Failure life prediction method for fluororubber seal of range extender for aging and fatigue coupled vehicle

Abstract

The invention relates to the technical field of fluororubber, in particular to an aging and fatigue coupling lower vehicle range extender fluororubber sealing failure life prediction method, which comprises the following steps of S1, establishing multi-physics field coupling finite element simulation analysis aiming at the working condition of a vehicle range extender; s2, establishing a fluororubber aging kinetic model and a fatigue crack propagation model; s3, integrating the fluororubber aging model and the fatigue model, and establishing a coupling effect model; s4, acquiring working condition environment change data of fluororubber of the vehicle range extender based on multiple sensors; s5, data cleaning and optimization of key features influencing the fatigue life of the fluororubber; and S6, improving the diversity of fluororubber performance change data based on the generative adversarial network. According to the method, firstly, two key factors of fluororubber aging and fatigue influencing the service life under the operation working condition of the vehicle range extender are integrated, a coupling effect model is established, and the performance change process of fluororubber under the complex working condition is reflected;

Description

Technical Field

[0001] The present invention relates to the technical field of fluororubber, and particularly to a method for predicting the failure life of fluororubber seals in vehicle range extenders under the coupling of aging and fatigue. Background Art

[0002] Fluororubber is a high-performance elastomer, which is widely used in fields such as automotive, aerospace, and petrochemical due to its excellent high-temperature resistance, chemical corrosion resistance, and good sealing performance. In vehicle range extenders, fluororubber seals play a crucial role in ensuring the safe operation of fuel, electric drive, and cooling systems. However, during long-term use, fluororubber materials are affected by the coupling of aging and fatigue, resulting in performance degradation and failure. Predicting its life is a key issue for improving the reliability and safety of equipment.

[0003] The working environment of vehicle range extenders is complex and variable. The interaction of factors such as temperature fluctuations, chemical medium effects, and mechanical vibrations makes fluororubber seals prone to aging and fatigue failure. Thermal-oxidative aging can lead to molecular chain breakage or crosslinking, causing material hardening and decreased elasticity; ozone aging may cause surface cracking, especially more significantly under dynamic stress; long-term immersion in chemical media (such as fuel and coolant) can cause swelling and affect the sealing performance; the cyclic action of mechanical loads can lead to fatigue crack propagation until fracture failure. The combined action of these complex factors significantly shortens the service life of the seals. However, existing methods for predicting the failure life of fluororubber seals are mostly based on single factors, and the aging and fatigue coupling mechanism in a multi-physical field coupling environment such as temperature field and mechanical load field has not been fully clarified.

[0004] At present, the methods for predicting the sealing failure life of fluororubber mainly include the empirical model method, the experimental statistics method, and the numerical simulation method, etc. The empirical model method usually based on the results of accelerated aging tests, and calculates the life through establishing empirical formulas related to environmental variables. However, this method has poor adaptability to specific working conditions. The experimental statistics method analyzes the life distribution characteristics of materials through a large number of experimental data, and uses statistical models such as Weibull distribution and Cox regression for life prediction. However, the experimental statistics method requires a large amount of data support, and it is difficult to describe the overall distribution of samples under small sample conditions, resulting in limited prediction accuracy. The numerical simulation method relies on tools such as finite element analysis, and combines the constitutive model and failure criterion of materials for life prediction. However, due to the complex acquisition of material parameters and large computational amount, it is difficult to be widely applied in actual engineering. In recent years, neural networks have been preliminarily applied in material life prediction due to their powerful data processing capabilities. By training experimental data, neural networks can capture the complex non-linear relationships between multi-factors such as temperature, vibration, and medium and the failure life. However, traditional neural network models usually ignore the physical properties of materials, resulting in the lack of constraints of physical laws in prediction results; at the same time, limited by the difficulty of obtaining experimental data, the effect of neural network models on small samples still needs to be improved. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems existing in the prior art, and propose a method for predicting the sealing failure life of fluororubber for vehicle range extenders under the coupling of aging and fatigue.

[0006] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A method for predicting the sealing failure life of fluororubber for vehicle range extenders under the coupling of aging and fatigue, comprising the following steps:

[0008] S1. Establish a finite element simulation analysis of multi-physical field coupling for the working conditions of vehicle range extenders;

[0009] S2. Establish an aging kinetics model and a fatigue crack growth model of fluororubber;

[0010] S3. Integrate the aging and fatigue models of fluororubber and establish a coupling action model;

[0011] S4. Obtain the data of the environmental change of the fluororubber for vehicle range extenders based on multi-sensors;

[0012] S5. Clean the data and select the key features that affect the fatigue life of fluororubber;

[0013] S6. Improve the diversity of the data of the performance change of fluororubber based on the generative adversarial network;

[0014] S7. Evaluate the effectiveness of data - enhanced samples in combination with the fluororubber aging and fatigue coupling effect model;

[0015] S8. Divide the data - enhanced samples based on the K - fold cross - validation method;

[0016] S9. Construct the components for predicting the fluororubber life based on the temporal convolutional network;

[0017] S10. Introduce a physical loss term based on the fluororubber aging and fatigue coupling effect model;

[0018] S11. Design a time - varying weight to balance each component of the total loss function and train the model to output the life distribution.

[0019] Preferably, in the step S1, first determine the geometric model of the vehicle - mounted range extender, including the fluororubber seal and its surrounding related components, then precisely construct the geometric shapes of these components using 3D modeling software, and finally define the physical fields and their interactions.

[0020] Preferably, in the step S2, first establish the fluororubber aging kinetics model, and its formula is:

[0021]

[0022] where k is the aging rate (unit: d), A is the frequency factor (unit: d -1 )), E a is the activation energy (unit: J / mol), R is the molar gas constant (unit: J / (mol·K)), and T is the absolute temperature (unit: Kelvin);

[0023] Then establish the fatigue crack growth model, and its formula is:

[0024]

[0025] where a is the crack length, N is the number of cycles, ΔK is the stress intensity factor range, C const and m are material constants.

[0026] Preferably, in the step S3, integrate the fluororubber aging kinetics model and the fatigue crack growth model in step S2, and its integration formula is:

[0027]

[0028] where β is the aging influence coefficient, Y is the geometric factor related to the crack shape and loading mode, k0 is the aging rate without fatigue damage, γ is the fatigue influence coefficient, and D is the fatigue damage.

[0029] Preferably, in the step S5, data cleaning includes outlier removal and data standardization; the preferred methods for key features are one of principal component analysis and correlation analysis;

[0030] Among them, using the principal component analysis method to achieve the characteristics affecting the fatigue life of fluororubber, the original data of fluororubber collected by multiple sensors in the operation of the vehicle range extender in step S4, after data cleaning, are arranged and sorted in time series to form a standardized data matrix Z.

