A fatigue life simulation evaluation method for a new energy vehicle lightweight aluminum alloy material

Abstract

The application discloses a kind of fatigue life simulation evaluation methods of new energy vehicle light weight aluminum alloy material, it is related to material life evaluation technical field;Microstructure image is collected and coupled feature is extracted by machine learning, complete heterogeneous data fusion and enhancement;Based on GAN generation topological optimization structure, in stress gradient dynamic grid, introduce VPSC model to describe anisotropy, construct dynamic finite element model;Fusion vehicle driving data, to LSTM prediction load, VMD decomposition and environmental correction, realize space-time correlation load spectrum reconstruction;Microscopically with phase-field model, macroscopically with XFEM crack tracking, bidirectional coupling transmission damage parameter, carry out multi-physical field coupling simulation;Bayesian inference is used to fuse simulation and test data, calculate life probability distribution, error is over limit when correcting parameter;The application ultimately reduces fatigue life prediction error, reduces the time consumption of single simulation, realizes life cycle assessment and visual early warning, provides efficient scheme for light weight design, promotes industrial technology upgrading.

Description

Technical Field

[0001] This invention relates to the field of material life assessment technology, and in particular to a fatigue life simulation assessment method for lightweight aluminum alloy materials for new energy vehicles. Background Technology

[0002] Against the backdrop of the rapid development of the new energy vehicle industry, lightweight design has become a key approach to improving vehicle range and energy efficiency. Aluminum alloys, due to their low density and high specific strength, are widely used in core components such as automobile bodies, chassis, and battery trays. However, fatigue failure of aluminum alloys under complex alternating load conditions seriously threatens the safety and reliability of vehicles. Traditional fatigue life assessment relies on physical testing, which is time-consuming and costly, and it is difficult to simulate the multi-condition coupled environment in actual service. Although finite element simulation can shorten the development cycle, existing methods have shortcomings in material microstructure characterization, multi-physics field coupling simulation, and dynamic load prediction, resulting in large deviations between simulation results and actual lifespan, failing to meet engineering design requirements.

[0003] As new energy vehicles develop towards higher integration and intelligence, the structural complexity and service environment severity of aluminum alloy components are constantly increasing. On the one hand, the influence mechanism of factors such as grain orientation and second-phase particle distribution of aluminum alloys on fatigue performance at the microscale has not been fully incorporated into simulation models. On the other hand, the synergistic effects of temperature, humidity, corrosive media, and mechanical loads during vehicle operation, as well as the dynamic load changes caused by different driving behaviors and road conditions, all exceed the processing capabilities of traditional simulation methods. Furthermore, existing technologies lack a systematic assessment of the evolution of material performance throughout its entire life cycle, making it difficult to support sustainable design goals under a circular economy model.

[0004] At the simulation technology level, traditional finite element model mesh generation and parameter setting largely rely on manual experience, resulting in low computational efficiency and susceptibility to subjective factors. The single load spectrum equivalent method cannot accurately reflect the spatiotemporal characteristics of actual working conditions. Furthermore, multiphysics coupling simulations often consume excessive computational resources, making it difficult to achieve refined simulations across scales. These technical bottlenecks lead to repeated trial and error in the design process of aluminum alloy components for new energy vehicles, not only extending the R&D cycle but also increasing production costs. Therefore, innovative fatigue life simulation and evaluation methods are urgently needed to achieve technological breakthroughs. Summary of the Invention

[0005] This invention proposes a fatigue life simulation evaluation method for lightweight aluminum alloy materials for new energy vehicles to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a fatigue life simulation evaluation method for lightweight aluminum alloy materials for new energy vehicles, comprising:

[0007] Multi-source heterogeneous data fusion and feature enhancement steps: Field emission scanning electron microscopy (FE-SEM) is used to acquire images of the second-phase particle distribution of aluminum alloy, and electron backscatter diffraction (EBSD) technology is used to obtain grain orientation and grain boundary characteristics; Digital image processing (DIC) technology is used to monitor the surface strain field in real time during tensile tests; Convolutional neural network is used to automatically extract the coupling features of microstructure and mechanical response to construct material feature vectors;

[0008] The steps for constructing a dynamic adaptive finite element model are as follows: A candidate topology for aluminum alloy components is generated based on a Genetic Algorithm (GAN), and optimized using a genetic algorithm in conjunction with fatigue life and lightweight objectives; an adaptive mesh refinement technique is employed, dynamically adjusting the mesh based on stress gradient and crack propagation prediction; and a viscoplastic self-consistent VPSC model is introduced to describe the anisotropic plastic behavior of the aluminum alloy, improving the accuracy of stress calculation under loading.

