Damage evolution prediction methods, devices, equipment, media, and products based on dynamic order reduction

Abstract

This application discloses a method, apparatus, equipment, medium, and product for predicting damage evolution based on dynamic order reduction, relating to the field of aero-engine health management and monitoring. The method includes: conducting finite element fatigue / creep-fatigue simulation tests on structural components covering stress states under various operating conditions, and extracting the time-physics field matrix from the finite element fatigue simulation results under each operating condition; dynamically reducing the order of the time-physics field matrix to reduce its dimension, obtaining a reduced-order matrix; constructing a reduced-order model; restoring the equivalent stress field distribution based on the reduced-order model to obtain the reduced-order physical field; predicting the damage distribution evolution and fatigue life under fatigue or creep-fatigue loading based on the reduced-order physical field, and making corrections based on the fatigue life prediction model. This application can improve the efficiency and universality of life prediction.

Description

Technical Field

[0001] This application relates to the field of aero-engine health management and monitoring, and in particular to a method, device, equipment, medium and product for predicting damage evolution based on dynamic order reduction. Background Technology

[0002] High-temperature rotating components, such as gas turbine rotors and engine turbine disks, often operate under complex stress states and high-temperature environments, making them prone to failure due to high- and low-cycle fatigue, creep-fatigue, and other forms of damage. Furthermore, many components undergo different processing techniques at connections and mating points, such as different welding processes and surface strengthening processes, depending on specific application requirements. These processing techniques alter the surface properties of the machined areas, resulting in discontinuous distribution of surface integrity, which significantly impacts the overall fatigue performance and fatigue life of the component. Surface integrity parameters, including residual stress, microhardness, and surface roughness, are essential parameters affecting the initiation of fatigue cracks and changes in fatigue life. In particular, the surface residual compressive stress field can be superimposed on fatigue loads, reducing or increasing the component load, and is one of the most important parameters affecting the fatigue life of components. Quantifying the influence of these parameters and predicting the life of such structures has significant engineering value and importance.

[0003] Currently, there are numerous studies on life prediction of such components based on data-driven methods. These methods are simple to operate and provide accurate prediction results, but they lack physical meaning and have poor interpretability. Life prediction methods based on nominal stress-strain and local stress-strain models often involve specific process parameters, thus limiting their application scope. At the same time, in order to obtain local stress-strain, prediction is often accompanied by complex and time-consuming finite element fatigue simulation or even crystal plastic fatigue simulation, resulting in low computational efficiency and making it impossible to achieve real-time health monitoring of power units.

[0004] For many rotating components and parts whose surface integrity changes significantly after machining, traditional fatigue life prediction methods based on local stress-strain methods often involve complex and time-consuming finite element fatigue simulations or even crystal plastic fatigue simulations, resulting in low computational efficiency and an inability to achieve real-time health monitoring of engine conditions. Therefore, there is an urgent need for an efficient and high-precision life prediction method to simplify the model for real-time health monitoring and life assessment, thereby improving the universality and computational efficiency of fatigue life prediction for such components. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, equipment, medium, and product for predicting damage evolution based on dynamic order reduction, which can improve the efficiency and universality of lifetime prediction.

[0006] To achieve the above objectives, this application provides the following solution:

[0007] Firstly, this application provides a damage evolution prediction method based on dynamic order reduction, including:

[0008] Finite element fatigue / creep-fatigue simulation tests were conducted on structural components covering stress states under various operating conditions, and the time-physics field matrix of the finite element fatigue simulation results under each operating condition was extracted.

[0009] The time-physics field matrix is ​​dynamically reduced in order to decrease its dimension, resulting in a reduced-order matrix.

[0010] A reduced-order model is constructed; the reduced-order model is obtained by training it under various finite element operating conditions using machine learning methods based on the reduced-order matrix and the external time-series load spectrum.

[0011] Based on the reduced-order model, the equivalent stress field distribution is restored to obtain the reduced-order physical field;

[0012] The damage distribution evolution and fatigue life prediction under fatigue or creep-fatigue loading are based on the reduced-order physical field, and are corrected based on the fatigue life prediction model. The fatigue life prediction model is determined after simulation experiments based on information data. The fatigue life prediction model includes a damage model and a surface integrity evolution model. The information data is obtained after conducting material-level and structural-level mechanical property tests and surface integrity characterization tests based on structural components. The information data includes: damage data, surface integrity evolution data, and life verification data.

[0013] Secondly, this application provides a damage evolution prediction device based on dynamic order reduction, comprising:

[0014] The testing and extraction module is used to conduct finite element fatigue / creep-fatigue simulation tests on structural components covering stress states under various operating conditions, and to extract the time-physics field matrix of the finite element fatigue simulation results under various operating conditions.

[0015] The dynamic order reduction module is used to dynamically reduce the order of the time-physics field matrix to reduce the matrix dimension and obtain a reduced-order matrix.

[0016] The order reduction model construction module is used to construct the order reduction model; the order reduction model is obtained by training it under various finite element operation conditions using machine learning methods based on the order reduction matrix and the external time-series load spectrum.

[0017] The order reduction module is used to restore the equivalent stress field distribution based on the reduced order model to obtain the reduced order physical field.

