A method for fatigue analysis of automobile leaf spring based on simulation model optimization

Abstract

The application relates to the technical field of simulation analysis, in particular to a vehicle leaf spring fatigue analysis method based on simulation model optimization, which comprises the following steps: sampling a multi-channel load spectrum under a test working condition to obtain a load sequence; performing mesh division on a leaf spring model to obtain a plurality of mesh nodes; applying a load to the leaf spring model according to the load sequence, and performing finite element analysis to obtain a stress sequence of each mesh node under the test working condition; using a rain flow counting method to statistically count the cumulative fatigue damage and damage saturation of each mesh node on the stress sequence in real time; and constructing a damage cloud map according to the cumulative fatigue damage of each mesh node and a second-order difference sequence of the stress sequence at the termination moment of the load sequence. The technical scheme can improve the efficiency and precision of vehicle leaf spring fatigue analysis.

Description

Technical Field

[0001] This application relates to the field of simulation analysis technology, and in particular to a fatigue analysis method for automotive leaf springs based on simulation model optimization. Background Technology

[0002] As the automotive industry moves towards lightweighting and longer lifespans, the structural reliability of leaf springs, a core load-bearing component of the suspension system, directly impacts vehicle safety. During the research and development phase, road tests are typically used to evaluate the fatigue performance of leaf springs. However, road testing is time-consuming, costly, and difficult to reproduce extreme random loads. Therefore, using finite element method (FEM) simulation for leaf spring fatigue analysis has become the mainstream approach in the industry.

[0003] Currently, the common fatigue analysis methods for leaf springs in the industry mostly use a uniform time step to sample multi-channel load spectra and use finite element analysis technology to obtain the stress sequence of each grid node under the multi-channel load spectrum. In terms of fatigue damage statistics, the stress sequence of all grid nodes is calculated in the entire time domain based on the rainflow counting method. Finally, the stress cycle of all grid nodes in the model is extracted on the entire time axis by the rainflow counting method, and the cumulative damage is calculated by referring to the SN curve of the material, thereby generating the damage distribution results of the entire field.

[0004] However, fixed-step load sampling cannot balance computational efficiency and transient impact capture, easily leading to the loss of key fatigue extreme points and generating a large amount of data redundancy. Furthermore, the aforementioned method only performs linear cumulative damage superposition based on the material's SN curve, ignoring the damage caused by stress impact to automotive leaf springs, resulting in low efficiency and accuracy in automotive leaf spring fatigue analysis. Summary of the Invention

[0005] To address the technical problem of low efficiency and accuracy in automotive leaf spring fatigue analysis, this application provides a simulation model-based fatigue analysis method for automotive leaf springs, which can improve the efficiency and accuracy of automotive leaf spring fatigue analysis.

[0006] This application provides a fatigue analysis method for automotive leaf springs based on simulation model optimization. The analysis method includes: sampling a multi-channel load spectrum under test conditions to obtain a load sequence; meshing the leaf spring model to obtain multiple mesh nodes; applying loads to the leaf spring model according to the load sequence and performing finite element analysis to obtain the stress sequence of each mesh node under test conditions; using the rainflow counting method to statistically analyze the cumulative fatigue damage and damage saturation of each mesh node in real time on the stress sequence, including: setting the corresponding mesh node as a dormant node and pausing the rainflow counting method calculation when the growth rate of damage saturation is less than a first threshold; monitoring the load change in the load sequence, and activating each dormant node to execute the rainflow counting method again when the load change is greater than a second threshold; and constructing a damage cloud map based on the cumulative fatigue damage of each mesh node and the second-order difference sequence of the stress sequence at the end of the load sequence.

[0007] By determining the dormant state of grid nodes through damage saturation and triggering the activation mechanism of grid nodes based on load changes, dynamic allocation of computing resources is achieved. This eliminates redundant calculations in a large number of non-critical areas and uses the second-order difference sequence of stress sequence to correct cumulative fatigue damage, preventing the omission of potential fatigue weak points caused by sudden load changes and ensuring the accuracy of fatigue life assessment.