[0009] The load spectrum reconstruction steps based on spatiotemporal correlation are as follows: 1. Integrate vehicle GPS trajectory, road roughness data and driving behavior patterns to establish a three-dimensional spatiotemporal load prediction model; 2. Analyze historical load data through a long short-term memory network (LSTM) to predict the dynamic load distribution of the next 500 kilometers of travel; 3. Design a variational mode decomposition (VMD) algorithm to decompose the measured load into sub-signals, and reconstruct the composite loading sequence with physical meaning by combining the fatigue damage equivalence principle.

[0010] The steps of cross-scale multiphysics coupling simulation are as follows: A cross-scale simulation framework is constructed. At the microscale, a phase-field model is used to simulate the interaction between dislocations and second-phase particles. At the macroscale, the extended finite element method (XFEM) is used to track crack propagation. Stress and damage transfer between the micro and macroscales is achieved through a two-way coupling algorithm. An electrochemical corrosion model coupling humidity, temperature, and stress is introduced, with the following formula: Where D is the degree of corrosion damage and t is time. Let R be the stress level, R be the gas constant, and T be the absolute temperature. Chloride ion concentration, The material-related parameters are used to quantify the accelerating effect of environmental factors on fatigue damage.

[0011] Furthermore, it also includes:

[0012] The confidence-enhanced life prediction and verification steps are as follows: a Bayesian inference framework is used to fuse simulation results and experimental data, and the posterior distribution of model parameters is dynamically updated; an uncertainty propagation network including material batches and manufacturing process factors is constructed to calculate the probability density function of fatigue life; when the confidence interval width of the prediction result exceeds the threshold, sensitivity analysis of parameters and model correction are automatically triggered.

[0013] The steps for co-optimizing microstructure and fatigue performance are as follows: Establish a response surface model of microstructure parameters and fatigue life; search for the optimal microstructure control scheme using a particle swarm optimization algorithm, and verify manufacturability through simulation of hot working processes; introduce a microstructure evolution dynamics model, with the following formula: ,in Let t be the average grain size, A be the material constant, Q be the activation energy, R be the gas constant, T be the absolute temperature, and n be the grain growth index. This method predicts the impact of microstructure evolution on fatigue performance during heat treatment, thereby achieving a balance between material properties and processing costs.

[0014] Multi-lifecycle scenario fusion assessment steps: Construct a full lifecycle simulation platform including design, manufacturing, service, and recycling; simulate the generation and release process of residual stress during the manufacturing stage; superimpose the impact of different maintenance strategies on fatigue life during the service stage; use digital twin technology to update vehicle operation data in real time and dynamically correct the simulation model; evaluate the deterioration effect of material reprocessing on fatigue performance during the recycling stage, providing a basis for circular economy design.

[0015] Furthermore, in the multi-source heterogeneous data fusion and feature enhancement step, a self-supervised contrastive learning algorithm is employed. A self-supervised learning framework based on Siamese neural networks is designed. Positive and negative sample pairs are constructed through data enhancement operations such as rotation and scaling of the material image; the objective function is optimized. Automatically extract high-discrimination microstructural features, among which For positive sample pairs feature vectors, Let be the negative sample feature vector, and sim be the cosine similarity calculation function. This refers to the temperature parameter.

[0016] Furthermore, in the dynamic adaptive finite element model construction step, a physics-guided neural network (PINN) is used to embed the mechanical equilibrium equations and constitutive physical laws into the neural network loss function, as shown in the formula: ,in For data fitting loss, For physical constraint loss, The equilibrium coefficient is used; the stress distribution in the geometric region is predicted using PINN, replacing the local mesh refinement of traditional finite element methods; and the model boundary conditions are automatically corrected by minimizing the residuals.

[0017] Furthermore, in the load spectrum reconstruction step based on spatiotemporal correlation, a spatiotemporal graph convolutional network STGCN is used to abstract the road network into a graph structure, where nodes represent geographical locations and edges represent load transmission relationships; the spatiotemporal convolutional layer captures the propagation patterns of loads in the spatial and temporal dimensions; and an attention mechanism is introduced to dynamically allocate load weights for different road segments and time periods to predict the load sequence for the next 1000-kilometer journey.