[0018] The prediction and correction module is used to predict the damage distribution evolution and fatigue life under fatigue or creep-fatigue loading based on the reduced-order physical field, and to correct it based on the fatigue life prediction model. The fatigue life prediction model is determined after simulation experiments based on information data. The fatigue life prediction model includes a damage model and a surface integrity evolution model. The information data is obtained after conducting material-level and structural-level mechanical property tests and surface integrity characterization tests based on structural components. The information data includes: damage data, surface integrity evolution data, and life verification data.

[0019] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned damage evolution prediction method based on dynamic order reduction.

[0020] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned damage evolution prediction method based on dynamic order reduction.

[0021] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned damage evolution prediction method based on dynamic order reduction.

[0022] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0023] This application provides a method, apparatus, equipment, medium, and product for damage evolution prediction based on dynamic order reduction. It involves conducting finite element fatigue / creep-fatigue simulation tests on structural components covering various operating conditions and stress states, and extracting the time-physics field matrix from the finite element fatigue simulation results for each operating condition. The time-physics field matrix is ​​then dynamically reduced in order to decrease its dimension, resulting in a reduced-order matrix. The equivalent stress field distribution is then reconstructed based on the reduced-order model to obtain the reduced-order physical field. This avoids the time-consuming and laborious finite element simulation process in traditional life prediction, thus rapidly obtaining the physical field distribution. Subsequently, the damage distribution evolution and fatigue life prediction under fatigue or creep-fatigue loading are predicted based on the reduced-order physical field, and the fatigue life prediction model is corrected, improving the efficiency of life prediction. Furthermore, the combination of the reduced-order physical field and the fatigue life prediction model makes it applicable to various processes that affect the surface integrity of materials. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 The flowchart shows a damage evolution prediction method based on dynamic order reduction.

[0026] Figure 2 This is a flowchart of the real-time prediction method for damage evolution of high-temperature rotating components based on a dynamic order reduction model, as described in this application.

[0027] Figure 3 This is a dimension drawing of a structural component sample according to an embodiment of this application;

[0028] Figure 4 The image shows the results of a high-temperature fatigue test on a GH4169 original smooth specimen and a specimen after roll strengthening, according to an embodiment of this application.

[0029] Figure 5 The image shows the high-temperature creep test results of a GH4169 smooth specimen according to an embodiment of this application.

[0030] Figure 6 This is a schematic diagram of the finite element simulation geometric model and mesh generation according to an embodiment of this application;

[0031] Figure 7 This is a comparison chart of the creep-fatigue life prediction SN curve and quantitative analysis of a GH4169 perforated plate structure at 650°C according to an embodiment of this application; wherein, Figure 7 (a) in the diagram is a schematic diagram of the corresponding creep-fatigue life and maximum axial stress; Figure 7 (b) in the diagram is a schematic diagram of the corresponding test life and predicted life;

[0032] Figure 8 This is a diagram showing the residual stress and hardness distribution of a GH4169 smooth round bar specimen after ultrasonic rolling according to an embodiment of this application; wherein, Figure 8 (a) in the diagram is a schematic diagram of residual stress and depth; Figure 8 (b) in the diagram is a schematic diagram of hardness and depth;

[0033] Figure 9 The figure shows the results of the maximum residual stress relaxation test on the surface of an ultrasonically rolled GH4169 smooth round bar under different loads according to an embodiment of this application.

[0034] Figure 10A schematic diagram illustrating the residual stress field of a geometric model according to an embodiment of this application;

[0035] Figure 11 This is a comparison chart of the fatigue life prediction SN curve and quantitative analysis of a GH4169 cold-extruded reinforced small-hole structural component at 650°C according to an embodiment of this application; wherein, Figure 11 (a) in the diagram is a schematic diagram of the corresponding fatigue life and stress amplitude; Figure 11 (b) in the diagram is a schematic diagram of the corresponding test life and predicted life;

[0036] Figure 12 This is a structural diagram of a damage evolution prediction device based on dynamic order reduction. Detailed Implementation

[0037] This application can replace the traditional finite element simulation process in prediction, greatly improving the computational efficiency of prediction while maintaining high prediction accuracy. In view of the phenomenon that welding and surface strengthening processes commonly seen in rotating parts cause significant changes in the surface integrity of components, this application corrects the damage model based on surface integrity parameters, which can be applied to different processing technologies and has a wide range of applications.

[0038] To simultaneously improve the real-time performance and computational efficiency of health monitoring for such components, this application performs lifetime prediction based on surface integrity status and integrates it into lifetime prediction software that visualizes the damage and surface integrity evolution process.

[0039] This application addresses the shortcomings of poor interpretability of data-driven models and the time-consuming and laborious prediction of local stress-strain models by developing a dynamic order reduction model. Combined with a data-driven approach, it avoids the extensive high-precision, high-density mesh finite element simulation process required in traditional lifetime prediction, thus rapidly obtaining the physical field distribution constructed under the predicted operating conditions. Simultaneously, it corrects the damage model based on surface integrity parameters and performs damage calculations and lifetime predictions, making it applicable to various processes that affect the surface integrity of materials.

[0040] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] In one exemplary embodiment, such as Figure 1 As shown, a damage evolution prediction method based on dynamic order reduction is provided, including:

[0042] Step 100: Conduct finite element fatigue / creep-fatigue simulation tests on structural components covering stress states under various operating conditions, and extract the time-physics field matrix of the finite element fatigue simulation results under each operating condition.

[0043] Step 200: Perform dynamic order reduction on the time-physics matrix to reduce the matrix dimension and obtain a reduced-order matrix.