[0008] Preferably, the step of sampling the multi-channel load spectrum under the test condition to obtain the load sequence includes: obtaining the load change at each moment in the multi-channel load spectrum and calculating the average load change; comparing the load change at any moment with the average load change to divide the multi-channel load spectrum into a stable condition segment and a sudden change condition segment; sampling with a first step length in the stable condition segment and sampling with a second step length in the sudden change condition segment to obtain the load sequence, wherein the second step length is less than the first step length.

[0009] An adaptive sampling step size driven by load variation is adopted. A large step size is used to reduce the data size in the stable working condition section, and a small step size is used to capture the stress peak in the sudden working condition section. While ensuring that the random impact characteristics of the road surface are not lost, the amount of preprocessing data for finite element analysis is greatly reduced.

[0010] Preferably, the step of meshing the leaf spring model to obtain multiple mesh nodes includes: meshing the leaf spring model with a preset mesh size; calculating the geometric sensitivity factor of each position point in the leaf spring model; and, in response to the geometric sensitivity factor being greater than the sensitivity threshold, further subdividing the mesh of the corresponding position point to achieve local mesh refinement.

[0011] By re-dividing the local mesh based on the geometric sensitivity factor, computational resources are tilted toward structurally weak areas. This solves the problem that the global coarse mesh cannot accurately reconstruct the stress field in high-stress gradient regions such as leaf spring lugs without increasing the global mesh size.

[0012] Preferably, the geometric sensitivity factor is positively correlated with both the principal curvature and local thickness of the surface at the corresponding location point.

[0013] Preferably, the step of obtaining the stress sequence of each grid node under the test condition through finite element analysis includes: constructing a nonlinear model from load vector to node stress using the Kriging algorithm; inputting each load vector in the load sequence into the nonlinear model to obtain the stress of each grid node under each load vector, thereby obtaining the stress sequence of each grid node.

[0014] Preferably, the construction process of the nonlinear model includes: performing Latin hypercube sampling in a hypercube space composed of six-axis loads to obtain feature sample points, each feature sample point corresponding to a load vector; obtaining the stress of each grid node under the feature sample points as sample data, and splitting the sample data into a training set and a validation set; training the nonlinear model using the training set, and calculating the root mean square error of the nonlinear model using the validation set; and, in response to the root mean square error being not less than a preset error upper limit, adding sample data to continue training the nonlinear model.

[0015] By using the Kriging algorithm to construct a nonlinear surrogate model of load to stress, and combining it with Latin hypercube sampling and root mean square error feedback to evolve the nonlinear model, we can obtain a rapid prediction of stress sequence with a small amount of sample data, eliminating the huge time overhead caused by frame-by-frame simulation.

[0016] Preferably, the real-time statistical analysis of cumulative fatigue damage and damage saturation of each grid node on the stress sequence using the rainflow counting method includes: extracting the stress cycles of any grid node in each active segment using the rainflow counting method; calculating the ratio of the actual number of stress cycles at each stress amplitude to the fatigue life limit of the material at the corresponding stress amplitude to obtain a single-level damage value; summing the single-level damage values ​​at each stress amplitude to obtain the damage increment of the grid node in each active segment; statistically analyzing the cumulative fatigue damage of the grid node in real time based on the damage increment within each active segment; dividing the cumulative fatigue damage by a failure threshold to obtain the damage saturation; the active segment is the time interval during which the grid node is not in a dormant node state.

[0017] Preferably, the method for obtaining the cumulative fatigue damage of any grid node includes: obtaining the damage increment of the grid node in each active segment in real time; and adding the damage increment of the grid node in the current active segment to the damage increment of each previous active segment to obtain the updated cumulative fatigue damage.

[0018] The cumulative fatigue damage of each grid node is statistically analyzed in real time by accumulating the damage increment of active fragments. This time-series accumulation ensures the continuity of data statistics and can completely trace the full life cycle damage record of each grid node even during the dynamic switching of computing resources.