[0018] Furthermore, in the cross-scale multiphysics coupling simulation step, a multi-scale reduced-order model MSROM is used to establish a reduced-order model at the microscale to replace the computationally expensive phase-field simulation; cluster analysis is used to divide similar regions of the macrostructure, and the microscale reduced-order model is shared for similar regions; and an error compensation algorithm is used to quantify the information transfer loss between scales.

[0019] Furthermore, in the confidence-enhanced lifetime prediction and verification step, a reinforcement learning optimization strategy is employed to model the model correction process as a Markov decision process, obtaining rewards by trying different parameter adjustment strategies; the reward function is defined as follows: ,in This represents the change in lifetime prediction error. To calculate the change in cost, The cost weight coefficient is used; the optimal correction strategy is learned through the Q-learning algorithm.

[0020] Furthermore, prior to the multi-source heterogeneous data fusion and feature enhancement steps, a dynamic data quality assessment step is also included: constructing an assessment index system that includes data integrity, consistency, and timeliness; using DS evidence theory to integrate expert experience and historical data to calculate the credibility weight of each data source; and automatically triggering data cleaning or supplementary collection processes when the credibility of a data source is lower than a threshold.

[0021] Furthermore, following the cross-scale multiphysics coupling simulation step, a fatigue failure mode visualization and early warning step is also included. A virtual reality-based visualization platform is developed to dynamically display the microcrack initiation and macrocrack propagation process in three dimensions; risk areas are highlighted through color coding and transparency adjustment; a fatigue damage threshold is set to trigger an early warning mechanism, and a visualization report containing failure modes, remaining life, and improvement suggestions is automatically generated when the simulation results approach the critical state.

[0022] Compared with existing technologies, the beneficial effects of this invention are:

[0023] In terms of accuracy, by using multi-source heterogeneous data fusion and cross-scale multi-physics coupling technology, the influence of microstructure characteristics and multi-condition interaction on fatigue performance is fully captured. Compared with traditional methods, the fatigue life prediction error is reduced by 67.9%, effectively avoiding the risk of structural failure caused by assessment bias.

[0024] In terms of efficiency, the application of technologies such as dynamic adaptive finite element model construction and physically guided neural networks significantly reduces computation time, shortening the time for a single finite element simulation from 48 hours to 15 hours, and improving feature extraction efficiency by 92%, thus significantly accelerating product development. Meanwhile, load spectrum reconstruction and model optimization strategies based on machine learning automatically process complex dynamic load data, reducing the number of model correction iterations by 63.3% and lowering the cost of manual intervention.

[0025] Furthermore, this method innovatively integrates full life-cycle scenarios with visual early warning functions. It can not only simulate the performance evolution of materials from manufacturing to recycling, providing a basis for circular economy design, but also intuitively present the fatigue failure process through VR technology, providing early warnings of potential risks. This technological system provides a high-precision, high-efficiency, and intelligent solution for lightweight design of new energy vehicles, which is of great significance for promoting industrial technology upgrades and ensuring driving safety. Attached Figure Description

[0026] Figure 1 This is a schematic block diagram of a fatigue life simulation evaluation method for lightweight aluminum alloy materials for new energy vehicles proposed in this invention.

[0027] Figure 2 This is a schematic diagram comparing the fatigue life prediction errors of different methods for fatigue life simulation evaluation of lightweight aluminum alloy materials for new energy vehicles proposed in this invention.

[0028] Figure 3 This diagram illustrates the variation of finite element calculation time with model complexity for a fatigue life simulation evaluation method for lightweight aluminum alloy materials used in new energy vehicles, as proposed in this invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0031] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0032] Reference Figures 1 to 3 : Specific Implementation Method of Fatigue Life Simulation Evaluation Method for Lightweight Aluminum Alloy Materials in New Energy Vehicles

[0033] Taking the fatigue life simulation evaluation of an aluminum alloy battery tray for a new energy vehicle as an example, the implementation process of this patented method is explained in detail.

[0034] 1.1 Steps for Multi-Source Heterogeneous Data Fusion and Feature Enhancement

[0035] Microstructure images of aluminum alloy samples were acquired using a FEITalos F200X field emission scanning electron microscope with a resolution of 3 nm, obtaining data such as the size and distribution density of second-phase particles. Simultaneously, EBSD technology was used to scan the samples at 0.5 μm steps to obtain grain orientation distribution maps and grain boundary angle information. In the mechanical property testing phase, tensile tests were performed using an Instron 5982 electronic universal testing machine equipped with a DIC system, recording surface strain field changes at a frame rate of 500 fps.

