge groove and the dynamic characteristics of the brake, obtain the trained physical constraint graph neural network.

[0011] Step 6: Obtain real-time operating parameters and input them directly into the physical constraint graph neural network. Based on the physical constraint graph neural network, predict the stress change curve of each sensitive node in the sensitive node set, and calculate the cumulative damage value accordingly and perform self-locking correction on the cumulative damage value.

[0012] Step 7: Analyze the fatigue life assessment value of the wedge brake based on the corrected cumulative damage value.

[0013] Furthermore, the original simulation dataset includes: self-locking state labels and instantaneous force increase ratio curves;

[0014] The self-locking status label is calculated based on the real-time normal and tangential forces at the contact surfaces of the brake wedge-roller assembly.

[0015] The instantaneous force ratio curve is obtained by analyzing the input force of the piston and the output force of the brake pads.

[0016] Furthermore, the calculation formula for the self-locking state tag is as follows:

[0017]

[0018] Where LockState(t) is the self-locking state label, t is the time index, and F T (t) represents the tangential force, F N (t) represents the normal force, μ represents the friction coefficient, LockState(t) = 1 represents the self-locking state, and LockState(t) = 0 represents the sliding state.

[0019] Furthermore, the formula for calculating the instantaneous force ratio is as follows:

[0020]

[0021] Among them, K boost (t) represents the instantaneous force increase ratio, F out (t) represents the output force of the brake pads, F in (t) represents the input force of the plunger.

[0022] Furthermore, the dynamic characteristics of the brake include: self-locking correction characteristics and force-enhancing coupling characteristics;

[0023] The self-locking correction feature is: Force coupling characteristics are

[0024] Among them, a s (t) is the sliding acceleration of the brake wedge, T r (t) represents the torque of the roller assembly, LockState(t) is the self-locking state label, and K boost (t) represents the instantaneous increase in force ratio.

[0025] Furthermore, the step of obtaining the set of sensitive nodes at the root of the brake wedge groove includes:

[0026]

[0027] Wherein, CriticalNodes is the set of sensitive nodes at the root of the brake wedge groove, σ threshold The stress gradient threshold, This represents the global stress gradient of the sensitive node.

[0028] Furthermore, the training process of the physical constraint graph neural network includes:

[0029] The dynamic data and stress field data of the wedge structure were synchronized and normalized in time, and divided into training and testing sets.

[0030] Define the graph nodes and edges in the graph structure of the physical constraint graph neural network based on the set of sensitive nodes at the root of the brake wedge groove;

[0031] The self-locking correction feature is input into the motion channel to learn the stress change of the braking wedge, and the force-enhancing coupling feature is input into the force-enhancing channel to learn the stress change of the roller assembly. The outputs of the two channels are fused by a graph convolutional layer to generate a comprehensive feature vector.

[0032] Obtain the Newtonian mechanical residual constraints and energy conservation constraints of the roller assembly, and construct a comprehensive loss function;

[0033] Training is performed on the training set until the comprehensive loss function of the physical constraint graph neural network reaches the convergence value, at which point training is terminated, and the training results are validated based on the test set.

[0034] After successful verification, the trained physical constraint graph neural network is output.

[0035] Furthermore, the comprehensive loss function of the physical constraint graph neural network is:

[0036] L total =L pred +λ·L physics +β·L energy +γ·L smooth

[0037] Among them, L total For the comprehensive loss value, L pred For the prediction error loss, λ, β, and γ are the weighting coefficients, respectively, and L physics L is the residual constraint term in Newtonian mechanics. energy L is the energy conservation constraint term. smooth This represents the smoothness loss in stress prediction;

[0038] The prediction error loss is Lpred =‖σ pred (t)-σ true (t)‖ 2 , σ pred (t) represents the predicted stress value, σ true (t) represents the actual stress value, and t is the time index;

[0039] The Newtonian mechanics residual constraint term is: J is the moment of inertia of the roller, and α is the angular acceleration. This is a force-coupled feature;

[0040] The energy conservation constraint term is: V represents the volume domain of the wedge brake, and ρ represents the material density. Let F be the displacement velocity, σ be the stress tensor, ε be the strain tensor, and σ:ε be the dot product of the stress and strain tensors. Calculate the elastic strain energy density, F. in (t) represents the input force of the plunger. Let x be the displacement velocity of the brake wedge, x be the displacement of the brake wedge, and the upper point represent the time derivative.