[0019] Preferably, the multi-channel load spectrum is a time series of six-axis loads at the leaf spring seat, front lug, and rear lug, wherein the loads include forces in three axes and torques in three axes.

[0020] Preferably, the step of constructing a damage cloud map based on the second-order difference sequence of the cumulative fatigue damage and stress sequence of each grid node includes: in the second-order difference sequence of the stress sequence of any grid node, counting the number of targets whose second-order difference value is greater than the difference threshold, and calculating the final damage value of the grid node, wherein the final damage value is positively correlated with both the cumulative fatigue damage and the number of targets.

[0021] By introducing the second-order difference of the stress sequence as a correction factor, the influence of high-frequency pulse load on damage assessment is enhanced. It can accurately capture the transient impact effect caused by road potholes or emergency braking, effectively identify and warn of brittle damage risk areas induced by drastic stress changes, and improve the accuracy of damage cloud map in locating weak points.

[0022] The technical solution of this application has the following beneficial technical effects:

[0023] The load sequence is adaptively sampled based on the load variation, and the grid nodes are locally refined according to the geometric complexity. In the rainflow calculation process, a dynamic dormancy mechanism for the grid nodes is introduced based on their contribution to the overall fatigue damage, which can significantly reduce the computational power and time cost required for fatigue analysis. Furthermore, the cumulative fatigue damage obtained from the rainflow calculation is corrected by using the second-order difference of the stress sequence of the grid nodes, which solves the problem of ignoring transient impact damage in traditional fatigue analysis and improves the accuracy of fatigue analysis. Attached Figure Description

[0024] Figure 1 This is a flowchart of a fatigue analysis method for automotive leaf springs based on simulation model optimization, according to an embodiment of this application.

[0025] Figure 2 This is a schematic diagram of a leaf spring model according to an embodiment of this application.

[0026] Figure 3 This is a schematic diagram of the mesh nodes in the leaf spring model according to an embodiment of this application.

[0027] Figure 4 This is a schematic diagram of a damage cloud map according to an embodiment of this application. Detailed Implementation

[0028] This application provides a simulation model-based fatigue analysis method for automotive leaf springs, primarily applicable to structural reliability assessment during the R&D phase of vehicle suspension systems. Specifically, it addresses the accurate assessment of fatigue life and weak point location of multi-leaf laminated automotive leaf springs during durability road testing. The road test conditions include, but are not limited to, braking, steering, and random road impact conditions. Figure 1 This is a flowchart illustrating a simulation model-based fatigue analysis method for automotive leaf springs according to an embodiment of this application. Figure 1 As shown, the fatigue analysis method for automotive leaf springs based on simulation model optimization includes steps S101 to S105, which are described in detail below.

[0029] S101, sample the multi-channel load spectrum under the test conditions to obtain the load sequence.

[0030] In one embodiment, the multi-channel load spectrum refers to the time series of six-axis loads acquired during the actual vehicle road spectrum acquisition process at the leaf spring seat, front hanger, and rear hanger. The loads include forces in three axes and torques in three axes. The load sequence refers to discrete data, after adaptive sampling processing, used as input to the finite element model for strength calculation.

[0031] To achieve adaptive adjustment of the sampling step size, it is necessary to accurately quantify the transient characteristics of the load signal, leading to the process of sampling the multi-channel load spectrum under the test conditions. The load change at each moment in the multi-channel load spectrum is obtained, and the average load change is calculated. By comparing the load change at any moment with the average load change, the multi-channel load spectrum is divided into a stable condition segment and a sudden change condition segment. Sampling is performed with a first step size in the stable condition segment and a second step size in the sudden change condition segment, resulting in a load sequence, where the second step size is smaller than the first step size. The average load change reflects the load fluctuation level under the test conditions; using it as a threshold benchmark can effectively identify high-frequency impacts caused by road potholes or emergency braking.