[0036] The above data, along with the material composition analysis results (determined by an EDX spectrometer), were imported into a self-developed data fusion platform. A convolutional neural network (CNN) was built using the PyTorch framework, with a network structure containing 5 convolutional layers, 3 pooling layers, and 2 fully connected layers. A training dataset was constructed by performing data augmentation operations such as rotation and scaling on 1000 sets of material images. During training, the stochastic gradient descent (SGD) algorithm was used to optimize the network parameters, with a learning rate set to 0.001 and 100 iterations. Finally, a material feature vector with 326 dimensions, including grain size distribution, second-phase particle spacing, and strain concentration coefficient, was extracted.

[0037] 1.2 Steps for Constructing a Dynamic Adaptive Finite Element Model

[0038] A 3D geometric model of the battery tray was constructed using SolidWorks software, accurately reproducing structural details such as reinforcing ribs and mounting holes. The model was imported into ANSYS Workbench, and adaptive mesh generation was enabled, with the initial mesh size set to 2mm. A Generative Adversarial Network (GAN) model was invoked using a Python script. This GAN consists of a generator with 6 layers of transposed convolutions and a discriminator with 5 layers of convolutions. Optimization objectives included lightweight design and fatigue life, generating 50 candidate topology optimization structures through 1000 iterations.

[0039] A multi-objective genetic algorithm (NSGA-II) was used to screen candidate structures, with a crossover probability of 0.8, a mutation probability of 0.2, and a population size of 100. The optimal structure was obtained after 20 generations of evolution. During the simulation, stress distribution was monitored in real time. When the local stress gradient exceeded 1.5 times the average stress, mesh refinement was automatically triggered, refining the mesh in critical areas to 50 μm. Simultaneously, the VPSC model was used to describe the anisotropy of the aluminum alloy. Model parameters were set based on grain orientation data measured by EBSD to complete the finite element model construction.

[0040] 1.3 Load spectrum reconstruction steps based on spatiotemporal correlation

[0041] Driving data for 3000 kilometers was continuously collected using a GPS positioning module (accuracy ±1m), an accelerometer (range ±50g, resolution 0.001g), and a road roughness detector installed on the test vehicle. The raw data was cleaned using Python's pandas library to remove outliers. The processed data was then fed into an LSTM network built using TensorFlow. The network contains two hidden layers, each with 128 neurons. During training, the Adam optimizer was used with a learning rate of 0.0005 and 50 iterations to predict the load distribution for the next 500 kilometers of travel.

[0042] A variational mode decomposition (VMD) program was written using MATLAB to decompose the measured load signal into eight sub-signals. Based on Miner's linear fatigue cumulative damage theory and combined with the material's SN curve (obtained through fatigue testing), the decomposed sub-signals were equivalent to a standard sinusoidal loading sequence. Simultaneously, an environmental factor correction model was established, where temperature data was collected by a K-type thermocouple mounted on the battery tray surface, and humidity data was acquired by a HIH-4000 humidity sensor. A neural network model was trained using historical experimental data to obtain the correction coefficient calculation formula K = 0.98 + 0.002T - 0.001RH, where T is temperature (°C) and RH is relative humidity (%), which corrects for the equivalent load.

[0043] 1.4 Steps for Multi-Scale Multiphysics Coupling Simulation

[0044] In the microscale simulation, the open-source phase-field simulation software Marmot was used, with a computational domain of 10 μm × 10 μm × 10 μm and a mesh size of 0.05 μm. Dislocation motion parameters and second-phase particle properties were set based on material characteristic data to simulate the interaction between dislocations and second-phase particles. At the macroscale, extended finite element analysis was performed using ANSYS Mechanical APDL, with an initial crack length of 0.1 mm and a direction perpendicular to the maximum principal stress direction.

[0045] A Python interface program was developed to achieve bidirectional coupling between micro and macro scales. Every 10 macro loading steps, macroscopic stress was transferred to the microscopic model as boundary conditions, while damage parameters obtained from the microscopic simulation were mapped back to the macroscopic model. An electrochemical corrosion model coupling humidity, temperature, and stress was introduced. ,in Ea = 80000 J / mol, R = 8.314 J / (mol·K). Chloride ion concentration. The environmental simulation test setting was 0.01 mol / L, and the effect of corrosion on fatigue damage was calculated in real time.