[0041] The smoothness loss in the stress prediction: σ pred (t) represents the predicted stress value.

[0042] Furthermore, the step of statistically calculating the cumulative damage value and performing self-locking correction on the cumulative damage value includes:

[0043] By extracting cyclic features from the stress variation curve of each sensitive node, the stress cyclic feature (Δσ) of each node is obtained. i ,n i ), where Δσ i Let n be the i-th type of stress amplitude of the material. i The number of cycles for the material under the i-th type of stress amplitude;

[0044] Query the maximum fatigue life corresponding to various stress amplitudes for each node;

[0045]

[0046] Where, N i Let be the maximum fatigue life of the material under the i-th type of stress amplitude, k be the stress sensitivity coefficient of the material, and C be the intrinsic strength of the material.

[0047] Based on the number of cycles and maximum fatigue life of each sensitive node material under the i-th type of stress amplitude, the cumulative damage value of each sensitive node material under the i-th type of stress amplitude is obtained;

[0048] When the cumulative damage value of a sensitive node is not less than 1, the sensitive node is determined to have experienced fatigue failure; otherwise, the cumulative damage value is determined to indicate that the sensitive node has exhausted its lifespan.

[0049] When the self-locking status label is detected to be in the self-locking state, the cumulative damage value of the material corresponding to the sensitive node at the root of the brake wedge groove under the first type of stress amplitude is obtained, and the fretting wear amplification correction is performed to obtain the corrected cumulative damage value.

[0050]

[0051] Among them, D' i D is the corrected cumulative damage value. i The cumulative damage value before correction is γ, which is the damage amplification factor.

[0052] Furthermore, the steps for obtaining the fatigue life assessment value include:

[0053] The number of cycles required for the cumulative damage value to be 1 at each sensitive node under the i-th stress amplitude is counted, and denoted as the fatigue life N of each sensitive node. f ,

[0054] The location with the largest cumulative damage value among all sensitive nodes is taken as the fatigue life assessment value of the wedge brake.

[0055] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0056] 1. By embedding the Newtonian mechanics residual constraints and energy conservation constraints into the comprehensive loss function of the physical constraint graph neural network, the model output is forced to conform to the dynamic behavior of the brake and the principle of energy balance. This achieves deep coupling between the machine learning model and physical laws, effectively solving the problem that the prediction of pure data-driven models deviates from the actual physical behavior under extreme conditions in the existing technology.

[0057] 2. By constructing a dual-channel input mechanism that combines self-locking correction features with force-enhancing coupling features, and by embedding the self-locking state label of the brake wedge-roller assembly contact edge, the dynamic feature fusion of the two physical dimensions of brake motion and force transmission is realized. This allows for the accurate capture of the accelerating effect of the surge in contact force under self-locking conditions on fatigue damage, thus solving the problem of decoupling nonlinear mechanical behavior from fatigue damage.

[0058] 3. By deeply integrating physical and mechanical constraints with graph neural networks, the stress curve is efficiently predicted by using physical priors to guide data learning. This maintains the physical interpretability of nonlinear behaviors such as brake self-locking while significantly improving computational efficiency, solving the problems of long processing time and insufficient coverage of extreme working conditions in traditional methods.