[0044] Step 300: Construct the reduced-order model. The reduced-order model is obtained by training it under various finite element operation conditions using machine learning methods, based on the reduced-order matrix and the applied time-series load spectrum.

[0045] Step 400: Restore the equivalent stress field distribution based on the reduced-order model to obtain the reduced-order physical field.

[0046] Step 500: Based on the reduced-order physical field, predict the damage distribution evolution and fatigue life under fatigue or creep-fatigue loading, and then correct the fatigue life prediction model. The fatigue life prediction model is determined after simulation experiments based on information data; the fatigue life prediction model includes a damage model and a surface integrity evolution model; the information data is obtained after conducting material-level and structural-part-level mechanical property tests and surface integrity characterization tests based on structural components; the information data includes: damage data, surface integrity evolution data, and life verification data.

[0047] In one embodiment, finite element fatigue / creep-fatigue simulation tests are performed on structural components covering stress states under various operating conditions, and the time-physics matrix of the finite element fatigue simulation results under each operating condition is extracted, specifically including:

[0048] Finite element fatigue / creep-fatigue simulation tests were conducted on the structural components to obtain finite element fatigue simulation results under various operating conditions; the time-physics field matrix of the finite element fatigue simulation results under various operating conditions was extracted.

[0049] Finite element fatigue / creep-fatigue simulation tests were conducted on the structural components to obtain finite element fatigue simulation results under various operating conditions. The number of simulation cycles was consistent across all operating conditions. During the tests, the number of simulation cycles in the finite element fatigue / creep-fatigue simulation should be greater than the number of cycles required for the critical parts of the component to reach a steady state. Reaching a steady state is determined based on discrimination conditions and constraint conditions. The constraint conditions include: relative standard deviation constraint and mean trend constraint. The expression corresponding to the relative standard deviation constraint is:

[0050] .

[0051] The expression corresponding to the mean trend constraint is:

[0052] .

[0053] The expression corresponding to the discrimination condition is:

[0054] .

[0055] .

[0056] in, The standard deviation is relative. The width of the window; For the first i The week represents the average peak stress within the window of the initial week. For the first j Peak stress of the period; i, j All are serial numbers; For the first The week represents the average peak stress within the window of the initial week. The threshold for discrimination based on the mean; The threshold is the relative standard deviation. It should be determined based on the characteristics and magnitude of the data; The determination should be based on the characteristics and magnitude of the data. If both the relative standard deviation constraint and the mean trend constraint are satisfied, the fatigue cycle is considered to have entered a steady state.

[0057] When extracting finite element simulation results, the calculation results of each element in the same frame are saved in the same row according to the element number. The calculation results of each frame correspond to one row, forming the result physical field matrix.

[0058] In one embodiment, the time-physics matrix is ​​dynamically reduced in order to decrease its dimension, resulting in a reduced-order matrix. Specifically, this includes:

[0059] The singular value decomposition method is used to dynamically reduce the order of the time-physics field matrix, thereby reducing the matrix dimension and obtaining a reduced-order matrix. The decomposition process corresponding to the dynamic order reduction includes:

[0060] .

[0061] in, The time-physics field matrix; ; It is a left orthogonal matrix; Its column vectors are called left singular vectors; It is a diagonal matrix composed of singular values; ; It is a right orthogonal matrix; Its column vectors are right singular vectors; T is the transpose of the matrix.

[0062] reserve Center front r The largest singular value and its corresponding left and right singular vectors approximate the time-physics matrix in a low-rank form:

[0063] .

[0064] in, It is a reduced-order matrix; This is the reduced-order left orthogonal matrix; This is the diagonal matrix composed of singular values ​​after the order has been reduced; This is the reduced-order right orthogonal matrix; r The dimensionality reduction order is usually determined by the proportion of accumulated energy.

[0065] In one embodiment, the equivalent stress field distribution is restored according to the reduced-order model to obtain the reduced-order physical field, specifically including:

[0066] Using the KNN nearest neighbor algorithm, select k For each nearest neighbor operating condition, the corresponding left singular vector is determined based on a reduced-order model, and matrix multiplication is used to find the corresponding... , Matrix multiplication yields the Mises equivalent stress field distribution under nearby operating conditions.

[0067] The weighted average vector of the Mises equivalent stress field under nearby working conditions is calculated based on Euclidean distance. This serves as the result of the stress field distribution under the predicted working condition, in order to obtain the reduced physical field.

[0068] As an optional implementation, damage distribution evolution and fatigue life prediction under fatigue or creep-fatigue loading are performed based on a reduced-order physical field, and the fatigue life prediction model is then corrected. Specifically, this includes:

[0069] Damage distribution evolution and fatigue life prediction under fatigue or creep-fatigue loading are based on reduced-order physical fields. For homogeneous structures, characteristic physical parameters are extracted for each loading cycle according to the load type and input into the corresponding damage model for fatigue or creep-fatigue life calculation and accumulation. For heterogeneous structures, a fatigue life prediction model considering average stress is used to fit fatigue test data of smooth specimens, and the results of relaxation test of material residual stress with cycle count are fitted, as well as the changes of surface residual stress and hardness with depth are fitted.

[0070] The external load is obtained from the equivalent stress field, and the residual stress at a set time / cycle number is determined. The residual stress and hardness are considered to modify the lifetime model based on the reduced physical field, resulting in a modified lifetime model. The external load, residual stress at the set time / cycle number, and hardness are input into the modified lifetime model to determine the damage in the current cycle. The cumulative damage is determined as the number of cycles increases according to the Miner linear damage accumulation criterion.