[0031] First, calculate the covariance matrix R. The covariance matrix is used to measure the correlation between different features, and the formula is as follows:

[0032]

[0033] where n is the number of samples, Z T is the transpose matrix of Z, and the elements of the covariance matrix R are expressed as r ij the covariance between feature i and feature j;

[0034] Then, perform eigenvalue decomposition. Performing eigenvalue decomposition on the covariance matrix R can obtain eigenvalues λ1, λ2, …, λ p (p is the number of features) and the corresponding eigenvectors a1, a2, …, a p , satisfying R = 1AP -1 , where P = [a1, a2, …, a p is the matrix composed of eigenvectors, and A = diaj(λ1, λ2, …, λ p ) is a diagonal matrix, and the elements on the diagonal are eigenvalues;

[0035] Finally, calculate the variance contribution rate and cumulative variance contribution rate of the principal components. Calculate the variance contribution rate η i of each principal component, and the formula is as follows:

[0036]

[0037] The cumulative variance contribution rate is the sum of the variance contribution rates of the first num principal components, that is:

[0038]

[0039] Preferably, in the step S6, the generative adversarial network includes a generator and a discriminator. The purpose of the generator is to generate data similar to the real fluororubber performance change data according to the input random noise, and the discriminator is used to judge whether the input data is real data or data generated by the generator;

[0040] The loss function and optimizer of the generative adversarial network. The objective function of the generative adversarial network is as follows:

[0041]

[0042] Among them, G is the generator, D is the discriminator, x is the real data, z is the random noise vector, and the Adam optimizer is selected;

[0043] Then, the data distribution of the generated samples is judged by statistical indicators to determine whether it is consistent with the original data distribution.

[0044] Preferably, in the step S7, after data augmentation is completed, combined with the previously established aging and fatigue coupling model, the effectiveness of the augmented samples is strictly tested to ensure that the augmented data can truly reflect the performance changes of fluororubber under actual working conditions;

[0045] First, the augmented sample data is preprocessed to make it conform to the input format and requirements of the aging and fatigue coupling model, which includes the working condition parameters of fluororubber, such as the time-varying sequences of temperature, stress, and chemical medium concentration. Ensure that the time step, unit, etc. of these parameters in the augmented data are consistent with the model settings, and perform necessary normalization or format conversion operations on the data. Then, the preprocessed augmented sample data is input into the aging and fatigue coupling model group by group according to the time series; the augmented sample contains data of N time steps. For each time step t, the corresponding temperature T(t), stress σ(t), and chemical medium concentration C(t) parameters are input into the model;

[0046] Subsequently, the model outputs include the change of the aging degree a(t) of fluororubber over time and the change of the fatigue crack growth length a(t) over the number of cycles N. Compare the aging degree and fatigue crack growth results output by the model when inputting the original data; calculate the differences of the key output indicators, and use the average relative error to evaluate the output aging degree and fatigue crack growth results. The average relative error of the aging degree is:

[0047]

[0048] The average relative error of the fatigue crack growth length is:

[0049]

[0050] Among them, a augmented (t) is the aging degree of the augmented data at time step t, a orjginal (t) is the aging degree of the input original data at time step t, a augmented (t) is the fatigue crack growth length of the augmented data at time step t, a orjginal (t) is the fatigue crack growth length of the input original data at time step t;

[0051] Finally, deeply analyze the aging degree output by the model and the changing trend of fatigue crack growth over time. When observing the input enhanced data, observe the growth curve of the aging degree over time and the changing curve of the fatigue crack growth length over the number of cycles.

[0052] Preferably, in the step S9, use a temporal convolutional network to regard the fluororubber life prediction as a temporal prediction problem affected by multiple factors. The temporal convolutional network includes a temporal convolutional network main body, a fully connected layer, and a dual-head output layer;

[0053] The temporal convolutional network main body is stacked by residual blocks. Each residual block includes a one-dimensional causal convolutional layer, layer normalization, a ReLU activation function, and a Dropout layer;

[0054] Fully connected layer and dual-head output layer: Flatten the TCN output features and connect them to the fully connected layer. Set two fully connected layers. The first layer has 128 nodes, and the activation function is ReLU, which is used for preliminary feature fusion; the second layer has 64 nodes, also using ReLU, to further refine the features;

[0055] The dual-head output layer is set with two output branches. One of them outputs the Weibull distribution parameters (i.e., the shape parameter s and the scale parameter λ). To ensure that the output parameters conform to the physical meaning (both are positive values), apply the Softplus activation function to the two nodes of this branch respectively. The Softplus function is defined as follows:

[0056] Softplus(x) = ln(1 + e x )

[0057] For the component of the fluororubber life prediction of the temporal convolutional network, its loss function is composed of the life value prediction loss. The mean squared error is used to measure the deviation between the predicted life value and the true life value. Let the predicted life value be The true life value is y, and N is the number of samples. Then the life value prediction loss is as follows:

[0058]

[0059] Preferably, in the step S10, the temporal convolutional network in the step S9 outputs the life value T pred , and convert it to the corresponding number of cycles N according to the loading frequency of the data used pred , that is, there is N pred = f × T pred , and establish a connection between the life value and the number of cycles in the fatigue crack growth equation through this method;

[0060] In the step S3, the relationship between the crack length a and the number of cycles N can be obtained by integrating the fatigue crack growth equation. Its integration formula is:

[0061]

[0062] Substitute the converted number of cycles N pred into the above equation to calculate the theoretical crack length a(N pred ) at the predicted life;

[0063] Integrating the aging equation in step S3 gives:

[0064]

[0065] Substitute the life value T output by the time convolutional network pred into this equation to calculate the theoretical aging degree a(T pred ) at the predicted life.

[0066] Preferably, in step S11, based on the loss term of the time convolutional network in step S9 and the physical loss term in step S10, the total loss function formula of the physics-informed neural network at this time is:

[0067] L total = L data + L phy

[0068] To better guide the model to learn accurate quality estimation, reasonable weight settings are made for each loss of the physics-informed neural network, and the weight calculation formula is:

[0069]

[0070] L total = a(t)L total + (1 - a(t))L phy

[0071] Finally, use this total loss function to guide the model training, and finally output the Weibull distribution shape parameter s and scale parameter λ of the fluororubber, as well as a representative predicted life value.

[0072] Compared with the existing technologies, the advantages of the present invention are as follows:

[0073] 1. The method for predicting the failure life of the fluororubber seal of the vehicle-mounted range extender under the coupling of aging and fatigue proposed by the present invention first integrates two key factors affecting the life, namely fluororubber aging and fatigue, under the operating conditions of the vehicle-mounted range extender, and establishes a coupling action model, which reflects the performance change process of the fluororubber under complex working conditions.