[0046] 1.5 Steps for Confidence Enhancement-Based Lifetime Prediction and Validation

[0047] A Bayesian inference model was constructed using the PyMC3 library. The simulated fatigue life prediction was used as the prior distribution, and the posterior distribution was updated by combining it with fatigue test data from three sets of identical conditions (test lives of 120,000, 135,000, and 118,000 cycles, respectively). Markov chain Monte Carlo (MCMC) sampling was set up with 4 chains, 2000 sampling steps, and 500 combustion steps. The probability density function of fatigue life was calculated, with a 95% confidence interval of [112,000, 140,000] cycles.

[0048] Sensitivity analysis is automatically triggered when the confidence interval width exceeds 20,000 cycles. The Sobol method is used to calculate the influence coefficients of each input parameter (such as material elastic modulus, load amplitude, ambient temperature, etc.) on fatigue life, identifying load amplitude and material yield strength as key parameters. The model parameters are then corrected by returning to the corresponding steps, and the simulation is re-performed until the prediction error is below 10%.

[0049] 2. Microstructure-fatigue performance synergistic optimization steps

[0050] A response surface model of microstructure parameters and fatigue life was established using Design-Expert software, and a center composite design (CCD) test with 50 test points was designed. Hot working process simulation was performed using Deform-3D software, with a forging temperature of 450℃ and a strain rate of 0.1 s⁻ 1 To simulate the manufacturability of different micro-organism control schemes.

[0051] Introducing a dynamic model of micro-organism evolution ,in Given Q = 150,000 J / mol and n = 2, the optimal microstructure control scheme was searched using the Particle Swarm Optimization (PSO) algorithm. The particle swarm size was set to 30, the number of iterations to 50, and the inertia weight was linearly decreased from 0.9 to 0.4. The final optimal scheme was a grain size of 15 μm and a second phase volume fraction of 8%, resulting in a 28% improvement in fatigue life compared to the initial scheme.

[0052] 3. Multi-lifecycle scenario fusion evaluation steps

[0053] A full life-cycle simulation platform for battery trays was built using Simcenter3D software. During the manufacturing phase, the temperature and stress fields during die-casting were simulated to predict the magnitude and distribution of residual stress. During the service phase, actual driving load spectra and different maintenance strategies (such as periodic bolt tightening and surface anti-corrosion treatment) were imported to assess their impact on fatigue life. Digital twin technology was employed, using real-time vibration and temperature data collected from onboard sensors to update the simulation model hourly.

[0054] During the recycling phase, the effects of material remelting and casting processes on microstructure and fatigue properties were simulated. The remelting temperature was set at 700℃, and the holding time was 2 hours. Phase field simulation was used to predict grain growth and the degree of second-phase coarsening, assessing the deteriorating effect of reprocessing on fatigue properties and providing data support for optimizing the material recycling process.

[0055] 4. Implementation of Self-Supervised Comparative Learning Algorithm

[0056] A Siamese neural network was built using PyTorch. The network structure includes two convolutional layers (3×3 kernels, 64 channels), two pooling layers (2×2 max pooling), and two fully connected layers (128 neurons). Data augmentation operations, including random rotation (±30°), scaling (0.8-1.2x), and adding Gaussian noise (mean 0, variance 0.05), were performed on 2000 acquired microscopic tissue images to construct positive and negative sample pairs.

[0057] Optimize objective function Among them, temperature parameters Cosine similarity calculation function The Adam optimizer was used with a learning rate of 0.0003 for 50 training epochs. The trained network can automatically extract highly discriminative microstructural features, achieving a 92% efficiency improvement in feature extraction compared to traditional manual annotation methods.

[0058] 5. Implementation of Physically Guided Neural Network (PINN)

[0059] PINN is built using the TensorFlow framework. The network contains four hidden layers, each with 64 neurons, and uses tanh as the activation function. The balance equation... ( (where b is the stress tensor and b is the body force) and constitutive relations (C is the elasticity matrix, (For strain tensor) Embedded loss function ,in =0.5.

[0060] In complex geometric regions (such as near battery tray mounting holes), PINN predicts stress distribution, replacing the local mesh refinement of traditional finite element methods. It automatically corrects model boundary conditions by minimizing residuals (the degree to which equilibrium equations and constitutive relations are not satisfied). Testing shows that, under the same accuracy requirements, computational efficiency is 3.2 times higher than traditional finite element methods.