[0071] This application develops a dynamic order reduction model, combined with a data-driven approach, avoiding the time-consuming and laborious finite element simulation process in traditional lifetime prediction, thereby quickly obtaining the physical field distribution constructed under the prediction conditions. Simultaneously, based on surface integrity parameters, the damage model is corrected for damage calculation and lifetime prediction, making it applicable to various processes that affect the surface integrity of materials. For example... Figure 2 As shown, in practical applications, the following steps are included.

[0072] S1: Structural components designed to cover stress states under various operating conditions, based on the stress state of critical / dangerous parts of the structure.

[0073] S2: Conduct material-level and structural component-level mechanical property tests and surface integrity characterization tests to obtain damage model and surface integrity evolution model data and lifetime prediction result verification data required for prediction.

[0074] Step S2 involves conducting fatigue tests and surface integrity characterization tests to obtain model fitting data, including fatigue tests on original smooth specimens under different mean stresses, fatigue tests on processed smooth specimens, residual stress relaxation tests, and hardness tests. Fatigue tests on structural components are also conducted to verify the accuracy of the fatigue life prediction results.

[0075] S3: Perform finite element fatigue / creep-fatigue simulations covering extreme and intermediate working conditions for original structural components.

[0076] S31: Perform fatigue simulation on the component to be predicted, where the number of incremental steps in the analysis steps corresponding to each fatigue cycle is consistent, and the number of simulation cycles performed under each working condition is consistent. The applied load conditions should include extreme working conditions under the component's operating environment as well as multiple intermediate working conditions.

[0077] S32: Use discrimination conditions, relative standard deviation constraints, and mean trend constraints to ensure that the number of finite element simulation cycles is greater than the number of cycles necessary for the critical point of the component to enter steady state.

[0078] S4: Extract the time-physics matrix from the finite element calculation results, perform dynamic model order reduction, and reduce the matrix dimension.

[0079] S41: Extract the time-physics matrix of the finite element fatigue simulation results under various working conditions. The matrix dimension is m×n, where m is the total number of incremental steps (frames) in the fatigue simulation, and n is the number of finite element model elements.

[0080] S42: Use Singular Value Decomposition (SVD) to decompose the original physical field matrix (time-physical field matrix) under each working condition to reduce the data dimensionality.

[0081] .

[0082] in, For the first i A singular value, To accumulate energy, the percentage is usually 99%, 99.9%, or higher.

[0083] S5: Using the external time-series load spectrum as input and the reduced-order matrix as output, construct and train a machine learning data-driven model.

[0084] Step S5 involves machine learning training, using the external time-series load spectrum as input and the reduced-order U matrix as output, and selecting a suitable model for training to form a reduced-order model library.

[0085] S6: Combining the damage model with the reduced-order physical field, calculate the damage distribution and accumulation, and lifetime under fatigue or creep-fatigue loading. For heterogeneous structures formed after surface strengthening, welding, and other processing, establish a model to characterize the distribution and evolution of surface integrity parameters, and correct the damage model by considering the influence of residual stress on mean stress and the change in hardness before and after processing.

[0086] S61: Calculate the distribution of the applied load field of the component at a certain moment under the predicted working condition, such as the Mises equivalent stress field. Using the KNN (K-Nearest Neighbor) algorithm, select the k nearest neighbor working conditions. Combined with the reduced-order model library saved in step S5, calculate the corresponding left singular vector for each working condition, and use matrix multiplication and corresponding... , Matrix multiplication yields the Mises equivalent stress field distribution of the component under the adjacent working conditions at that moment. The weighted average vector of the stress field under the adjacent working conditions is calculated based on the Euclidean distance d. This serves as the result of the stress field distribution under the predicted working condition.

[0087] .

[0088] in, d Euclidean distance. x These are the sample points for the operating conditions to be predicted. y This is a sample point for a certain neighboring working condition. The first working condition to be predicted i One input; For the first working condition of the neighboring unit i One input.

[0089] .

[0090] in, Let Y be the stress field distribution vector of the component under the predicted working condition, Y be the nearest neighbor working condition vector matrix, and w be the corresponding weight matrix. For the th jWeight of each nearest neighbor condition , which can be determined by The calculation yielded, where To prevent extremely small smooth terms with a denominator of 0. For the first j The Euclidean distance between the nearest neighbor conditions and the condition to be predicted.

[0091] S62: Based on the restored physical field matrix, damage distribution evolution and fatigue life are predicted. For homogeneous structures, according to the load type, characteristic physical parameters such as stress amplitude, average stress, and holding time are extracted for each loading cycle, and substituted into the corresponding damage model to calculate and accumulate the damage of each element, and calculate fatigue or creep-fatigue life.

[0092] S63: For heterogeneous structures, the influence of surface integrity parameter variations needs to be considered. A fatigue life prediction model considering average stress is used to fit the fatigue test data of the original smooth specimen. A residual stress relaxation model is used to fit the relaxation test results of the material's residual stress with varying cycle counts. A residual stress and hardness distribution model with depth is used to fit the variations of residual stress and hardness on the surface of the component to be predicted with depth.

[0093] S64: Based on the distance of each element from the component surface, an initial residual stress and hardness are assigned to each near-surface element using a residual stress and hardness distribution model. The residual stress of each element is calculated using a residual stress relaxation model at a specified time / cycle number.