[0032] Specifically, representing time The rate of load mutation in terms of the degree of load change Satisfying the relation:

[0033]

[0034] In the formula, and The first one in the multi-channel load spectrum Road passage at all times and time The size of the load; The total number of channels is 18, which includes six-way loads at three locations: leaf spring seat, front hanger, and rear hanger. This represents the average load variation.

[0035] Understandably, when a vehicle is traveling on a smooth road, the real-time load change is less than the average load change, resulting in a higher load abrupt change rate. If the value is less than or equal to 1, the current time is determined to be a stable operating condition segment. A larger first step length is used to sample the stable operating condition segment to reduce redundant calculations. An example value for the first step length is 0.5 seconds. When the vehicle encounters a random impact causing a sudden change in load rate... When the value is greater than 1, the current time is determined to be a sudden change segment of the working condition. The sampling of the sudden change segment of the working condition is automatically switched to a smaller second step size in order to capture the complete stress peak characteristics. An example value of the second step size is 0.1 seconds.

[0036] In this way, by adaptively setting the sampling step size based on the load change, the data scale of non-critical working conditions is significantly reduced without losing key fatigue damage features, thereby improving the efficiency of subsequent simulation calculations.

[0037] S102, mesh the leaf spring model to obtain multiple mesh nodes.

[0038] In one embodiment, the leaf spring model is a pre-built numerical model that reflects the geometry and assembly relationships of the automotive leaf spring. (See [link to relevant documentation]). Figure 2 This is a schematic diagram of a leaf spring model according to an embodiment of this application. The leaf spring model includes a leaf spring seat 100, a front lug 200, and a rear lug 300. A mesh node refers to the smallest computational unit formed after discretizing the leaf spring model during the finite element preprocessing stage.

[0039] To achieve a balance between computational accuracy and computational scale, the model topology needs to be optimized. Therefore, the leaf spring model needs to be meshed. The leaf spring model is meshed with a preset mesh size; the geometric sensitivity factor at each location point in the leaf spring model is calculated; in response to the geometric sensitivity factor being greater than the sensitivity threshold, the mesh of the corresponding location point is further subdivided to achieve local mesh refinement; the geometric sensitivity factor is positively correlated with the principal curvature and local thickness of the surface at the corresponding location point.

[0040] Understandably, based on the geometric sensitivity factor, key areas with abrupt curvature changes or small thicknesses in the leaf spring structure, such as the lug part or the end of the spring, can be automatically identified, thereby improving the accuracy of mesh generation in these high strain gradient areas without changing the global mesh density.

[0041] Specifically, the geometric sensitivity factor is used to measure the mechanical sensitivity of a corresponding location point to the load response. Geometric sensitivity at the location Satisfying the relation:

[0042]

[0043] In the formula, Location point The principal curvature of the surface at that point, its dimension is ; This represents the local thickness at the corresponding location point, in units of... ; The standard thickness of the leaf spring main leaf is exemplarily taken as [value]. ; This is a correction coefficient used to reflect the gain intensity of the geometric sensitivity factor as it changes with geometric shape; its preferred value is [value missing]. .

[0044] Understandably, by coupling the principal curvature to the thickness ratio, it is possible to accurately identify complex locations of the leaf spring, such as the lug forming point and the end taper change point. When When the sensitivity exceeds a preset threshold, the region is identified as a potential stress concentration area, and sub-model mesh refinement is performed. The sensitivity threshold is exemplarily set to a value of... Please see. Figure 3 This is a schematic diagram of the mesh nodes in the leaf spring model according to an embodiment of this application. It can be seen that the mesh node density at the leaf spring seat, the front lug, and the rear lug is greater than the mesh node density in other areas.

[0045] In this way, computing resources are tilted toward geometrically complex regions, which significantly improves the accuracy of capturing local high-stress gradients while ensuring overall computing efficiency.

[0046] S103, apply loads to the leaf spring model according to the load sequence, and perform finite element analysis to obtain the stress sequence of each mesh node under the test conditions.