[0061] 6. Implementation of Spatiotemporal Graph Convolutional Network (STGCN)

[0062] The test vehicle's driving route was abstracted as a graph structure containing 100 nodes, where nodes represent key road locations, and edge weights are determined based on road connectivity and historical load transfer data. An STGCN network was built using the PyTorchGeometric library, comprising two spatiotemporal convolutional layers and one fully connected layer. The spatiotemporal convolutional layers use 1D convolutional kernels to capture temporal features, and graph convolution operations to capture spatial features.

[0063] By training the network, the propagation patterns of loads in spatial and temporal dimensions are learned. An attention mechanism is introduced to dynamically allocate load weights to different road segments and time periods based on road type (urban roads, highways, etc.) and driving behavior (rapid acceleration, constant speed driving, etc.). In predicting load sequences for a future 1000-kilometer journey, the average error is reduced to 4.8%.

[0064] 7. Implementation of Multiscale Reduced-Order Model (MSROM)

[0065] At the microscale, a Kriging surrogate model is used to replace phase-field simulation. A high-fidelity phase-field simulation is performed using 20 sample points selected through Latin hypercube sampling to construct the Kriging model.

[0066] In the macroscopic structural analysis, the K-means clustering algorithm was used to divide the battery tray into 10 similar regions, and the microscopic Kriging model was shared for similar regions. An error compensation algorithm was developed to calculate the information transfer loss between scales by comparing the microscopic high-fidelity simulation results and the prediction results of the reduced-order model for local regions. In the overall simulation process, the efficiency was improved by 82% compared with the full-scale high-fidelity simulation, while the fatigue life prediction error was controlled within 9.5%.

[0067] 8. Strengthen the implementation of learning optimization strategies

[0068] The model correction process is modeled as a Markov decision process. The state space includes 10 dimensions such as material parameters, load conditions, and simulation results, while the action space includes 8 parameter adjustment strategies such as adjusting the material elastic modulus and correcting the load amplitude. A reward function is defined. Among them, the cost weighting coefficient =0.1, This represents the change in lifetime prediction error. To calculate the change in cost (measured by simulation time).

[0069] The Q-learning algorithm was used for policy learning, with a learning rate of 0.1, a discount factor of 0.9, and an exploration rate that linearly decayed from 0.9 to 0.1. After 500 iterations, the optimal modified policy was obtained, reducing the number of model iterations by 63% compared to the random adjustment policy.

[0070] 9. Steps for Dynamic Data Quality Assessment

[0071] A three-dimensional evaluation index system was constructed, comprising data integrity (proportion of missing values), consistency (degree of inconsistency between different data sources), and timeliness (data update interval). The DS evidence theory was used to integrate the experience and historical data of three materials experts to calculate the credibility weight of each data source (SEM images, mechanical test data, etc.).

[0072] When the reliability of a data source falls below 0.6, the data cleaning process is automatically triggered. Missing data is handled using multiple imputation techniques; contradictory data is corrected through cross-validation and expert review. This ensures the reliability of the input data, providing a guarantee for subsequent simulation analysis.

[0073] 10. Steps for Visualizing and Early Warning of Fatigue Failure Modes

[0074] A virtual reality visualization system was developed based on the Unity3D platform to perform 3D modeling and rendering of the micro-crack initiation and macro-crack propagation processes. Color coding was used to represent stress levels (red for high stress areas, blue for low stress areas), and transparency adjustments were made to highlight potential crack propagation paths.

[0075] The fatigue damage threshold is set to 0.8 (calculated based on Miner's theory). When the cumulative damage in the simulation results approaches this threshold, the system automatically triggers an early warning mechanism. A visual report is generated that includes failure modes (such as structural fracture caused by crack propagation), remaining life (estimated 12,000 cycles), and improvement suggestions (adding local reinforcement structures), presented to designers in the form of 3D animation and text descriptions.

[0076] Performance evaluation data

[0077] Evaluation indicators Traditional methods This application method Increase ratio Feature extraction efficiency Manual labeling takes 20 hours per sample. Automatic extraction takes 1.6 hours per sample. 92% Finite element calculation efficiency A single simulation takes 48 hours. A single simulation takes 15 hours. 68.75% Load spectrum prediction error Average error 18% Average error 4.8% 73.3% Fatigue life prediction error Average error 28% Average error 9% 67.9% Model correction iteration count An average of 15 times 5.5 times on average 63.3%

[0078] Data Interpretation:

[0079] Regarding feature extraction efficiency, traditional manual annotation methods are not only time-consuming and labor-intensive but also easily affected by subjective factors. This application employs a self-supervised contrastive learning algorithm, which, through automatic learning from a large number of material images, can quickly and accurately extract microstructure features, significantly improving efficiency. The improved finite element computation efficiency is attributed to the physics-guided neural network and the multi-scale order reduction model. The former reduces the computational load for local mesh refinement, while the latter, through the order reduction processing of the microscopic model, greatly shortens the simulation time.

