Fatigue crack propagation prediction method and system for hybrid-driven fiber metal laminate

Abstract

The invention discloses a hybrid-driven fatigue crack propagation prediction method and system for a fiber metal laminate, and relates to the technical field of health prediction of fiber metal laminates. A bilinear cohesion model is improved by combining a displacement memory effect and a hysteretic damage evolution rule and fusing an ROE cumulative damage theory; an improved Paris formula is used for establishing a metal plastic damage evolution criterion, and a secondary nominal stress criterion is used as a layered damage initial criterion; a double-position cohesion unit is combined with an improved bilinear cohesion model to simulate layered propagation and crack propagation behaviors, and a co-evolution framework capable of describing interaction of metal layer fatigue crack propagation and interlayer interface layered propagation is constructed; the genetic algorithm is coupled with the BP neural network, a fiber metal laminate fatigue crack propagation prediction model based on coupling of the genetic algorithm and the BP neural network is constructed, and prediction precision and generalization performance are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of fiber metal laminates health prediction, and particularly relates to a hybrid-driven fiber metal laminate fatigue crack propagation prediction method and system. BACKGROUND

[0002] Fiber metal laminates are a new interlaminar composite material prepared by alternately stacking metal thin plates and fiber reinforced composite material layers and curing by high temperature and high pressure process. The innovation of this heterogeneous composite material is that it not only inherits the excellent impact toughness, forming processability and corrosion resistance of light metals such as aluminum alloy, but also has high specific strength, high specific stiffness and excellent fatigue resistance of carbon fiber / aramid fiber reinforced system. Therefore, predicting the fatigue crack propagation rate is crucial for engineering safety and life assessment. If the fatigue crack propagation rate of fiber metal laminates can be predicted in time, not only the service reliability of key components under cyclic loading can be ensured, but also the service life of mechanical equipment can be prolonged, and the dual improvement of economic benefit and safety can be realized.

[0003] Current prediction models of fiber metal laminates under constant amplitude load mainly fall into three categories: phenomenological model, fracture mechanics model and finite element model. The phenomenological model relies on analytical algorithms to construct a parameterized analysis framework, which has significantly lower computational complexity than traditional multi-scale models. However, this model cannot well reflect the complex mechanism in the fatigue process of the laminate. The fracture mechanics model can quantitatively describe the stress field characteristics at the crack tip and accurately predict the crack propagation rate and direction. It is especially suitable for analyzing the interaction of cracks between metal layers and fiber layers. However, fracture mechanics model faces challenges such as multiple damage modes, material heterogeneity and complex load response in the analysis of fiber metal laminate fatigue damage. It is urgent to develop more universal multi-scale damage prediction models by combining numerical simulation and experimental characterization to improve the precision of fatigue behavior prediction of fiber metal laminates. The finite element model can accurately simulate the multi-material coupling effect, consider the plastic deformation of the metal layer and the anisotropic damage of the fiber composite material, and accurately predict crack initiation, propagation path and life. However, the finite element model still has some problems in practical application. The sensitivity of mesh division often leads to significant dispersion of calculation results, the oversimplification of material constitutive model easily leads to deviation from the real physical mechanism, the accurate application of complex boundary conditions is difficult, and the high consumption of time and computing resources by large-scale calculation significantly restricts the timeliness of its engineering application. In summary, although the three methods have their own advantages and disadvantages, with the development of finite element technology, numerical modeling has become an important means to reveal the evolution mechanism of layered damage.

[0004] At present, there are many studies on the fatigue crack propagation of fiber metal laminates, but there are still many challenges in the actual application of using finite element simulation method to study the fatigue crack propagation of fiber metal laminates. Xia Zhong-ping et al. studied the crack propagation problem of GLARE laminates under constant amplitude load. Cohesive elements are used to simulate the delamination phenomenon between aluminum layers and prepreg layers, and a constitutive model based on traction-separation viscous elements is used to simulate the delamination damage, and the equivalent stress intensity factor is calculated according to VCCT. Jia Xue et al. use COH3D8 cohesive element in ABAQUS to simulate the occurrence and expansion process of UACS crack and interlaminar interface delamination. The constitutive equation of the viscous element is established according to the relative displacement and traction of the interface, and the linear elastic traction-relative displacement law is introduced before delamination to maintain the stability of the two surfaces of the upper and lower adhesive layers on the interface, and finally the interface strength of delamination is defined by the maximum traction force index. Che Lu et al. established an analytical model considering the interface slip effect of fiber metal laminates, solved the interface slip strain of the fiber layer and the metal layer along the fiber direction, and simulated the interface slip effect by using SPRINGA element in ABAOUS. Through the simulation of the strain evolution process in the metal layer and the fiber layer during the solidification and cooling stage, the influence of the size of the laminated plate, the thickness of the metal layer and the transverse ratio of the part on the solidification deformation of the laminated plate is obtained. However, at present, the numerical simulation analysis of fiber metal laminates is relatively scarce. At the same time, in the process of finite element numerical analysis, there is a problem that the crack propagation behavior of the metal layer and the delamination expansion behavior of the interlayer under the action of fatigue load cannot be accurately coupled and characterized at the same time. SUMMARY