[0059] 4. By directly obtaining real-time operating parameters, i.e., real-time operating parameters can be directly input into the physical constraint graph neural network, there is no need to repeatedly call the physics engine for operating condition simulation. Through historical data matching and real-time data fusion, the work that relies on repeated simulation in traditional methods is transformed into fast parameter mapping of the model, which significantly reduces the consumption of computing resources and realizes real-time assessment and dynamic early warning of fatigue life. Attached Figure Description

[0060] Figure 1 A flowchart of a machine learning-based fatigue life assessment method for wedge brakes provided in this application embodiment. Detailed Implementation

[0061] This application provides a machine learning-based method for assessing the fatigue life of wedge brakes, which solves the problem in the prior art where pure data-driven models predict deviations from actual physical behavior under extreme conditions. By embedding Newtonian mechanical residual constraints and energy conservation constraints into the comprehensive loss function of a physical constraint graph neural network, the model output is forced to conform to the dynamic behavior and energy balance principle of the brake, thereby achieving deep coupling between the machine learning model and physical laws.

[0062] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0063] like Figure 1 As shown, this application provides a machine learning-based method for assessing the fatigue life of a wedge brake. The method includes the following steps: Step 1, collecting dynamic data and stress field data of the wedge structure during multibody dynamics simulation based on the CAD model and working parameters of the wedge brake.

[0064] The CAD model includes at least: a brake wedge, a roller assembly, and an adjusting plunger, wherein the roller assembly includes rollers and roller isolation frames;

[0065] The operating parameters include at least: the plunger input force range F. in Initial braking velocity v0, friction coefficient μ;

[0066] The dynamic data of the wedge structure includes: the normal force F at the contact surface of the brake wedge-roller assembly. N (t) and tangential force F T (t), also includes the brake pad output force F out (t), brake wedge displacement x(t), and roller assembly rotational angular velocity ω(t).

[0067] Step 2: Preprocess the dynamic data of the wedge structure to obtain the original simulation dataset;

[0068] The original simulation dataset includes: self-locking state labels and instantaneous force increase ratio curves;

[0069] The self-locking status label is based on the real-time normal force F of the brake wedge-roller assembly contact surface. N (t) and tangential force F T (t) is calculated;

[0070] The formula for calculating the self-locking status tag is:

[0071]

[0072] Where LockState(t) is the self-locking state label, F T (t) represents the tangential force, F N (t) represents the normal force, μ represents the coefficient of friction, LockState(t) = 1 represents the self-locking state, and LockState(t) = 0 represents the sliding state;

[0073] The instantaneous force increase ratio curve is obtained by inputting force F through the plunger. in (t) and brake pad output force F out (t) analysis yielded the following results;

[0074] The formula for calculating the instantaneous force increase ratio is:

[0075]

[0076] Among them, K boost (t) represents the instantaneous force increase ratio, F out (t) represents the output force of the brake pads, F in (t) represents the input force of the plunger.

[0077] Step 3: Extract the dynamic features of the brakes from the original simulation dataset;

[0078] The dynamic characteristics of the brake include: self-locking correction characteristics and force-enhancing coupling characteristics;

[0079] The self-locking correction feature is: Force coupling characteristics are

[0080] Among them, a s (t) is the sliding acceleration of the brake wedge, T r (t) represents the torque of the roller assembly, LockState(t) is the self-locking state label, and K boost (t) represents the instantaneous force increase ratio;

[0081] Step 4: Obtain the stress gradient field of each node at the root of the brake wedge groove from the stress field data, and extract the sensitive nodes based on the stress gradient field of each node to obtain the set of sensitive nodes at the root of the brake wedge groove.

[0082]

[0083] Wherein, CriticalNodes is the set of sensitive nodes at the root of the brake wedge groove, σ threshold The stress gradient threshold, The global stress gradient of the sensitive node;

[0084] The stress field data includes at least the stress distribution data of the brake wedge, brake assembly, and brake backing plate.

[0085] Step 5: Define the graph nodes and edges in the graph structure of the physical constraint graph neural network based on the set of sensitive nodes at the root of the brake wedge groove. Input the self-locking correction feature into the motion channel to learn the stress change of the brake wedge. Input the force-enhancing coupling feature into the force-enhancing channel to learn the stress change of the roller assembly.