[0080] The reduction in load spectrum prediction error is primarily attributed to the spatiotemporal graph convolutional network, which effectively captures the propagation patterns of loads in the spatiotemporal dimension, making it more closely aligned with actual working conditions compared to traditional methods. The improvement in fatigue life prediction error is the result of a multi-step collaborative process, from multi-source data fusion and cross-scale simulation to confidence-enhancing prediction methods, comprehensively improving prediction accuracy. The reduction in model correction iterations is due to the reinforcement learning optimization strategy, which intelligently selects the optimal parameter adjustment scheme, avoiding blind iteration and improving model optimization efficiency. These data fully demonstrate the significant advantages of the proposed method in fatigue life simulation and evaluation of aluminum alloy materials for new energy vehicles.

[0081] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A fatigue life simulation evaluation method for a lightweight aluminum alloy material for new energy vehicles, characterized by, The method comprises the following steps: Multi-source heterogeneous data fusion and feature enhancement step: collect aluminum alloy second phase particle distribution images using field emission scanning electron microscope FE-SEM, and obtain grain orientation and grain boundary characteristics by combining electron backscatter diffraction EBSD technology; Real-time monitoring of surface strain field in tensile test by digital image DIC technology; automatic extraction of coupling characteristics of microstructure and mechanical response by convolutional neural network, and construction of material feature vector; Dynamic adaptive finite element model construction step: generate topological optimization candidate structure of aluminum alloy component based on generative adversarial network GAN, and combine fatigue life and lightweight target to carry out genetic algorithm optimization; adopt adaptive grid refinement technology, and dynamically adjust the grid according to the stress gradient and crack propagation prediction; introduce visco-plastic self-consistent VPSC model to describe the anisotropic plastic behavior of aluminum alloy, and improve the stress calculation accuracy under loading; Load spectrum reconstruction step based on space-time correlation: fuse vehicle GPS trajectory, road roughness data and driving behavior mode to establish a three-dimensional space-time load prediction model; analyze historical load data by long short-term memory network LSTM to predict the dynamic load distribution of the next 500 kilometers of travel; design a variational mode decomposition VMD algorithm to decompose the measured load into sub-signals, and reconstruct the composite loading sequence with physical meaning according to the fatigue damage equivalence principle; Cross-scale multi-physical field coupling simulation step: build a cross-scale simulation framework, simulate the interaction of dislocation-second phase particles at the microscale by using the phase field model, and track crack propagation at the macroscale based on the extended finite element method XFEM; The stress and damage transfer between micro and macro scales is realized by the two-way coupling algorithm. The electrochemical corrosion model coupled with humidity-temperature-stress is introduced, which is given by where D is the corrosion damage degree, t is the time, is the stress level, R is the gas constant, T is the absolute temperature, is the chloride ion concentration, is the material related parameter, which quantifies the acceleration effect of environmental factors on fatigue damage. 2.The fatigue life simulation evaluation method of a new energy vehicle lightweight aluminum alloy material according to claim 1, characterized in that, The method further comprises the following steps: Life prediction and verification step based on confidence enhancement: adopt Bayesian inference framework to fuse simulation results and test data, and dynamically update the posterior distribution of model parameters; build an uncertainty propagation network including material batch and manufacturing process factors, and calculate the probability density function of fatigue life; when the confidence interval width of the prediction result exceeds the threshold, automatically trigger the sensitivity analysis and model correction of the parameters; Microstructure-fatigue performance synergistic optimization step: a response surface model of microstructure parameters and fatigue life is established; the optimal microstructure control scheme is searched by a particle swarm optimization algorithm, and the manufacturability is verified by combining a hot working process simulation; a microstructure evolution dynamics model is introduced, and the formula is The influence of microstructure evolution during heat treatment on fatigue performance is predicted, and the balance optimization of material performance and processing cost is realized, wherein is an average grain size, t is time, A is a material constant, Q is an activation energy, R is a gas constant, T is an absolute temperature, and n is a grain growth index. 