[0005] In view of the shortcomings of the prior art, the present application provides a hybrid driven fiber metal laminate fatigue crack propagation prediction method and system, which solves the problem that the prior art cannot accurately couple and characterize the crack propagation behavior of the metal layer and the delamination expansion behavior of the interlayer under the action of fatigue load at the same time, so as to further improve the prediction accuracy of the finite element high-fidelity model.

[0006] The technical scheme of the present application is as follows:

[0007] On the one hand, the present application provides a hybrid driven fiber metal laminate fatigue crack propagation prediction method, comprising the following steps:

[0008] A finite element model of the fiber metal laminate is established, and a double-position cohesive layer is set and the cohesive element is divided during the establishment process to obtain a double-position cohesive element; the double-position cohesive layer includes a zero-thickness cohesive layer created on the two contact surfaces of the metal layer and the fiber layer and two cohesive layers created on the two metal layers, which are used as crack propagation paths;

[0009] The improved bilinear cohesive force model is obtained by introducing a cumulative damage formula to improve the traditional bilinear cohesive force model, which is used to calculate the cumulative damage of each part of the fiber metal laminates to realize the fatigue damage evolution simulation of the multi-damage behavior of the fiber metal laminates; the multi-damage behavior includes metal layer crack propagation and interfacial delamination;

[0010] The improved Paris formula is introduced to establish the damage evolution criterion of the metal layer of the laminates, and the criteria in the fatigue damage evolution simulation of the multi-damage behavior of the fiber metal laminates include the crack propagation initiation criterion, the interfacial delamination damage initiation criterion and the interfacial delamination damage failure criterion;

[0011] Based on the criteria in the fatigue damage evolution simulation of the multi-damage behavior of the fiber metal laminates, the finite element model of the fiber metal laminates is simulated by using the double-position cohesive force unit combined with the improved bilinear cohesive force model and the improved Paris formula, and the metal layer fatigue crack propagation and the interfacial delamination propagation are considered in the simulation process to obtain the crack propagation length and the delamination shape;

[0012] A number of simulation data samples are obtained by setting different simulation conditions for the finite element model of the fiber metal laminates, and a number of real experiment samples are obtained, which together form a sample set, the sample set is divided into a training set, a validation set and a test set after preprocessing to obtain the preprocessed training set, the validation set and the test set; each simulation data sample or real experiment sample includes an initial crack length, a stress ratio, a cycle number, a load and a crack propagation length;

[0013] The weights and thresholds of the BP neural network are improved based on the preprocessed training set by using a genetic algorithm to obtain an improved BP neural network;

[0014] The improved BP neural network is trained using the training set to adjust the weights and thresholds, then the hyperparameters of the trained BP neural network are adjusted using the validation set, and finally the BP neural network is further evaluated and optimized to calculate the error, verify the generalization ability, obtain the final BP neural network and use it to obtain the predicted crack propagation length.

[0015] Further, the finite element model of the fiber metal laminates specifically comprises:

[0016] A1: a geometric model of the fiber metal laminates is established in a finite element analysis software according to the shape and size of the fiber metal laminates; the middle layer of the geometric model of the fiber metal laminates is a fiber layer, and the upper and lower layers are metal layers;

[0017] A2: a cohesive force layer is created on the geometric model of the fiber metal laminates;

[0018] A3: setting the cohesive force parameter and assigning the cohesive force parameter to the cohesive force layer; for the cohesive force layer on the metal layer and the two contact surfaces of the metal layer and the fiber layer, the cohesive force parameter is the initial stiffness, including the normal stiffness , the first shear direction stiffness , and the second shear direction stiffness ;

[0019] A4: setting the material parameters of the geometric model of the fiber metal laminate;

[0020] A5: meshing the metal layer, the fiber layer and the cohesive force layer in the geometric model of the fiber metal laminate, wherein the metal layer and the fiber layer are divided into a plurality of three-dimensional solid elements, and the cohesive force layer is divided into a plurality of cohesive force elements, to complete the construction of the finite element model.