[0086] The nodes include: brake wedge grid points, roller assembly center points, and adjusting plunger contact points;

[0087] The edge includes: the contact edge of the brake wedge-roller assembly;

[0088] Embed the self-locking status label into the properties of the brake wedge-roller assembly contact edge.

[0089] The training process of the physical constraint graph neural network includes:

[0090] The dynamic data and stress field data of the wedge structure were synchronized and normalized in time, and divided into training and testing sets.

[0091] Define the graph nodes and edges in the graph structure of the physical constraint graph neural network based on the set of sensitive nodes at the root of the brake wedge groove;

[0092] The self-locking correction feature is input into the motion channel to learn the stress change of the braking wedge, and the force-enhancing coupling feature is input into the force-enhancing channel to learn the stress change of the roller assembly. The outputs of the two channels are fused by a graph convolutional layer to generate a comprehensive feature vector.

[0093] Obtain the Newtonian mechanical residual constraints and energy conservation constraints of the roller assembly, and construct a comprehensive loss function;

[0094] Training is performed on the training set until the comprehensive loss function of the physical constraint graph neural network reaches the convergence value, at which point training is terminated, and the training results are validated based on the test set.

[0095] After successful verification, the trained physical constraint graph neural network is output.

[0096] The comprehensive loss function of the physical constraint graph neural network is:

[0097] Ltotal =L pred +λ·L physics +β·L energy +γ·L smooth

[0098] Among them, L total For the comprehensive loss value, L pred For the prediction error loss, λ, β, and γ are the weighting coefficients, respectively, and L physics L is the residual constraint term in Newtonian mechanics. energy L is the energy conservation constraint term. smooth This represents the smoothness loss in stress prediction;

[0099] The prediction error loss is L pred =‖σ pred (t)-σ true (t)‖ 2 , σ pred (t) represents the predicted stress value, σ true (t) represents the actual stress value;

[0100] The Newtonian mechanics residual constraint term is: J is the moment of inertia of the roller, and α is the angular acceleration;

[0101] The energy conservation constraint term is: V represents the volume domain of the wedge brake, and ρ represents the material density. Let σ be the displacement velocity, σ be the stress tensor, and ε be the strain tensor. σ:ε is the dot product of the stress and strain tensors. Calculate the elastic strain energy density. Let x be the displacement velocity of the brake wedge, x be the displacement of the brake wedge, and the upper point represent the time derivative.

[0102] The smoothness loss in the stress prediction: σ pred (t) represents the predicted stress value.

[0103] Step 6: Obtain real-time operating parameters and input them directly into the physical constraint graph neural network. Based on the physical constraint graph neural network, predict the stress change curve of each sensitive node in the sensitive node set. Extract cyclic features from the stress change curve of each sensitive node to obtain the stress cyclic feature (Δσ) of each node. i ,n i ), where Δσ i Let n be the i-th type of stress amplitude of the material. i The number of cycles for the material under the i-th type of stress amplitude;

[0104] Query the maximum fatigue life corresponding to various stress amplitudes for each node;

[0105]

[0106] Where, N i Let be the maximum fatigue life of the material under the i-th type of stress amplitude, k be the stress sensitivity coefficient of the material, and C be the intrinsic strength of the material.

[0107] Based on the number of cycles and maximum fatigue life of each sensitive node material under the i-th type of stress amplitude, the cumulative damage value of each sensitive node material under the i-th type of stress amplitude is obtained. When the cumulative damage value of the sensitive node is not less than 1, it is determined that the sensitive node has failed due to fatigue; otherwise, it is determined that the cumulative damage value is the lifespan of the sensitive node.