3.The fatigue life simulation evaluation method of a new energy vehicle lightweight aluminum alloy material according to claim 1, characterized in that, The method further comprises the following steps: Multi-life cycle scenario fusion evaluation step: build a whole life cycle simulation platform including design, manufacturing, service and recycling; Simulate the generation and release process of residual stress in the manufacturing stage, and superimpose the influence of different maintenance strategies on fatigue life in the service stage; adopt digital twin technology to update the vehicle operation data in real time, and dynamically correct the simulation model; In the recycling stage, evaluate the deterioration effect of material reprocessing on fatigue performance, and provide a basis for circular economy design. 4.The fatigue life simulation and evaluation method of a new energy vehicle lightweight aluminum alloy material according to claim 1, characterized in that, In the multi-source heterogeneous data fusion and feature enhancement step, a self-supervised learning framework based on a twin neural network is designed by using a self-supervised contrast learning algorithm, positive and negative sample pairs are constructed through rotation and scaling data enhancement operations on the material image, and an objective function is optimized automatically extract high-differentiation microstructure features, wherein is a feature vector of a positive sample pair, is a feature vector of a negative sample, and sim is a cosine similarity calculation function, is a temperature parameter. 5.The fatigue life simulation and evaluation method of a new energy vehicle lightweight aluminum alloy material according to claim 1, characterized in that, In the dynamic adaptive finite element model construction step, a physically guided neural network PINN is used, mechanical equilibrium equations and physical laws of constitutive relationship are embedded into a neural network loss function, and the formula is wherein is a data fitting loss, is a physical constraint loss, is a balance coefficient; the stress distribution of the geometric region is predicted by using the PINN, and the local grid encryption of the traditional finite element is replaced; the model boundary conditions are automatically corrected through residual minimization. 6.The fatigue life simulation and evaluation method of a new energy vehicle lightweight aluminum alloy material according to claim 1, characterized in that, In the load spectrum reconstruction step based on space-time correlation, a space-time graph convolution network STGCN is adopted to abstract the road network as a graph structure, the nodes represent geographical positions, and the edges represent load transmission relationships; Through the space-time convolution layer, the propagation law of load in the space and time dimensions is captured; the attention mechanism is introduced to dynamically allocate the load weights of different road sections and time periods, and the load sequence of the next 1000 kilometers of travel is predicted. 7.The fatigue life simulation and evaluation method of a new energy vehicle lightweight aluminum alloy material according to claim 1, characterized in that, In the cross-scale multi-physical field coupling simulation step, a multi-scale reduced order model MSROM is used to establish a reduced order model at a micro scale to replace a phase field simulation with a high calculation cost; similar areas of a macro structure are divided by cluster analysis, and micro reduced order models are shared by the same type of area; and an error compensation algorithm is used to quantify the loss of information transmission between scales. 8.The fatigue life simulation and evaluation method of a new energy vehicle lightweight aluminum alloy material according to claim 2, characterized in that, In the life prediction and verification step based on confidence enhancement, a reinforcement learning optimization strategy is adopted, a model correction process is modeled as a Markov decision process, and a reward is obtained by trying different parameter adjustment strategies; a reward function is defined as wherein is a life prediction error change amount, is a calculation cost change amount, is a cost weight coefficient; and an optimal correction strategy is learned through a Q-learning algorithm. 9.The fatigue life simulation and evaluation method of a new energy vehicle lightweight aluminum alloy material according to claim 1, characterized in that, Before the multi-source heterogeneous data fusion and feature enhancement step, a data quality dynamic evaluation step is further included: an evaluation index system including data integrity, consistency and timeliness is constructed; D-S evidence theory is used to fuse expert experience and historical data to calculate the credibility weight of each data source; When the data source credibility is lower than the threshold, the data cleaning or supplementary acquisition process is automatically triggered. 10.The fatigue life simulation and evaluation method of a new energy vehicle lightweight aluminum alloy material according to claim 1, characterized in that, After the cross-scale multi-physical field coupling simulation step, a fatigue failure mode visualization early warning step is further included, a virtual reality-based visualization platform is developed, and the micro crack initiation and macro crack propagation process is dynamically displayed in three dimensions; Risk areas are highlighted through color coding and transparency adjustment; A fatigue damage threshold is set to trigger an early warning mechanism, and when the simulation result approaches a critical state, a visualization report including the failure mode, remaining life and improvement suggestions is automatically generated.

Images