[0094] S65: Using the fatigue life prediction model and experimental data of the original material, the fatigue life model is modified, taking into account the influence of residual stress on mean stress and the change in surface hardness before and after processing. The external load on the element, the residual stress in the current cycle, and the hardness data are substituted into the modified fatigue life model to calculate the damage of the element in the current cycle. Based on the Miner linear damage accumulation criterion, the cumulative damage of each element is calculated as the number of cycles increases. When the cumulative damage of any element reaches 1, the component is considered to have failed, and the corresponding cycle life is its fatigue life under that working condition.

[0095] S7: Integrated into life prediction software to enable rapid and accurate prediction of fatigue / creep-fatigue life of a given structure.

[0096] Step S7 integrates lifetime prediction software, providing functions such as model parameter fitting, fatigue lifetime prediction, and real-time visualization of damage and surface integrity evolution.

[0097] Example 1

[0098] Taking a porous structure made of GH4169 material as an example, creep-fatigue life prediction at 650℃ was conducted to verify the applicability and accuracy of the method. Mechanical property tests and creep-fatigue tests were performed on both smooth specimens and porous structures. The tests on smooth specimens were used to calibrate the material parameters of the damage model required for life prediction, including high-temperature fatigue and high-temperature creep tests. The porous structure was the component to be predicted, and its tests were used to verify the accuracy of the model prediction. The dimensions of the porous structure specimen are shown below. Figure 3 As shown, based on the design of cooling holes for a certain type of engine turbine disk, the hole diameter is 12mm. The life prediction steps for this component are as follows: a high-temperature fatigue test is performed on a smooth specimen at a temperature of 650℃ and a stress ratio of 0.1. The fatigue test results are as follows. Figure 4 As shown. High-temperature creep tests were conducted on smooth specimens at 650℃ and stress levels of 450-900 MPa. The creep test results are as follows. Figure 5 As shown.

[0099] Finite element creep-fatigue simulation was conducted on GH4169 perforated plate structures. A two-dimensional quarter-hole plate model was established using CPE4R quadrilateral elements. The mesh was refined (high degree of freedom) within a range of twice the hole diameter. Specific mesh generation, boundary condition settings, and loading methods are as follows: Figure 6 As shown in Table 1, the stress-strain relationship of the material during the finite element simulation was characterized using the Chaboche unified viscoplastic constitutive model. Stress-controlled creep-fatigue simulations were performed on porous structures under different working conditions, with maximum axial stress ranging from 200 to 1000 MPa, a step size of 25 MPa, a stress ratio of 0.1, and a holding time of 60 s.

[0100] Table 1. Chaboche Model Parameter Table

[0101]

[0102] Among them, C1, C2 C3 is the kinematic hardening parameter, and Qinf and b are the isotropic hardening parameters.

[0103] The Mises equivalent stress fields at each incremental step are extracted from the finite element simulation results under different working conditions to form the original physical field matrix of the component. Each row represents the Mises stress magnitude of each element in a time frame. The original physical field matrix under each working condition is decomposed based on SVD singular value decomposition. The accumulated energy ratio threshold is 99.9%. By setting the accumulated energy ratio, r can be determined, and the corresponding Ur is obtained. The Vr matrix, the order reduction process is shown in step S42.

[0104] For each operating condition, a time-series load spectrum matrix is ​​generated, with the first column representing time, the second column representing the nominal load magnitude, and the third column representing the cycle number. Using this matrix as input, the resulting Ur matrix is ​​decomposed and used as output. Machine learning models are then trained and saved to form a reduced-order model library. The machine learning model uses the XGboost distributed gradient boosting library and Bayesian optimization to search for and optimize hyperparameters.

[0105] For the load condition to be predicted, steady-state fatigue external load is calculated based on KNN weighted interpolation. The nearest neighbor group number is selected as k=2. After generating the time-series load spectrum input, the trained model is called to calculate the corresponding left singular vectors for the two load conditions. Matrix multiplication is then used to multiply these vectors with the corresponding Σ and V matrices to reconstruct the Mises equivalent stress field distribution of the component under the adjacent load conditions. The stress field distribution under the load condition to be predicted is obtained by weighted averaging the stress fields of the adjacent load conditions based on the Euclidean distance d.

[0106] For any given cycle, extract the Mises equivalent stress loading end value and calculate the stress amplitude and mean stress for that cycle. The Modified Basquin model is selected as the fatigue damage model, and the model is written as an explicit function of fatigue life as follows:

[0107] .

[0108] in, For fatigue life, For average stress, The stress amplitude, For tensile strength, α and b are model parameters. From this, the fatigue damage for that cycle can be obtained. The fatigue damage model was fitted using fatigue test data from smooth round bars.

[0109] The creep damage for the current week is calculated using the time fraction method. ,in To ensure the load time, This represents the creep life of the material under the current temperature and stress level. In this embodiment, a power function is used to perform a simple fitting of the creep life of GH4169 material under different loads at 650℃, so as to calculate the creep life of the material under different stress levels, such as... Figure 5 As shown. In summary, creep-fatigue damage in a given cycle can be determined as the sum of fatigue damage and creep damage, i.e. .

[0110] Based on Miner's linear damage accumulation criterion, the damage of each element is linearly accumulated. When the accumulated damage of any element reaches 1, the corresponding number of cycles is the creep-fatigue life of the component.