[0074] 2. Secondly, based on the generative adversarial network, data augmentation is performed on the small sample of fluororubber performance change data under actual working conditions, and the effectiveness of the augmented samples is verified by combining the coupling action model, providing sufficient samples similar to the actual working condition data distribution for data-driven life prediction methods.

[0075] 3. Finally, a physical information neural network is constructed based on the components of the fluororubber life prediction built by the temporal convolutional network and the physical loss term introducing the coupling action model. The network training is constrained by physical laws, enabling the Weibull distribution parameters of the predicted fluororubber seal failure life by the model to closely fit the actual physical scenario. Brief Description of the Drawings

[0076] Figure 1 It is the framework flow chart of the method for predicting the failure life of fluororubber seals in vehicle range extenders under the coupling of aging and fatigue proposed by the present invention;

[0077] Figure 2 It is the linear regression analysis diagram in the embodiment. Detailed Embodiment

[0078] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0079] Refer to Figure 1 - Figure 2 , the method for predicting the failure life of fluororubber seals in vehicle range extenders under the coupling of aging and fatigue includes the following steps:

[0080] S1. Establish a finite element simulation analysis of multi-physical field coupling for the working conditions of vehicle range extenders;

[0081] S2. Establish a fluororubber aging kinetics model and a fatigue crack growth model;

[0082] S3. Integrate the fluororubber aging and fatigue models and establish a coupling action model;

[0083] S4. Obtain the data of the working condition environment change of fluororubber in vehicle range extenders based on multi-sensors;

[0084] S5. Data cleaning and selection of key features affecting the fatigue life of fluororubber;

[0085] S6. Improve the diversity of fluororubber performance change data based on the generative adversarial network;

[0086] S7. Evaluate the effectiveness of the data-augmented samples in combination with the fluororubber aging and fatigue coupling action model;

[0087] S8. Divide the data augmentation samples based on the K-fold cross-validation method;

[0088] S9. Construct the components for predicting the life of fluororubber based on the temporal convolutional network;

[0089] S10. Introduce the physical loss term based on the coupling action model of fluororubber aging and fatigue;

[0090] S11. Design a time-varying weight to balance each component of the total loss function and train the model to output the life distribution.

[0091] Specifically:

[0092] In the first step, establish a finite element simulation analysis of multi-physical field coupling for the working conditions of vehicle range extenders. First, determine the geometric model of the vehicle range extender, including fluororubber seals and related components around them. Use 3D modeling software (such as SolidWorks, Catia, etc.) to accurately construct the geometric shapes of these components. Then, define the physical fields and their interactions. Consider the temperature field, stress field, chemical field, etc.

[0093] 1.1 For the temperature field, according to the actual working temperature range and heat conduction characteristics of the range extender, set the thermal boundary conditions, such as the heat source temperature of the engine, heat dissipation conditions, etc. The heat conduction equation can be expressed as:

[0094]

[0095] Among them, ρ is the material density, c p is the specific heat capacity, T is the temperature, t is the time, k is the thermal conductivity, and Q is the internal heat source.

[0096] 1.2 For the stress field, considering the vibration and mechanical load during the operation of the range extender, apply the corresponding displacement boundary conditions and force boundary conditions. Use the linear elastic constitutive equation to describe the mechanical behavior of the material. At the same time, considering the hyperelastic characteristics of fluororubber, hyperelastic constitutive models such as the Mooney-Rivlin model can be used.

[0097] σ = Eε

[0098] Among them, σ is the stress, E is the elastic modulus, and ε is the strain.

[0099] 1.3 For the chemical field, according to the diffusion and reaction characteristics of chemical media such as fuel and lubricating oil, set the concentration boundary conditions of chemical substances and the diffusion equation. Fick's first law describes the diffusion flux J as:

[0100]

[0101] Among them, D is the diffusion coefficient and C is the concentration of chemical substances.

[0102] Import the equations and boundary conditions of the above physical fields into a finite element analysis software (such as ANSYS, etc.), perform mesh generation, select appropriate element types (such as tetrahedral elements, hexahedral elements, etc.), and solve to obtain results such as the temperature, stress, and chemical substance concentration distribution of fluororubber under the coupling action of multiple physical fields.

[0103] In the second step, establish the aging kinetics model and fatigue crack growth model of fluororubber.

[0104] 2.1 Establish the aging kinetics model of fluororubber: In the complex working condition environment of vehicle range extenders, the aging process of fluororubber is affected by various factors, and temperature is one of the most critical factors. To accurately describe the aging behavior of fluororubber, we construct an aging kinetics model based on the Arrhenius equation, which can well reflect the influence of temperature on the chemical reaction rate, and the aging of fluororubber is essentially a chemical reaction process. The equation is as follows:

[0105]

[0106] where, k is the aging rate (unit: d), A is the frequency factor (unit: d -1 ), E a is the activation energy (unit: J / mol), R is the molar gas constant (unit: J / (mol·K)), and T is the absolute temperature (unit: Kelvin). To more intuitively describe the change of the aging degree of fluororubber with time, establish the relationship between the aging degree a and time t. Considering the general law of the aging process, the first-order reaction kinetics equation is used to describe it:

[0107]

[0108] Integrate the first-order reaction kinetics equation with respect to time t to obtain the following equation:

[0109] a = 1 - exp(-kt)

[0110] In summary, a kinetics model that can quantitatively describe the change of the aging degree of fluororubber with time at different temperatures is obtained, providing an important theoretical basis for subsequent analysis of the performance degradation of fluororubber.

[0111] 2.2 Establish the fatigue crack growth model: In the complex working condition environment of vehicle range extenders, fluororubber often also bears alternating stress, and the growth of fatigue cracks is another important factor leading to its failure. To describe this failure factor, the Paris law is used to establish a fatigue crack growth model, which is widely used in the field of material fatigue and can well describe the relationship between the fatigue crack growth rate and the stress intensity factor range. The law is as follows:

[0112]

[0113] where a is the crack length, N is the number of cycles, ΔK is the stress intensity factor range, C const and m are material constants. By establishing such a fatigue crack growth model, the growth law of cracks in fluororubber under alternating stress can be deeply understood, providing a theoretical support for predicting the fatigue life of fluororubber and evaluating its reliability in vehicle range extenders.

[0114] Step 3: Integrate the fluororubber aging and fatigue models and establish a coupling model. After establishing the fluororubber aging kinetics model and the fatigue crack growth model respectively, these two models need to be integrated to consider the mutual influence between aging and fatigue, so as to more realistically reflect the performance changes of fluororubber under the working conditions of vehicle range extenders.