[0108] When the self-locking status label is detected to be in the self-locking state, the cumulative damage value of the material corresponding to the sensitive node at the root of the brake wedge groove under the first type of stress amplitude is obtained, and the fretting wear amplification correction is performed to obtain the corrected cumulative damage value.

[0109]

[0110] Among them, D i D' is the cumulative damage value before correction. i γ is the corrected cumulative damage value, and γ is the damage amplification factor.

[0111] The damage amplification factor is the ratio of the average fatigue life under non-locking conditions to the average fatigue life under locking conditions. Since locking conditions accelerate damage, the average fatigue life under locking conditions does not exceed the average fatigue life under non-locking conditions.

[0112] Step 7: Analyze the fatigue life assessment value of the wedge brake based on the corrected cumulative damage value:

[0113] The number of cycles required for the cumulative damage value to be 1 at each sensitive node under the i-th stress amplitude is counted, and this number is recorded as the fatigue life of each sensitive node.

[0114] The location with the largest cumulative damage value among all sensitive nodes is taken as the critical area fatigue life assessment value of the wedge brake.

[0115] In summary, the embodiments of this application embed Newtonian mechanics residual constraints and energy conservation constraints into the comprehensive loss function of the physical constraint graph neural network, thereby forcing the model output to conform to the dynamic behavior of the brake and the principle of energy balance, and thus realizing the deep coupling between the machine learning model and physical laws.

[0116] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0117] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0118] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0119] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0120] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0121] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A machine learning-based method for assessing the fatigue life of a wedge brake, characterized in that, Includes the following steps: Step 1: Collect dynamic data and stress field data of the wedge structure during multibody dynamics simulation based on the CAD model and working parameters of the wedge brake; Step 2: Preprocess the dynamic data of the wedge structure to obtain the original simulation dataset; Step 3: Extract the dynamic features of the brakes from the original simulation dataset; Step 4: Obtain the stress gradient field of each node at the root of the brake wedge groove from the stress field data, and extract the sensitive nodes based on the stress gradient field of each node to obtain the set of sensitive nodes at the root of the brake wedge groove. Step 5: Based on the set of sensitive nodes at the root of the brake wedge groove and the dynamic characteristics of the brake, obtain the trained physical constraint graph neural network. Step 6: Obtain real-time operating parameters and input them directly into the physical constraint graph neural network. Based on the physical constraint graph neural network, predict the stress change curve of each sensitive node in the sensitive node set, and calculate the cumulative damage value accordingly and perform self-locking correction on the cumulative damage value. Step 7: Analyze the fatigue life assessment value of the wedge brake based on the corrected cumulative damage value; The steps for obtaining the set of sensitive nodes at the root of the brake wedge groove include: ； in, This is the set of sensitive nodes at the root of the brake wedge groove. The stress gradient threshold, The global stress gradient of the sensitive node; The steps of calculating the cumulative damage value and performing self-locking correction on the cumulative damage value include: By extracting cyclic features from the stress change curves of each sensitive node, the stress cyclic features of each node are obtained. ,in, Let i be the amplitude of the i-th type of stress in the material. The number of cycles for the material under the i-th type of stress amplitude; Query the maximum fatigue life corresponding to various stress amplitudes for each node; ； in, Let be the maximum fatigue life of the material under the i-th type of stress amplitude, k be the stress sensitivity coefficient of the material, and C be the intrinsic strength of the material. Based on the number of cycles and maximum fatigue life of each sensitive node material under the i-th type of stress amplitude, the cumulative damage value of each sensitive node material under the i-th type of stress amplitude is obtained; When the cumulative damage value of a sensitive node is not less than 1, the sensitive node is determined to have experienced fatigue failure; otherwise, the cumulative damage value is determined to indicate that the sensitive node has exhausted its lifespan. When the self-locking status label is detected to be in the self-locking state, the cumulative damage value of the material corresponding to the sensitive node at the root of the brake wedge groove under the first type of stress amplitude is obtained, and the fretting wear amplification correction is performed to obtain the corrected cumulative damage value. ； in, This is the corrected cumulative damage value. The cumulative damage value before correction. This is the damage amplification factor.