[0047] In one embodiment, the load sequence includes a time series of six-axis loads at the leaf spring seat, front lug, and rear lug. At any given time, the six-axis loads at the leaf spring seat, front lug, and rear lug reflect the stress state of the leaf spring. By applying loads to the leaf spring model based on the stress state of the leaf spring at each time and performing finite element analysis, the stress at each mesh node at each time can be obtained. For any given mesh node, the stress at each time constitutes the stress sequence of that mesh node.

[0048] In another embodiment, since the load sequence is too long, applying loads to the leaf spring model based on the stress state of the leaf spring at each moment would undoubtedly result in a huge workload. Therefore, a nonlinear model can be constructed to predict the stress sequence of each grid node. Specifically, obtaining the stress sequence of each grid node under the test condition through finite element analysis includes: constructing a nonlinear model from the load vector to the node stress using the Kriging algorithm; inputting each load vector in the load sequence into the nonlinear model to obtain the stress of each grid node under each load vector, thereby obtaining the stress sequence of each grid node; the load vector includes the leaf spring seat, the front lug, and the six-axis load at the front lug.

[0049] The construction process of the nonlinear model includes: performing Latin hypercube sampling within a hypercube space composed of six-directional loads to obtain feature sample points, each corresponding to a load vector; acquiring the stress of each grid node under the feature sample points as sample data, and splitting the sample data into a training set and a validation set; training the nonlinear model using the training set, and calculating the root mean square error of the nonlinear model using the validation set; and adding sample data to continue training the nonlinear model in response to the root mean square error not being less than a preset error upper limit. Understandably, since the stress at each grid node of an automotive leaf spring exhibits geometric nonlinearity during the loading process, the nonlinear model constructed using the Kriging algorithm can capture the microscopic details of the stress field evolution with load.

[0050] Specifically, the prediction accuracy of the nonlinear model is measured by the root mean square error. An evaluation is conducted, and the following relationship is satisfied:

[0051]

[0052] In the formula, To verify the quantity of sample data in the set; For the nonlinear model, the first in the validation set The output results for each sample data point, in units of... ; To verify the first set The actual results obtained from finite element analysis of sample data, in units of Both the output and actual results include the stress of all mesh nodes.

[0053] Understandably, the root mean square error (RMSE) reflects the degree to which the output of the nonlinear model deviates from the true result. A preset error upper limit is set, for example, a value of 2. When the RMSE exceeds the preset error upper limit, it indicates that the current feature sample point coverage density is insufficient to support the reconstruction of the complex nonlinear stress field. Latin hypercube sampling points are automatically added in the hypercube space composed of six-axis loads until the model converges, ensuring that the stress sequence output by the nonlinear model conforms to the true situation.

[0054] In this way, by using a high-precision nonlinear surrogate model to replace the traditional time-consuming frame-by-frame finite element simulation, the time to obtain the stress sequence under the full load condition sequence is greatly shortened.

[0055] S104 uses the rainflow counting method to statistically analyze the cumulative fatigue damage and damage saturation of each grid node in real time on the stress sequence.

[0056] In one embodiment, cumulative fatigue damage refers to the dimensionless value of the accumulation of micro-damage at a mesh node over time under the action of a stress sequence; damage saturation refers to the ratio of the current cumulative fatigue damage to the material failure limit, used to characterize the degree to which a node approaches fatigue fracture.

[0057] Since the leaf spring model is divided into multiple mesh nodes, it is necessary to perform real-time statistics on the cumulative fatigue damage and damage saturation of each mesh node. To improve computational efficiency, a dynamic computational resource allocation mechanism needs to be established. Specifically, the real-time statistics of the cumulative fatigue damage and damage saturation of each mesh node on the stress sequence using the rainflow counting method include: extracting the stress cycles of any mesh node in each active segment using the rainflow counting method; calculating the ratio of the actual number of stress cycles at each stress amplitude to the fatigue life limit of the material at the corresponding stress amplitude to obtain the single-level damage value; summing the single-level damage values ​​at each stress amplitude to obtain the damage increment of the mesh node in each active segment; calculating the cumulative fatigue damage of the mesh node in real time based on the damage increment within each active segment; and dividing the cumulative fatigue damage by the failure threshold to obtain the damage saturation.