[0111] In this example, the creep fatigue test parameters are: temperature 650℃, stress amplitude range 650-850MPa, and stress ratio 0.1. The comparison results between predicted life and test life are as follows: Figure 7 As shown. Among them, Figure 7 (a) in the diagram is a schematic diagram of the corresponding creep-fatigue life and maximum axial stress; Figure 7 (b) in the diagram shows the corresponding experimental lifetime and predicted lifetime. Observing the prediction results, it can be seen that most prediction results fall within the 2x error range, demonstrating good prediction accuracy.

[0112] Example 2

[0113] Taking a porous structure made of GH4169 material reinforced by cold extrusion as an example, fatigue life prediction at 650℃ was performed to verify the applicability and accuracy of the method. In this example, fatigue tests and surface integrity characterization were performed on two types of specimens: smooth specimens and porous structures. The tests based on smooth specimens were used to calibrate the material parameters required for life prediction; the porous structure was the component to be predicted, and its tests were used to verify the accuracy of the model prediction. The specimen size of the perforated plate structure was the same as in Example 1, and the reinforcement method was multi-stage convex hull rotary cold extrusion reinforcement. The life prediction steps for this component were as follows: the smooth round bar specimen was reinforced by ultrasonic rolling. The rolling process parameters were: static pressure 200N, amplitude 20%, and 15 rolling passes. The radial residual stress and hardness distribution along the depth of the round bar after rolling are shown below. Figure 8 As shown. Among them, Figure 8 (a) in the diagram is a schematic diagram of residual stress and depth; Figure 8 (b) in the diagram shows the relationship between hardness and depth. High-temperature fatigue tests were conducted on the round bar specimens before and after rolling at 650℃ and a stress ratio of 0.1. The fatigue test results are as follows: Figure 4 As shown. Residual stress relaxation tests were conducted on the rolled round bar specimens under different loads. The stress relaxation test results are as follows. Figure 9 As shown.

[0114] Finite element fatigue simulation was conducted on unreinforced GH4169 perforated plate structural components. The simulation was identical to Example 1, except that the load type was pure fatigue. Figure 6 As shown. The stress-strain relationship of the material after fatigue reaches steady state is characterized using the cyclic Ramberg-Osgood constitutive model. The Ramberg-Osgood constitutive expression is as follows:

[0115] .

[0116] in, For strain amplitude, Let E be the stress amplitude, E be the cyclic modulus of elasticity, and K' and n' be the material parameters. The RO model parameters for GH4169 material at 650℃ are as follows: K'=2717.9, n'=0.192.

[0117] Fatigue simulations were performed on porous structural components under different nominal stress amplitudes. The stress amplitude was -1, the loading waveform was a triangular fatigue waveform, the loading frequency was 10Hz, the nominal stress amplitude ranged from 20 to 500 MPa, and the step size was 20 MPa, for a total of 25 working conditions. The data extraction, model order reduction, and training process of the finite element results physical field were the same as in Example 1.

[0118] Based on the element and node coordinate information, the distance between the center of each element and the edge of the hole is calculated. A cubic polynomial is used to fit the depth distribution data of the residual stress after hole strengthening. Initial residual stress is assigned to each element based on the distance from the element to the edge of the hole, such as... Figure 10 As shown, a cubic polynomial was used to fit the hardness distribution data along the depth after micro-hole reinforcement to obtain the maximum surface hardness along the depth.

[0119] Based on uniaxial fatigue test, fatigue interruption test, and surface integrity test data, a surface integrity evolution damage model was established. A residual stress relaxation model was also established to characterize the rapid relaxation behavior of residual stress with increasing cycle count, and this model was fitted based on the surface integrity test data.

[0120] .

[0121] Where N is the current week, The residual stress for the current cycle, R represents the initial residual stress after strengthening, and R is the stress ratio. The maximum surface hardness, The initial hardness of the material. The stress amplitude, The yield strength is used. Model parameters are fitted using residual stress test data from interrupted fatigue tests on smooth uniaxial specimens, with the residual stress taken as the maximum value along the depth from the specimen surface.

[0122] Similar to Example 1, the Modified Basquin model is selected as the fatigue damage model, and the model is written as an explicit function of fatigue life as follows:

[0123] .

[0124] Considering the influence of residual stress introduced by surface strengthening process on the average stress of applied load, and considering the effect of surface hardness change brought about by material surface microstructure refinement before and after processing on fatigue performance, residual stress and correction factor k are introduced to modify the original damage model.

[0125] .

[0126] Where k is the correction factor, and its expression is as follows:

[0127] .

[0128] in, The fatigue limit of the material. Let be the surface strengthening threshold stress amplitude, which represents the stress amplitude at which the uniaxial fatigue performance of the material begins to show a significant improvement as the stress amplitude decreases. It has a simple linear relationship with the initial residual stress of the strengthened material and can be written as the following expression:

[0129] .

[0130] Where m represents the model parameters. In this example, fatigue data of GH4169 material at 650℃ and its rolled hardening were used to fit the model. The fitted model parameters are summarized in Table 2.

[0131] Table 2 Summary of Damage Calculation Model Parameters

[0132]

[0133] Based on Miner's linear damage accumulation model, and combined with the stress field of each element obtained by order reduction, the applied load, residual stress data and hardness data are substituted into the damage model to calculate the damage of each element in the current cycle. As the cycle number increases, the damage of each element is linearly accumulated. When the accumulated damage of any element reaches 1, failure is considered to have occurred at that element.