[0115] 3.1 Influence of aging on fatigue: Aging will cause changes in the mechanical properties of fluororubber. Among them, the change of elastic modulus has a greater impact on fatigue performance. There is the following relationship between the aging degree a and the elastic modulus E:

[0116] E = E0(1 - βa)

[0117] where E0 is the initial elastic modulus of fluororubber and β is the aging influence coefficient. When fluororubber ages, its elastic modulus decreases, which will cause the strain of fluororubber to increase under the same stress, resulting in more obvious local stress concentration phenomena and accelerating the initiation and propagation of fatigue cracks.

[0118] Substitute the aged elastic modulus into the calculation formula of the stress intensity factor in the fatigue crack growth model. For common crack shapes (such as type I cracks), the stress intensity factor is related to parameters such as stress and crack length, and its expression is:

[0119]

[0120] where Y is the geometric factor related to the crack shape and loading mode. Substitute the corrected one into the fatigue crack growth rate equation to obtain the fatigue crack growth rate equation considering the influence of aging:

[0121]

[0122] 3.2 Influence of fatigue on aging: The microcracks generated during fatigue will provide channels for the intrusion and diffusion of chemical media (such as fuel, lubricating oil, etc.), thus accelerating the aging of fluororubber. We assume that there is the following linear relationship between the fatigue damage ρ (which can be represented by the crack density D, that is, D = fρ) and the aging rate k:

[0123] k = k0(1 + γD)

[0124] Among them, k0 is the aging rate without fatigue damage, and γ is the fatigue influence coefficient. As the fatigue crack expands, D increases, and the aging rate k also increases accordingly, accelerating the aging process of fluororubber. Substituting this formula into the relationship between the aging degree a and time t, the following formula is obtained, realizing the coupling of the influence of fatigue on aging in the model:

[0125]

[0126] In summary, a coupling model of fluororubber aging and fatigue is obtained. This model consists of the following two coupled differential equations, and the parameter meanings in the formula remain consistent:

[0127]

[0128] Fourthly, based on multi-sensors, data on the changing working conditions and environments of fluororubber in vehicle range extenders are obtained. The following is described according to temperature change data, mechanical pressure change data, and chemical medium concentration data.

[0129] Install temperature sensors (such as thermocouples or infrared temperature sensors) on the vehicle range extender, and arrange them at key positions near the fluororubber seal, such as near the engine heat source, the contact between the seal and metal parts, etc., to monitor the temperature changes of the fluororubber in real time. The sampling frequency of the temperature sensor is set to ensure that rapid temperature changes and fluctuations can be captured.

[0130] Install pressure sensors and vibration sensors (such as acceleration sensors) to measure the mechanical stress borne by the fluororubber during the operation of the range extender. The pressure sensor can be installed in parts related to the fluororubber seal, such as oil circuits and gas circuits, to measure the medium pressure; the vibration sensor is installed on the housing of the range extender or components connected to the fluororubber seal to measure parameters such as vibration acceleration and frequency. The sampling frequencies of these sensors can be set to dozens to hundreds of times per second according to the actual situation to accurately reflect the dynamic changes of mechanical stress.

[0131] For the monitoring of chemical medium concentration, chemical sensors (such as electrochemical sensors) can be installed in parts such as the fuel tank and lubricating oil circuit to regularly measure the composition and concentration changes of chemical media such as fuel and lubricating oil.

[0132] Connect all sensors through a data acquisition system (such as a data acquisition card, industrial control computer, etc.) to collect and store sensor data in real time, forming an original data set of the changing working conditions and environments of fluororubber in vehicle range extenders.

[0133] Step 5: Data cleaning and optimization of key features affecting the fatigue life of fluororubber. Data cleaning includes outlier removal and data standardization; the method for optimizing key features can be one of the methods such as principal component analysis, correlation analysis, etc.; the following will be described separately.

[0134] 5.1 Data cleaning:

[0135] First, deal with the outliers in the collected data, and identify and process outliers by setting reasonable thresholds. For example, for vibration acceleration data, the 3 - sigma principle can be adopted. Calculate the mean μ and standard deviation σ of the vibration acceleration data, and regard the data outside the range of [μ - 3σ, μ + 3σ] as outliers. Secondly, perform data standardization to eliminate the influence of the dimension between different features. The commonly used standardization method is Z - score standardization.

[0136] 5.2 Use the principal component analysis method to optimize the features affecting the fatigue life of fluororubber: For the original data of temperature, pressure, vibration, chemical medium concentration, etc. of fluororubber collected by multi - sensors in the operation of vehicle range extenders in the fourth step, after the data cleaning in 5.1, arrange and organize them in time series to form a standardized data matrix Z.

[0137] First, calculate the covariance matrix R. The covariance matrix is used to measure the correlation between different features, and the formula is as follows:

[0138]

[0139] where n is the number of samples, Z T is the transpose matrix of Z, and the element of the covariance matrix R is expressed as r ij the covariance between feature i and feature j;

[0140] Then perform eigenvalue decomposition. Perform eigenvalue decomposition on the covariance matrix R to obtain eigenvalues λ1, λ2, …, λ p (p is the number of features) and the corresponding eigenvectors a1, a2, …, a p , satisfying R = 1AP -1 , where P = [a1, a2, …, a p is the matrix composed of eigenvectors, and A = diaj(λ1, λ2, …, λ p ) is a diagonal matrix, and the elements on the diagonal are eigenvalues. The magnitude of the eigenvalue reflects the variance size of the corresponding principal component. The larger the variance, the more original data information the principal component contains. Finally, calculate the variance contribution rate and cumulative variance contribution rate of the principal components. Calculate the variance contribution rate η i of each principal component, and the formula is as follows:

[0141]

[0142] The cumulative variance contribution rate is the sum of the variance contribution rates of the first num principal components, that is:

[0143]

[0144] By observing the cumulative variance contribution rate, determine the number of principal components to be retained. Usually, select the principal components whose cumulative variance contribution rate reaches a certain threshold (such as 85%, 90% or 95%). The original features corresponding to the retained principal components are the key features that have a greater impact on the fatigue life of fluororubber.

[0145] Step 6: Improve the diversity of fluororubber performance change data based on the generative adversarial network.

[0146] 6.1 Construct a generative adversarial network: The generative adversarial network consists of two parts: a generator and a discriminator. The purpose of the generator is to generate data similar to the real fluororubber performance change data according to the input random noise, that is, to be as close as possible to the same distribution as the data obtained in the fifth step. The structure of the generator adopts a multi-layer perceptron (MLP). The number of nodes in the input layer corresponds to the dimension of the random noise vector; multiple hidden layers are set in the middle and the activation function ReLU is set to enhance the ability of the generator; the number of nodes in the output layer is consistent with the dimension of the fluororubber performance characteristics, and the activation function uses Tanh to map the generated data to the same distribution interval as the data in the fifth step.