[0058] Among them, the rainflow counting method can simplify the complex stress sequence into a series of stress cycles with specific amplitudes and average values, thus conforming to the fatigue characteristics of materials. In this embodiment, the three-peak valley method in rainflow calculation can be used to perform real-time statistics on the stress cycles in the stress sequence. Each time a new stress is obtained, the extreme value stack is updated and the cycle closure is determined, thereby realizing real-time statistics on cumulative fatigue damage and damage saturation. Furthermore, the damage saturation is combined to avoid performing rainflow counting calculations on all mesh nodes in the full time domain, thus reducing the amount of computation.

[0059] Starting from the beginning of the load sequence, the mesh nodes At any moment damage saturation Satisfying the relation:

[0060]

[0061] In the formula, For grid nodes At any moment Cumulative fatigue damage; The failure threshold is preferably set to a value that is suitable for the following conditions: .

[0062] Understandably, by monitoring damage saturation in real time This allows for the quantification of the contribution weight of each region of the leaf spring to the overall fatigue life. The calculation involves mesh nodes. At any moment Damage saturation and time The difference in damage saturation is used to obtain the mesh node. At any moment The growth rate of the damage saturation is less than a first threshold, indicating that the mesh node... The impact on overall fatigue damage is small and negligible; therefore, the corresponding mesh node is set to a dormant node, pausing the rainflow counting method calculation. The first threshold is used to identify non-critical nodes that contribute very little to overall fatigue damage; an example value is... By putting these nodes into a dormant state, a large amount of redundant computation can be eliminated.

[0063] The method for obtaining the cumulative fatigue damage of any grid node includes: obtaining the damage increment of the grid node in each active segment in real time, wherein the active segment is the time interval during which the grid node is not in a dormant node state; and accumulating the damage increment of the grid node in the current active segment with the damage increment of each previous active segment to obtain the updated cumulative fatigue damage.

[0064] Grid nodes In the currently active segment Inner Time Real-time statistics of damage increment Satisfying the relation:

[0065]

[0066] In the formula, To the currently active segment Inner Time The number of real-time statistics The actual number of stress cycles at each stress level; For the material in the first The fatigue life limit under the highest stress amplitude is determined by the material. Curve determined; This represents the total number of stress amplitude levels. Based on the linear cumulative damage superposition theory, the mesh nodes are time-series... By accumulating the damage increments within each active segment, mesh nodes can be implemented. Real-time statistics of cumulative fatigue damage.

[0067] It should be noted that, due to the grid nodes during dormancy... The stress sequence contributes very little to the overall fatigue damage; therefore, when considering the mesh nodes... When performing real-time statistics on cumulative fatigue damage, the influence of the stress sequence during the dormant period on the cumulative fatigue damage is ignored. Response time... Grid nodes For dormant nodes, the damage increments of all previously active segments are simply accumulated and used as the time step. Cumulative fatigue damage, in response to time Grid nodes If it is a non-dormant node, i.e., in an active segment, then the grid node will be... Within the current active segment Real-time statistics of damage increment The time interval is obtained by summing the damage increments of each previous active segment. The cumulative fatigue damage.

[0068] In this embodiment, to avoid the risk that existing non-critical areas may evolve into new fatigue hotspots due to changes in driving conditions, it is necessary to monitor the load change in the load sequence. In response to a load change exceeding a second threshold, each dormant node is activated to re-execute the rainflow counting method; wherein, at time... The load change is Satisfying the relation:

[0069]

[0070] In the formula, and The first one in the multi-channel load spectrum Road passage at all times and time The size of the load; The total number of channels is 18, which includes six-way loads at three locations: leaf spring seat, front lug, and rear lug; the second threshold is 3.