[0134] .

[0135] Where D represents cumulative damage. For damage in a single week, i This represents the number of cycles.

[0136] Using the fitted model parameters, fatigue life prediction was performed on multi-stage convex hull rotary cold extrusion reinforced perforated plate structures under two different extrusion amounts: 30 μm and 50 μm. The fatigue test parameters were: temperature 650℃, loading frequency 10 Hz, stress amplitude range 400-600 MPa, and stress ratio 0.1. Comparison of predicted and experimental results is shown below. Figure 11 As shown, where, Figure 11 (a) in the diagram is a schematic diagram of the corresponding fatigue life and stress amplitude; Figure 11(b) in the figure is a schematic diagram of the corresponding test life and predicted life. It can be seen that the prediction method demonstrates good prediction accuracy and can achieve accurate prediction at all nominal load levels tested. Observing the quantitative comparison chart, it can be seen that, except for one point, all other predicted points fall within the 2x error band. The single data point with a large prediction error may be due to the dispersion of test data and the test life being much lower than expected.

[0137] In one exemplary embodiment, such as Figure 12 As shown, a damage evolution prediction device based on dynamic order reduction is provided, comprising:

[0138] The testing and extraction module is used to conduct finite element fatigue / creep-fatigue simulation tests on structural components covering stress states under various operating conditions, and to extract the time-physics field matrix of the finite element fatigue simulation results under various operating conditions.

[0139] The dynamic order reduction module is used to dynamically reduce the order of the time-physics matrix to reduce the matrix dimension and obtain a reduced-order matrix.

[0140] The order reduction model construction module is used to construct the order reduction model; the order reduction model is obtained by training it under various finite element operation conditions using machine learning methods based on the order reduction matrix and the external time-series load spectrum.

[0141] The order reduction module is used to restore the equivalent stress field distribution based on the reduced-order model to obtain the reduced-order physical field.

[0142] The prediction and correction module is used to predict the damage distribution evolution and fatigue life under fatigue or creep-fatigue loading based on the reduced-order physical field, and to correct it based on the fatigue life prediction model. The fatigue life prediction model is determined after simulation experiments based on information data. The fatigue life prediction model includes a damage model and a surface integrity evolution model. The information data is obtained after conducting material-level and structural-level mechanical property tests and surface integrity characterization tests based on structural components. The information data includes: damage data, surface integrity evolution data, and life verification data.

[0143] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0144] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0145] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0146] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0147] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A damage evolution prediction method based on dynamic reduction, characterized in that, The application relates to a method for predicting the fatigue life of a structure under fatigue or creep-fatigue load, and belongs to the technical field of fatigue life prediction. The method comprises the following steps: carrying out finite element fatigue / creep-fatigue simulation tests on a structure under various stress states of various operating conditions, and extracting time-physical field matrices of finite element fatigue simulation results under various operating conditions; performing dynamic reduction processing on the time-physical field matrices to reduce the matrix dimension, and obtaining a reduced matrix; constructing a reduced model; the reduced model is obtained by training each finite element operating condition by using a machine learning method based on the reduced matrix and an external time-varying load spectrum; performing equivalent stress field distribution restoration according to the reduced model to obtain a reduced physical field; performing damage distribution evolution and fatigue life prediction under fatigue or creep-fatigue load based on the reduced physical field, and correcting based on a fatigue life prediction model; the fatigue life prediction model is determined after simulation tests according to information data; the fatigue life prediction model comprises a damage model and a surface integrity evolution model; the information data is obtained based on a structure after material-level and structure-level mechanical property tests and surface integrity characterization tests; the information data comprises damage data, surface integrity evolution data and life verification data; performing damage distribution evolution and fatigue life prediction under fatigue or creep-fatigue load based on the reduced physical field, and correcting based on a fatigue life prediction model, specifically comprising: performing damage distribution evolution and fatigue life prediction under fatigue or creep-fatigue load based on the reduced physical field; wherein, for a homogeneous structure, according to the load type, the characteristic parameters of each loading cycle are extracted and input into the corresponding damage model to calculate and accumulate the fatigue or creep-fatigue life; the characteristic parameters comprise a stress amplitude, an average stress and a holding time; for a heterogeneous structure, a fatigue life prediction model considering the average stress is used to fit the smooth sample fatigue test data, and the relaxation test results of the material residual stress with the cycle number are fitted, and the surface residual stress and the hardness change with the depth size are fitted; obtaining external loads according to the equivalent stress field and determining the residual stress under a set time / cycle number; correcting the life model based on the reduced physical field by considering the influence of the residual stress on the average stress and the surface hardness change before and after processing based on the fatigue life prediction model and the information data, to obtain a corrected life model; inputting the external load, the residual stress and the hardness under the set time / cycle number into the corrected life model to determine the current cycle damage; 2. The dynamic reduced-order based damage evolution prediction method of claim 1, wherein, determining the cumulative damage according to the Miner linear damage accumulation criterion with the increase of the cycle number. The method comprises the following steps: carrying out finite element fatigue / creep-fatigue simulation tests on a structure under various stress states of various operating conditions, and extracting time-physical field matrices of finite element fatigue simulation results under various operating conditions; carrying out finite element fatigue / creep-fatigue simulation tests on the structure to obtain finite element fatigue simulation results under various operating conditions; extracting time-physical field matrices of finite element fatigue simulation results under various operating conditions.