[0147] The discriminator is used to judge whether the input data is real data or data generated by the generator. It is also constructed based on a multi-layer perceptron. The number of nodes in the input layer is the same as the dimension of the fluororubber performance characteristics; multiple hidden layers are set in the middle and the activation function ReLU is set; there is only one node in the output layer, and the Sigmoid activation function is used. The output value is between 0 and 1, and the closer it is to 1, the higher the probability that the data is real data.

[0148] 6.2 Loss function and optimizer of the generative adversarial network: Randomly initialize the weights and biases of the generator and the discriminator, usually using the normal distribution initialization method. The objective function of the generative adversarial network is as follows:

[0149]

[0150] where G is the generator, D is the discriminator, x is the real data, z is the random noise vector, and the optimizer is Adam.

[0151] 6.3 Determine whether the data distribution of the generated samples is consistent with the original data distribution through statistical indicators: Calculate statistical indicators such as the mean, standard deviation, skewness, and kurtosis of the original data and the augmented data. Use hypothesis testing methods, such as the Kolmogorov-Smirnov test (KS test), to determine whether the augmented data and the original data come from the same distribution. For a given feature, the KS test calculates the maximum distance between the cumulative distribution functions of the two data sets and calculates the p-value based on this distance. If the p-value is greater than the significance level (e.g., 0.05), the null hypothesis that the two data sets come from the same distribution cannot be rejected, that is, it is considered that there is no significant difference between the distribution of the augmented data and the original data distribution.

[0152] Step 7: Evaluate the effectiveness of the data-augmented samples in combination with the fluororubber aging and fatigue coupling model.

[0153] After data augmentation, it is necessary to strictly test the effectiveness of the augmented samples in combination with the previously established aging and fatigue coupling model to ensure that the augmented data can truly reflect the performance changes of fluororubber under actual working conditions. The specific steps are as follows:

[0154] 7.1 Preprocess the augmented sample data to make it conform to the input format and requirements of the aging and fatigue coupling model. The input of this model usually includes the working condition parameters of fluororubber, such as the time-varying sequences of temperature, stress, chemical medium concentration, etc. Ensure that the time step, unit, etc. of these parameters in the augmented data are consistent with the model settings, and perform necessary normalization or format conversion operations on the data. Input the preprocessed augmented sample data into the aging and fatigue coupling model group by group according to the time series. The augmented sample contains data for N time steps. For each time step t, input the corresponding temperature T(t), stress σ(t), and chemical medium concentration C(t) parameters into the model.

[0155] 7.2 The model outputs include the change of the aging degree a(t) of fluororubber over time and the change of the fatigue crack growth length a(t) over the number of cycles N. Compare the aging degree and fatigue crack growth results output by the model when inputting the original data; calculate the differences in key output indicators, and use the average relative error to evaluate the output aging degree and fatigue crack growth results. The average relative error of the aging degree is:

[0156]

[0157] The average relative error of the fatigue crack growth length is:

[0158]

[0159] where a augmented (t) is the aging degree of the augmented data at time step t, a orjginal$(t)$ is the aging degree of the input original data at time step $t$, $a$ augmented $(t)$ is the fatigue crack growth length of the enhanced data at time step $t$, $a$ orjginal $(t)$ is the fatigue crack growth length of the input original data at time step $t$. If these average relative errors are within a preset reasonable range (such as less than 10%), it is preliminarily considered that the enhanced data is somewhat similar to the results generated by the original data-driven model, and the data has a certain degree of effectiveness.

[0160] 7.3 Deeply analyze the changing trends of the aging degree and fatigue crack growth output by the model over time. Observe the growth curve of the aging degree over time and the curve of the fatigue crack growth length over the number of cycles when inputting the enhanced data. Compare these curves with the corresponding curves when inputting the original data to judge whether their changing trends are similar. Ideally, the growth curves of the aging degree and fatigue crack growth generated by the enhanced data-driven model should have similar shapes and changing rules as those of the original data-driven curves.

[0161] Step 8: Divide the data augmentation samples based on the K-fold cross-validation method.

[0162] Divide the performance change data of fluororubber after data augmentation and effectiveness evaluation into a training set, a validation set, and a test set. Each sample is composed of a mapping pair of "characteristics affecting life - life under this characteristic". Using the K-fold cross-validation method, randomly divide the data set into K subsets of equal size. Successively use (K - 1) of these subsets as the training set and the remaining one subset as the validation set for K training and validation processes. During each training process, use the training set to train the physics-informed neural network model and evaluate the performance of the model on the validation set. Select the best model parameters according to the performance on the validation set.

[0163] Step 9: Construct the components for predicting the life of fluororubber based on the temporal convolutional network.

[0164] Use the temporal convolutional network to regard the prediction of the life of fluororubber as a time series prediction problem affected by multiple factors. The temporal convolutional network in this step consists of three parts: the main body of the temporal convolutional network, the fully connected layer, and the dual-head output layer, which are described separately below.

[0165] 9.1 Temporal Convolutional Network (TCN) Body: It is mainly composed of stacked residual blocks. Each residual block contains: a one-dimensional causal convolutional layer, layer normalization, a ReLU activation function, and a Dropout layer. The one-dimensional causal convolutional layer follows the causal principle when processing time series data, that is, the convolution operation only depends on the data at the current and past moments, and does not introduce information from future moments, ensuring the rationality of time series prediction. Its convolutional kernel slides along the time dimension. According to the set convolutional kernel size, stride, and dilation rate, it extracts local features from the input data, mining different levels of local feature patterns from the input data. As the network depth increases and the dilation rate changes dynamically, the receptive field gradually expands, capturing dependencies over longer time spans. Layer normalization makes the data distribution more stable across each feature dimension; the ReLU activation function improves the network's non-linear transformation ability, capturing the potential laws between the characteristics related to the fluororubber life and the life results. The Dropout layer avoids excessive dependence between neurons, effectively preventing overfitting, and ensuring that the network has good generalization performance when facing diverse fluororubber working condition data.

[0166] 9.2 Fully Connected Layers and Dual-Head Output Layer: After flattening the TCN output features, they are connected to the fully connected layers. Two fully connected layers are set. The first layer contains 128 nodes, and the activation function is ReLU, which is used for preliminary feature fusion; the second layer has 64 nodes, also using ReLU, to further refine the features.

[0167] The dual-head output layer has two output branches. One of them outputs the Weibull distribution parameters (i.e., the shape parameter s and the scale parameter λ). To ensure that the output parameters conform to physical meanings (both are positive values), the Softplus activation function is applied to the two nodes of this branch respectively. The Softplus function is defined as follows:

[0168] Softplus(x) = ln(1 + e x )

[0169] The other output of the dual-head output layer is the specific life value of the fluororubber, which outputs a predicted fluororubber life value based on the current input working conditions with a linear activation function. This life value can be understood as a representative life estimate given according to the complex data patterns learned by the network. It is independent of the Weibull distribution parameter output branch and together serves the life prediction task.