[0071] In this way, by identifying non-critical nodes that contribute very little to the overall fatigue damage through damage saturation and placing these non-critical nodes in a dormant state, combined with a dynamic activation strategy based on load changes, the risk of missing fatigue hotspots caused by load changes can be eliminated while reducing computational load and improving computational efficiency.

[0072] S105, at the end of the load sequence, a damage cloud map is constructed based on the cumulative fatigue damage of each grid node and the second-order difference sequence of the stress sequence.

[0073] In one embodiment, the termination time corresponds to the last sampling point in the load sequence; the second-order difference sequence of the stress sequence can reflect the acceleration characteristics of stress changing with time and the impact of stress on the grid nodes. The larger the second-order difference value in the second-order difference sequence, the greater the impact of stress change on the corresponding grid node, and the greater the risk of microcrack propagation.

[0074] To ultimately determine the fatigue life of the leaf spring structure and improve the accuracy of transient impact damage assessment, it is necessary to dynamically correct the cumulative fatigue damage under the entire test condition, thereby completing the construction of the damage cloud map.

[0075] In the second-order difference sequence of stress sequence of any grid node, count the number of targets whose second-order difference value is greater than the difference threshold; construct a correction factor based on the number of targets, and use the product of cumulative fatigue damage and correction factor as the final damage value of the grid node.

[0076] Specifically, grid nodes Second-order difference Satisfying the relation:

[0077]

[0078] In the formula, , and Grid nodes At any moment ,time and time The stress value.

[0079] Understandably, by introducing differential operations, the impact effect of high-frequency pulses from the road surface on the leaf spring can be accurately captured. The differential threshold can be set to half of the fatigue limit stress of the leaf spring material, and for example, a value of 5 MPa can be taken.

[0080] Furthermore, grid nodes final damage value Satisfying the relation:

[0081]

[0082] In the formula, For grid nodes At the end of the load sequence Cumulative fatigue damage; For grid nodes The number of targets in the stress sequence whose second-order difference is greater than the difference threshold; The total length of the second-order difference sequence of the stress sequence; is the stress sensitivity coefficient, which characterizes the sensitivity of a material to impact loads, with a preferred reference value of 0.2.

[0083] Understandably, when the leaf spring is under stable operating conditions, the second-order difference is small, and the final damage value tends to be the cumulative fatigue damage. When encountering gravel roads or emergency braking causing frequent and drastic stress jumps, the correction factor increases, thereby enhancing the risk of fatigue weak points. After obtaining the final damage values ​​of all mesh nodes, a damage contour map is obtained; please refer to [link to relevant documentation]. Figure 4 This is a schematic diagram of a damage cloud map according to an embodiment of this application. The damage cloud map includes the final damage values ​​of all grid nodes. Furthermore, the grid node with the largest final damage value in the damage cloud map can be identified as a fatigue hotspot.

[0084] Thus, by combining the cumulative fatigue damage obtained by the rainflow counting method with the second-order difference value of the stress sequence, the accuracy of the simulation model in predicting the failure behavior of leaf springs under varying working conditions is significantly improved, providing a reference for the impact-resistant reliability design of structures.

[0085] It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application shall be determined by the appended claims.

Claims

1. A fatigue analysis method for automotive leaf springs based on simulation model optimization, characterized in that, The analysis method includes: sampling the multi-channel load spectrum under test conditions to obtain a load sequence; The leaf spring model is meshed to obtain multiple mesh nodes; Loads were applied to the leaf spring model according to the load sequence, and finite element analysis was performed to obtain the stress sequence of each mesh node under the test conditions. The cumulative fatigue damage and damage saturation of each grid node in a stress sequence are statistically analyzed in real time using the rainflow counting method. This includes: extracting stress cycles of any grid node in each active segment using the rainflow counting method; calculating the ratio of the actual number of stress cycles at each stress amplitude to the fatigue life limit of the material at the corresponding stress amplitude to obtain a single-level damage value; summing the single-level damage values ​​at each stress amplitude to obtain the damage increment of the grid node in each active segment, where the active segment is the time interval during which the grid node is not in a dormant node state; statistically analyzing the cumulative fatigue damage of the grid node in real time based on the damage increment within each active segment; dividing the cumulative fatigue damage by a failure threshold to obtain the damage saturation; setting the corresponding grid node as a dormant node and pausing the rainflow counting method calculation in response to a damage saturation growth rate less than a first threshold; monitoring the load change in the load sequence, and activating each dormant node to re-execute the rainflow counting method in response to a load change greater than a second threshold. At the end of the load sequence, a damage cloud map is constructed based on the cumulative fatigue damage of each grid node and the second-order difference sequence of the stress sequence.