3. The dynamic reduced-order based damage evolution prediction method of claim 2, wherein, The finite element fatigue / creep-fatigue simulation test is performed on the structural component, and finite element fatigue simulation results under various operating conditions are obtained, specifically including: The finite element fatigue / creep-fatigue simulation test is performed on the structural component, and finite element fatigue simulation results under various operating conditions are obtained, wherein the number of simulation cycles under each operating condition is consistent; during the test, the number of simulation cycles of the finite element fatigue / creep-fatigue simulation test should be greater than the number of cycles of the dangerous part of the component entering the steady state; the entry into the steady state is determined based on the discrimination condition and the constraint condition; The constraint condition includes a relative standard deviation constraint and a mean trend constraint; The expression corresponding to the relative standard deviation constraint is: ； The expression corresponding to the mean trend constraint is: ； The expression corresponding to the discrimination condition is: ； ； wherein, is the relative standard deviation; is the window width; is the mean of the peak stresses within a window starting at week i week; is the peak stress of the cycle; j is the peak stress of the cycle; i, j are sequence numbers; is the mean of the peak stresses within a window starting at week week; is the mean discrimination threshold; is the relative standard deviation discrimination threshold.

4. The dynamic reduced-order based damage evolution prediction method of claim 1, wherein, The time-physical field matrix is dynamically reduced in dimension to obtain a reduced matrix, specifically including: The time-physical field matrix is dynamically reduced in dimension by singular value decomposition to obtain a reduced matrix; The decomposition process corresponding to the dynamic reduction includes: ； wherein is a time-physical field matrix; is a left orthogonal matrix; is a diagonal matrix composed of singular values; is a right orthogonal matrix; T is the transpose of a matrix; reserved mid-fore r The time-physical field matrix is approximated in a low-rank form by retaining the largest singular value and its corresponding left and right singular vectors: ； wherein is a reduced order matrix; is a reduced left orthogonal matrix; is a reduced diagonal matrix composed of singular values; is a reduced right orthogonal matrix; r is a reduced dimension order.

5. The dynamic reduced-order based damage evolution prediction method of claim 4, wherein, The equivalent stress field distribution is restored based on the reduced model to obtain a reduced restored physical field, specifically including: KNN nearest neighbor algorithm is adopted, and 9 nearest neighbor working conditions are selected k The left singular vectors corresponding to each working condition are determined based on the reduced model, and the Mises equivalent stress field distribution of the adjacent working condition is obtained through matrix reduction. A weighted average vector of mises equivalent stress field under the adjacent working condition is calculated based on the Euclidean distance , as the stress field distribution result under the working condition to be predicted, to obtain the reduced physical field.

6. A device for damage evolution prediction based on dynamic reduction, characterized in that, including: The test and extraction module is configured to perform a finite element fatigue / creep-fatigue simulation test on a structural component covering various operating conditions, and extract a time-physical field matrix of finite element fatigue simulation results under various operating conditions; The dynamic reduction module is configured to dynamically reduce the time-physical field matrix in dimension to obtain a reduced matrix; The reduced model construction module is configured to construct a reduced model; the reduced model is obtained by training each finite element operating condition based on the reduced matrix and an external time sequence load spectrum using a machine learning method; The reduced restoration module is configured to restore the equivalent stress field distribution based on the reduced model to obtain a reduced restored physical field; The prediction and correction module is configured to predict the damage distribution evolution and fatigue life under fatigue or creep-fatigue load based on the reduced restored physical field, and correct based on a fatigue life prediction model; the fatigue life prediction model is determined based on information data obtained after simulation test; the fatigue life prediction model includes a damage model and a surface integrity evolution model; the information data is obtained based on a structural component, after material level and structural component level mechanical property test and surface integrity characterization test; the information data includes damage data, surface integrity evolution data, and life verification data; The prediction and correction module is configured to predict the damage distribution evolution and fatigue life under fatigue or creep-fatigue load based on the reduced restored physical field, and correct based on a fatigue life prediction model, specifically including: Based on the reduced physical field, the damage distribution evolution and fatigue life prediction under fatigue or creep-fatigue load are carried out; wherein, for homogeneous structure, according to the load type, the characteristic parameters of each loading cycle are extracted and input into the corresponding damage model for fatigue or creep-fatigue life calculation and accumulation; the characteristic parameters include stress amplitude, average stress and holding time; for heterogeneous structure, the fatigue life prediction model considering average stress is used to fit the smooth sample fatigue test data, and the relaxation test results of material residual stress with cycle number are fitted, and the surface residual stress and hardness change with depth are fitted; According to the equivalent stress field, the external load is obtained and the residual stress at the set time / cycle number is determined; Based on the fatigue life prediction model and the information data, considering the influence of residual stress on average stress and the change of surface hardness before and after processing, the life model based on the reduced physical field is modified to obtain the modified life model; The external load, residual stress and hardness at the set time / cycle number are input into the modified life model to determine the current cycle damage; According to the Miner linear damage accumulation criterion, the cumulative damage is determined with the increase of cycle number.

7. A computer device comprising: A memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the damage evolution prediction method based on dynamic reduction according to any one of claims 1-5.

8. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the damage evolution prediction method based on dynamic reduction according to any one of claims 1-5.

9. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the damage evolution prediction method based on dynamic reduction according to any one of claims 1-5.

Images