[0170] The last output of the dual-head output layer is the aging degree a pred and the fatigue crack length a pred , which prepares for introducing physical laws into the overall physical information neural network.

[0171] 9.3 Definition of Loss Function: For the fluororubber life prediction component of the temporal convolutional network, its loss function consists of the life value prediction loss. The mean squared error is used to measure the deviation between the predicted life value and the true life value. Let the predicted life value be the true life value be y, and N be the number of samples. Then the life value prediction loss is as follows:

[0172]

[0173] Step 10: Introduce a physical loss term based on the fluororubber aging and fatigue coupling model.

[0174] 10.1 In Step 9, the temporal convolutional network output the life value T pred , which is converted into the corresponding number of cycles N according to the loading frequency of the data used pred , that is, there is N pred = f × T pred , and through this method, a connection is established between the life value and the number of cycles in the fatigue crack growth equation.

[0175] 10.2 In Step 3, the aging and fatigue coupling model was obtained, that is, Equation (13). Based on the life value, theoretical physical quantities are calculated. For the fatigue part, the relationship between the crack length a and the number of cycles N can be obtained by integrating the fatigue crack growth equation. Assuming the initial crack length a = 0, its integration formula is:

[0176]

[0177] Substitute the converted number of cycles N pred into the above equation to calculate the theoretical crack length a(N pred ) at the predicted life;

[0178] For the integration of the aging equation in Step 3, it can be obtained:

[0179]

[0180] Assuming the initial aging degree a0 = 0, substitute the life value T pred output by the temporal convolutional network into this equation to calculate the theoretical aging degree a(T pred ) at the predicted life.

[0181] 10.3 Introduce the physical loss term as follows:

[0182] L phy = (a pred - a(N pred )) 2 + (a pred - a(T pred )) 2

[0183] where a pred is the predicted crack length, a(N pred ) is the theoretical crack length at the predicted life, a pred is the predicted aging degree, a(T pred ) is the theoretical aging degree at the predicted life.

[0184] The eleventh step is to design a time-varying weight to balance each component of the total loss function and train the model to output the life distribution.

[0185] Based on the loss term of the temporal convolutional network in the ninth step and the physical loss term in the tenth step, the formula for the total loss function of the physics-informed neural network at this time is:

[0186] L total = L data + L phy

[0187] When over-relying on the data-driven part, it may lead to underfitting of the model. At this time, appropriately increasing the weight of the physical loss term and leveraging the prior knowledge of the physical model can improve the performance of the model. When the sample data volume is rich and accurate, the weight can be appropriately adjusted to give the data-driven part more opportunities to learn the detailed information in the data and further optimize the quality estimation result. Therefore, in order to better guide the model to learn accurate quality estimation, reasonable weight settings need to be made for each loss of the physics-informed neural network. The weight calculation formula is:

[0188]

[0189] L total = a(t)L total + (1 - a(t))L phy

[0190] Finally, use this total loss function to guide the model training, and finally output the Weibull distribution shape parameter s and scale parameter λ of the fluororubber, as well as a representative predicted life value.

[0191] Next, taking the fluororubber seal of a certain type of vehicle-mounted range extender as an example, the implementation process of the failure life prediction technology scheme of the fluororubber seal of the vehicle-mounted range extender under the coupling of aging and fatigue is elaborated in detail.

[0192] The first step is multi-physics field coupling finite element simulation

[0193] A fluororubber seal of a vehicle-mounted range extender is used in the fuel system. Use SolidWorks to accurately construct the geometric model of it and its surrounding components. During operation, the temperature range is set to 50 - 120 °C, and heat dissipation relies on heat sinks and fans. According to the heat conduction equation, where the material density ρ = 1800 kg / m3 , the specific heat capacity c p = 1200 J / (kg·K), the thermal conductivity k = 0.2 W / (m·K), and the internal heat source Q = 500 W / m 3 . In terms of the stress field, due to the vibration and mechanical load of the range extender, the seal is subjected to an alternating pressure of 0.5 - 2 MPa, and the displacement boundary conditions are determined according to the installation structure. The Mooney-Rivlin model is used to describe the hyperelastic characteristics of fluororubber. In the chemical field, the diffusion coefficient D of the fuel at the seal is 5×10 -10 m 2 / s. The equations and boundary conditions of the above physical fields are imported into ANSYS, and tetrahedral elements are selected for mesh generation and solution, so as to obtain the temperature, stress, and fuel concentration distribution results.

[0194] Step 2: Establish a fluororubber aging kinetics model

[0195] Based on the aging-related data obtained from the finite element simulation in the first step, taking a group as an example, which are the compression set retention rate data at different temperatures and the tensile strength data at different temperatures, as shown in the following table:

[0196] Compression set retention rate P data at different temperatures

[0197]

[0198]

[0199] For the above table, based on the least squares principle, the Arrhenius formula is used to perform a regression analysis on the compression set retention rate P and the aging time t at different temperature points in Table 1 where a is a correction constant, and the reaction rate constant K at the corresponding temperature point is obtained i , i corresponds to different temperature points, and the estimated value of the experimental constant B i is as follows:

[0200]

[0201] Take the mean value of the experimental constant B i at six temperature points as: 0.8805, and the correction constant a is 1.01.

[0202] Based on the least squares principle, using the empirical relationship between the performance change rate constant K and the thermodynamic temperature T, that is, formula (1), a regression analysis is performed on the performance change rate constant K and the thermodynamic temperature T data at different temperature points in Table 1. After taking the logarithm of Ki at different temperature points in Table 1, a linear regression analysis is then performed on lnKi and 1 / T data, as Figure 2 shown.

[0203] The linear regression equation is: ln(ki)=-998.4856*(1 / T)+-2.7656

[0204] Similarly, the fatigue results obtained from the simulation analysis can be analyzed.

[0205] The third step is to integrate the fluororubber aging and fatigue models and establish a coupling model. According to experience, the aging influence coefficient β = 0.2, the fatigue influence coefficient γ = 0.1, and the relationship between crack density and fatigue damage is D = 0.01ρ. Aging will reduce the elastic modulus. According to E = E0 (1-βa), it is substituted into the stress intensity factor calculation formula to correct the fatigue crack growth rate equation. Fatigue will accelerate aging, k = k0 (1 + γD), which is substituted into the aging degree and time relationship equation to achieve the coupling of aging and fatigue.