2. The fatigue analysis method for automotive leaf springs based on simulation model optimization according to claim 1, characterized in that, The sampling of the multi-channel load spectrum under the test conditions to obtain the load sequence includes: Obtain the load change at each time point in the multi-channel load spectrum and calculate the average load change; By comparing the load change and the average load change at any given time, the multi-channel load spectrum is divided into a stable operating condition segment and a sudden change operating condition segment. The load sequence is obtained by sampling with a first step length during the stable operating condition segment and a second step length during the sudden operating condition segment, where the second step length is less than the first step length.

3. The fatigue analysis method for automotive leaf springs based on simulation model optimization according to claim 1, characterized in that, The process of meshing the leaf spring model to obtain multiple mesh nodes includes: The leaf spring model is meshed using a preset mesh size; Calculate the geometric sensitivity factor at each position point in the leaf spring model; In response to a geometric sensitivity factor exceeding a sensitivity threshold, the grid at the corresponding location point is further subdivided to achieve local grid refinement.

4. The fatigue analysis method for automotive leaf springs based on simulation model optimization according to claim 3, characterized in that, The geometric sensitivity factor is positively correlated with the principal curvature and local thickness of the surface at the corresponding location point.

5. The fatigue analysis method for automotive leaf springs based on simulation model optimization according to claim 1, characterized in that, The process of obtaining the stress sequence of each grid node under the test condition through finite element analysis includes: constructing a nonlinear model from load vector to node stress using the Kriging algorithm; inputting each load vector in the load sequence into the nonlinear model to obtain the stress of each grid node under each load vector, thereby obtaining the stress sequence of each grid node.

6. The fatigue analysis method for automotive leaf springs based on simulation model optimization according to claim 5, characterized in that, The process of constructing the nonlinear model includes: Latin hypercube sampling is performed in the hypercube space composed of six-axis loads to obtain feature sample points, each feature sample point corresponding to a load vector. The stress of each grid node under the feature sample point is obtained as sample data, and the sample data is split into training set and validation set; The nonlinear model is trained using the training set, and the root mean square error of the nonlinear model is calculated using the validation set. In response to the root mean square error being no less than a preset error upper limit, additional sample data is added to continue training the nonlinear model.

7. The fatigue analysis method for automotive leaf springs based on simulation model optimization according to claim 1, characterized in that, Methods for obtaining the cumulative fatigue damage of any grid node include: Real-time acquisition of the damage increment of the grid node in each active segment; The damage increment of the grid node in the current active segment is accumulated with the damage increment of each previous active segment to obtain the updated cumulative fatigue damage.

8. The fatigue analysis method for automotive leaf springs based on simulation model optimization according to claim 1, characterized in that, The multi-channel load spectrum is a time series of six-axis loads at the leaf spring seat, front lug, and rear lug. The loads include forces in three axes and torques in three axes.

9. The fatigue analysis method for automotive leaf springs based on simulation model optimization according to claim 1, characterized in that, The construction of the damage cloud map based on the second-order difference sequence of the cumulative fatigue damage and stress sequence of each grid node includes: in the second-order difference sequence of the stress sequence of any grid node, counting the number of targets whose second-order difference value is greater than the difference threshold, and calculating the final damage value of the grid node. The final damage value is positively correlated with both the cumulative fatigue damage and the number of targets.

Images