2. The machine learning-based fatigue life assessment method for wedge brakes as described in claim 1, characterized in that, The original simulation dataset includes: self-locking state labels and instantaneous force increase ratio curves; The self-locking status label is calculated based on the real-time normal and tangential forces at the contact surfaces of the brake wedge-roller assembly. The instantaneous force ratio curve is obtained by analyzing the input force of the piston and the output force of the brake pads.

3. The machine learning-based fatigue life assessment method for wedge brakes as described in claim 2, characterized in that, The calculation formula for the self-locking status tag is: ； in, Here, t represents the self-locking state label, and t represents the time index. For tangential force, For normal force, The coefficient of friction, It is in a self-locking state. It is in a sliding state.

4. The machine learning-based fatigue life assessment method for wedge brakes as described in claim 2, characterized in that, The formula for calculating the instantaneous force increase ratio is: ； in, For instantaneous force ratio, For the output force of the brake pads, Force is input to the plunger.

5. The machine learning-based fatigue life assessment method for wedge brakes as described in claim 2, characterized in that, The dynamic characteristics of the brake include: self-locking correction characteristics and force-enhancing coupling characteristics; The self-locking correction feature is: The force coupling characteristic is ; in, For the braking wedge sliding acceleration, For the torque of the roller assembly, This is a self-locking status label. This is the instantaneous force ratio.

6. The machine learning-based fatigue life assessment method for wedge brakes as described in claim 1, characterized in that, The training process of the physical constraint graph neural network includes: The dynamic data and stress field data of the wedge structure were synchronized and normalized in time, and divided into training set and test set; Define the graph nodes and edges in the graph structure of the physical constraint graph neural network based on the set of sensitive nodes at the root of the brake wedge groove; The self-locking correction feature is input into the motion channel to learn the stress change of the braking wedge, and the force-enhancing coupling feature is input into the force-enhancing channel to learn the stress change of the roller assembly. The outputs of the two channels are fused by a graph convolutional layer to generate a comprehensive feature vector. Obtain the Newtonian mechanical residual constraints and energy conservation constraints of the roller assembly, and construct a comprehensive loss function; Training is performed on the training set until the comprehensive loss function of the physical constraint graph neural network reaches the convergence value, at which point training is terminated, and the training results are validated based on the test set. After successful verification, the trained physical constraint graph neural network is output.

7. The machine learning-based fatigue life assessment method for wedge brakes as described in claim 6, characterized in that, The comprehensive loss function of the physical constraint graph neural network is: ； in, The total loss value, To predict error loss, , , These are the weighting coefficients, This is a residual constraint term in Newtonian mechanics. For energy conservation constraints, This represents the smoothness loss in stress prediction; The prediction error loss is , To predict stress values, The value represents the actual stress, and t is the time index. The Newtonian mechanics residual constraint term is: J is the moment of inertia of the roller. Angular acceleration, This is a force-coupled feature; The energy conservation constraint term is: V represents the volume domain of the wedge brake. For material density, For displacement velocity, For stress tensor, For strain tensor, Calculate the elastic strain energy density by taking the dot product of the stress tensor and the strain tensor. To input force to the plunger, Let x be the displacement velocity of the brake wedge, x be the displacement of the brake wedge, and the upper point represent the time derivative. The smoothness loss in the stress prediction: , This is the predicted stress value.

8. The machine learning-based fatigue life assessment method for wedge brakes as described in claim 7, characterized in that, The steps for obtaining the fatigue life assessment value include: The number of cycles required for the cumulative damage value to be 1 at each sensitive node under the i-th type of stress amplitude is counted, and this number is recorded as the fatigue life of each sensitive node. , ; The fatigue life corresponding to the position with the largest cumulative damage value among all sensitive nodes is taken as the fatigue life assessment value of the wedge brake.

Images