[0206] In the fourth step, a thermocouple (with an accuracy of ±0.5°C) is installed near the fluororubber seal to monitor the temperature, a pressure sensor (with an accuracy of ±0.05MPa) to measure the oil pressure, and a chemical sensor to monitor the fuel concentration. The data collection frequency is set to times per day, and the data is collected continuously for 50 days to obtain the original data set. The time unit of the data collected in this way is day, which is convenient for subsequent correspondence and analysis with the time parameters in the aging and fatigue models.

[0207] Step 5: Data cleaning and optimization of key features that affect rubber fatigue life. Data cleaning: Use the 3-times standard deviation principle to eliminate outliers. For example, the vibration acceleration data is calculated to have a mean value of μ = 0.5 m / s 2 , standard deviation σ=0.1m / s 2 , will exceed [0.2,0.8]m / s 2 The data in the range are eliminated. Then the Z-score is used to standardize the data so that the mean is 0 and the standard deviation is 1. Feature optimization: The covariance matrix of the temperature, pressure, vibration, and fuel concentration data is calculated and the eigenvalue decomposition is performed. The principal component with a cumulative variance contribution rate of 90% is selected to determine the key features.

[0208] Step 6: Enhance the diversity of fluororubber performance change data based on the generative adversarial network. The number of input layer nodes of the generator of the generative adversarial network is 10. Two hidden layers are set in the middle, with the number of nodes being 64 and 32 respectively. The ReLU activation function is adopted. The number of output layer nodes is consistent with the dimension of fluororubber performance characteristics, and the activation function is Tanh. The input layer of the discriminator has the same dimension as the performance characteristics. There are also two hidden layers in the middle, and the output layer has 1 node, with the activation function being Sigmoid. The loss function is the same as formula (17). Adam is selected as the optimizer. After generating the data, it is evaluated by calculating the mean, standard deviation, skewness, kurtosis and performing the KS test. Suppose the mean of the original data is 0.05, the standard deviation is 0.03, the mean of the enhanced data is 0.052, the standard deviation is 0.031, and the p-value of the KS test is 0.06 > 0.05, indicating that there is no significant difference between the enhanced data and the original data distribution.

[0209] Step 7: Evaluate the effectiveness of the data enhancement samples in combination with the fluororubber aging and fatigue coupling action model. Preprocess the enhanced data to make it conform to the input format and requirements of the coupling model, and ensure that the time step, unit, etc. of the data are consistent with the model settings. Input the preprocessed enhanced sample data into the aging and fatigue coupling model group by group according to the time series. Calculate that the average relative error of the aging degree is 8% < 10%, and the average relative error of the fatigue crack propagation length is 9% < 10%. Moreover, the aging degree growth curve and the fatigue crack propagation curve output by the model when inputting the enhanced data are similar to the corresponding curves when inputting the original data, thereby confirming the effectiveness of the data.

[0210] Step 8: Divide the data enhancement samples based on the K-fold cross-validation method. The 5-fold cross-validation method is used to divide the data. Each time during training, 4 subsets are selected as the training set and 1 subset is selected as the validation set.

[0211] Step 9: Construct the components for predicting the fluororubber life based on the temporal convolutional network. Main body of the temporal convolutional network (TCN): It consists of 4 residual blocks. The convolutional kernel size of the one-dimensional causal convolutional layer is 3, the stride is 1, and the dilation rate doubles each time starting from 1. The data distribution is stabilized through layer normalization, and the ReLU activation function enhances the non-linear transformation ability. The Dropout layer prevents overfitting. Fully connected layer and multi-head output layer: After flattening the output features of the TCN, they are connected to the fully connected layer. The first layer has 128 nodes, and the second layer has 64 nodes. The activation functions are all ReLU. One branch of the multi-head output layer outputs the Weibull distribution parameters, and the Softplus activation is used to ensure that the parameters are positive; the second branch outputs the life value through linear activation. The third branch outputs the predicted aging degree and fatigue crack length.

[0212] Steps 10 and 11 can be carried out according to the above instructions.

[0213] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention should cover within the protection scope of the present invention by making equivalent substitutions or changes according to the technical solution and inventive concept of the present invention.

Claims

1. A method for predicting the failure life of fluororubber seals of vehicle range extenders under the coupling of aging and fatigue, characterized in that, It includes the following steps: S1. Establish a finite element simulation analysis of multi-physical field coupling for the operating conditions of vehicle range extenders; S2. Establish a fluororubber aging kinetics model and a fatigue crack growth model; S3. Integrate the fluororubber aging and fatigue models and establish a coupling effect model; S4. Obtain data on the changes in the operating conditions of fluororubber for vehicle range extenders based on multi-sensors; S5. Data cleaning and optimization of the key features affecting the fatigue life of fluororubber; S6. Improve the diversity of data on the performance changes of fluororubber based on a generative adversarial network; S7. Evaluate the effectiveness of data-enhanced samples in combination with the fluororubber aging and fatigue coupling effect model; S8. Divide the data-enhanced samples based on the K-fold cross-validation method; S9. Construct the components for predicting the life of fluororubber based on a temporal convolutional network; S10. Introduce a physical loss term based on the fluororubber aging and fatigue coupling effect model; S11. Design a time-varying weight to balance each component of the total loss function and train the model to output the life distribution.

2. The method for predicting the failure life of the fluororubber seal of the vehicle range extender under the coupling of aging and fatigue according to claim 1, wherein, In the step S1, first determine the geometric model of the vehicle range extender, including the fluororubber seal and related components around it, then accurately construct the geometric shapes of these components using 3D modeling software, and finally define the physical fields and their interactions.

3. The method for predicting the failure life of the fluororubber seal of the vehicle range extender under the coupling of aging and fatigue according to claim 1, characterized in that, In the step S2, first establish a fluororubber aging kinetics model, and its formula is: where k is the aging rate (unit: d), A is the frequency factor (unit: d -1 ), E a is the activation energy (unit: J / mol), and R is the molar gas constant (unit: J / (mol·K)), T is the absolute temperature (unit: Kelvin); Then establish a fatigue crack growth model, and its formula is: where a is the crack length, N is the number of cycles, ΔK is the stress intensity factor range, C const and m are material constants.

4. The method for predicting the failure life of fluororubber seals of vehicle range extenders under the coupling of aging and fatigue according to claim 1, wherein, In the step S3, integrate the fluororubber aging kinetics model and the fatigue crack growth model in step S2, and its integration formula is: Among them, β is the aging influence coefficient, Y is the geometric factor related to the crack shape and loading mode, k0 is the aging rate without fatigue damage, γ is the fatigue influence coefficient, and D is the fatigue damage.

5. The method for predicting the failure life of fluororubber seals of vehicle range extenders under the coupling of aging and fatigue according to claim 1, wherein In the step S5, data cleaning includes outlier removal and data standardization; the method for optimizing key features is one of principal component analysis and correlation analysis; Among them, the principal component analysis method is used to realize the features affecting the fatigue life of fluororubber. For the original data of fluororubber collected by multi-sensors in the operation of vehicle range extenders in step S4, after data cleaning, they are arranged and sorted in time series to form a standardized data matrix Z. First calculate the covariance matrix R. The covariance matrix is used to measure the correlation between different features, and the formula is as follows: where n is the number of samples, Z T is the transpose matrix of Z, and the elements of the covariance matrix R represent r ij the covariance between feature i and feature j; Then perform eigenvalue decomposition. Perform eigenvalue decomposition on the covariance matrix R to obtain eigenvalues λ1, λ2, …, λ p (where p is the number of features) and the corresponding eigenvectors a1, a2, …, a p , satisfying R = 1AP -1 , where P = [a1, a2, …, a p is the matrix composed of eigenvectors, and A = diaj(λ1, λ2, …, λ p ) is a diagonal matrix, and the elements on the diagonal are eigenvalues; Finally, calculate the variance contribution rate and cumulative variance contribution rate of the principal components, and calculate the variance contribution rate η of each principal component i , and the formula is as follows: The cumulative variance contribution rate is the sum of the variance contribution rates of the first num principal components, that is:

6. The method for predicting the failure life of the fluororubber seal of the vehicle range extender under the coupling of aging and fatigue according to claim 1, characterized in that In the step S6, the generative adversarial network includes a generator and a discriminator. The purpose of the generator is to generate data similar to the real data on the performance changes of fluororubber according to the input random noise, and the discriminator is used to judge whether the input data is real data or data generated by the generator; The loss function and optimizer of the generative adversarial network. The objective function of the generative adversarial network is as follows: Among them, G is the generator, D is the discriminator, x is the real data, z is the random noise vector, and the optimizer selects Adam; Then judge whether the data distribution is consistent with the original data distribution for the generated samples through statistical indicators.

7. The method for predicting the failure life of the fluororubber seal of the vehicle range extender under the coupling of aging and fatigue according to claim 1, characterized in that, In the step S7, after data augmentation is completed, the effectiveness of the augmented samples is strictly tested in combination with the previously established aging and fatigue coupling model to ensure that the augmented data can truly reflect the performance changes of fluororubber under actual working conditions; First, the augmented sample data is preprocessed to meet the input format and requirements of the aging and fatigue coupling model. The input of this model includes the working condition parameters of fluororubber, such as the change sequences of temperature, stress, and chemical medium concentration over time. Ensure that the time step, unit, etc. of these parameters in the augmented data are consistent with the model settings, and perform necessary normalization or format conversion operations on the data. Then, the preprocessed augmented sample data is input into the aging and fatigue coupling model group by group according to the time series; The augmented sample contains data for N time steps. For each time step t, the corresponding temperature T(t), stress σ(t), and chemical medium concentration C(t) parameters are input into the model; Subsequently, the model outputs include the change of the aging degree a(t) of fluororubber over time and the change of the fatigue crack propagation length a(t) over the number of cycles N. Compare the aging degree and fatigue crack propagation results output by the model when inputting the original data; Calculate the differences in key output indicators, and use the mean relative error to evaluate the output aging degree and fatigue crack propagation results. The mean relative error of the aging degree is: The mean relative error of the fatigue crack propagation length is: Among them, a augmented (t) is the aging degree of the enhanced data at time step t, a orjginal (t) is the aging degree of the input original data at time step t, a augmented (t) is the fatigue crack growth length of the enhanced data at time step t, a orjginal (t) is the fatigue crack growth length of the input original data at time step t; Finally, deeply analyze the change trends of the aging degree and fatigue crack propagation output by the model over time. Observe the growth curve of the aging degree over time and the change curve of the fatigue crack propagation length over the number of cycles when inputting the augmented data.

8. The method for predicting the failure life of the fluororubber seal of the vehicle range extender under the coupling of aging and fatigue according to claim 4, wherein, In the step S9, the time convolutional network is used to regard the fluororubber life prediction as a time series prediction problem affected by multiple factors. The time convolutional network includes a time convolutional network main body, a fully connected layer, and a dual-head output layer; The time convolutional network main body is stacked by residual blocks. Each residual block includes a one-dimensional causal convolutional layer, layer normalization, a ReLU activation function, and a Dropout layer; Fully connected layer and dual-head output layer: After flattening the TCN output features, they are connected to the fully connected layer. Two fully connected layers are set. The first layer contains 128 nodes, and the activation function is ReLU, which is used for preliminary feature fusion; the second layer has 64 nodes, and ReLU is also used to further refine the features; The dual-head output layer is set with two output branches. One of them outputs the Weibull distribution parameters (i.e., the shape parameter s and the scale parameter λ). To ensure that the output parameters conform to physical meanings (both are positive values), the Softplus activation function is applied to the two nodes of this branch respectively. The Softplus function is defined as follows: Softplus(x) = ln(1 + e x ) For the component of the fluororubber life prediction in the temporal convolutional network, its loss function consists of the loss of life value prediction. The mean squared error is used to measure the deviation between the predicted life value and the true life value. Let the predicted life value be the true life value be y, and N be the number of samples. Then the loss of life value prediction is as follows:

9. The method for predicting the failure life of the fluororubber seal of the vehicle range extender under the coupling of aging and fatigue according to claim 8, characterized in that, In the step S10, the time convolutional network outputs a life value T in step S9 pred , which is converted into the corresponding number of cycles N according to the loading frequency of the data used pred , that is, there is N pred = f×T pred , and the life value is related to the number of cycles in the fatigue crack growth equation by this method; In step S3, the relationship between the crack length a and the number of cycles N can be obtained by integrating the fatigue crack propagation equation, and its integration formula is: Substitute the converted number of cycles N pred into the above equation to calculate the theoretical crack length a(N pred ) at the predicted life; Integrating the aging equation in step S3 gives: Substitute the life value T output by the temporal convolutional network pred into this equation to calculate the theoretical aging degree a(T pred ).

10. The method for predicting the failure life of the fluororubber seal of the vehicle range extender under the coupling of aging and fatigue according to claim 9, characterized in that, In the step S11, based on the loss term of the time convolutional network in step S9 and the physical loss term in step S10, the total loss function formula of the physical information neural network at this time is: L total = L data + L phy To better guide the model to learn accurate quality estimates, reasonable weight settings are made for the various losses of the physics-informed neural network. The weight calculation formula is as follows: L total = a(t)L total + (1 - a(t))L phy Finally, the total loss function is used to guide the model training, and finally the Weibull distribution shape parameter s and scale parameter λ of the fluororubber, as well as a representative predicted life value, are output.

Images