[0021] Further, the cumulative damage formula introduced in the improved bilinear cohesive force model includes a fatigue damage calculation formula under cyclic loading and a softening damage calculation formula under monotonic loading, which is specifically:

[0022] The fatigue damage calculation formula under cyclic loading is:

[0023] (1);

[0024] In the formula, is the fatigue damage increment, is the fatigue damage variable, is the damage amplification coefficient, is the separation displacement value, is the separation displacement, is the separation displacement increment, is the initial critical displacement, is the material constant, is the current equivalent stress, is the Heaviside function, is the maximum cohesive force, is a constant representing the stress threshold;

[0025] (2);

[0026] (3);

[0027] wherein, is the maximum stress under the mixed mode, is the overall damage variable, is the critical stress at which failure occurs;

[0028] In formula (1), the Heaviside function i.e. when ;

[0029] Equivalent separation displacement and its increment are given by equations (4) and (5) as:

[0030] (4);

[0031] (5);

[0032] where, is the normal separation displacement component, is the tangential separation displacement component, is the accumulated separation displacement vector at time T, is the accumulated separation displacement vector at time is the time interval;

[0033] The softening damage calculation formula under monotonic loading is:

[0034] (6);

[0035] where, is the softening damage increment under monotonic loading, and are the critical displacement and the failure displacement, respectively;

[0036] The relationship between the fatigue damage under cyclic loading and the softening damage under monotonic loading is:

[0037] (7); where,

[0038] is the time. Further, the improved Paris formula for establishing the metal damage evolution criterion of the laminate is:

[0039]

[0040] (8); where,

[0041] is the crack length, is the cycle number, and are material parameters, is the energy release rate amplitude at the crack tip;

[0042] ​​​(9);

[0043] (10);

[0044] wherein, E is the elastic modulus of the material, σ is the stress ratio of the load;

[0045] The crack propagation initiation criterion is:

[0046] (11);

[0047] wherein, F is the fatigue failure index, , C is the fatigue crack propagation coefficient;

[0048] The quadratic nominal stress criterion is used as the interface delamination damage initiation criterion:

[0049] (12);

[0050] wherein, , and are the normal cohesive force, the first shear direction cohesive force and the second shear direction cohesive force, respectively; , , are the normal damage initiation critical stress, the first shear direction damage initiation critical stress and the second shear direction damage initiation critical stress, respectively, and the calculation method is as follows:

[0051] (13);

[0052] (14);

[0053] (15);

[0054] wherein, σn is the normal cohesive strength, σt is the tangential cohesive strength;

[0055] The B-K criterion is used as the interface delamination damage failure criterion;

[0056] (16);

[0057] wherein, Gc is the critical fracture energy, Gnc is the normal fracture energy, Gsc is the shear fracture energy, Gt is the sum of the tangential fracture energy, and Gs G represents the sliding shear energy release rate. t To achieve the shear energy release rate, G T This refers to the total energy release rate; For material-related coefficients; when The BK criterion is triggered.

[0058] Furthermore, the criterion in the fatigue damage evolution simulation based on the multi-damage behavior of fiber-reinforced metal laminates is used to perform finite element simulation of the fiber-reinforced metal laminate model using an improved bilinear cohesive force model and an improved Paris formula, specifically as follows:

[0059] B1: Defines the analysis step for finite element simulation of the finite element model of the fiber-reinforced metal laminate;

[0060] B2: Set the loads for the finite element model of the fiber-reinforced metal laminate;

[0061] B3: Sets the material state variables, critical stress, fracture energy, BK criterion coefficient, damage amplification factor, and constants characterizing the stress threshold;

[0062] B4: Set the number of the current cohesive unit to b=0;

[0063] B5: When the current cohesive unit satisfies the interface delamination damage initiation criterion, damage evolution simulation is performed using the improved bilinear cohesive model. During the process, it is monitored in real time whether the crack propagation initiation criterion or the interface delamination damage failure criterion is reached. If the crack propagation initiation criterion is reached, B6 is executed. If the interface delamination damage failure criterion is reached, the current cohesive unit is deleted and the improved bilinear cohesive model is used to continue the damage evolution simulation.

[0064] B6: After the crack propagation initiation criterion is met, the improved Paris formula is used to simulate the fatigue crack propagation of the metal layer, and finally the crack propagation length and delamination shape are obtained.

[0065] Furthermore, the damage evolution simulation using the improved bilinear cohesion model specifically includes:

[0066] C1: For the current cohesive element, calculate the critical displacement and failure displacement based on the set critical stress, initial stiffness and fracture energy;

[0067] (17);

[0068] in, This represents the critical displacement in the normal direction. This represents the critical displacement in the first shear direction. This represents the critical displacement in the second shear direction.

[0069] (18);

[0070] wherein, is the normal direction failure displacement, is the first shear direction failure displacement, is the second shear direction failure displacement;

[0071] C2: obtaining the relative displacement of the cohesive element at the current time, and calculating the displacement change amount in the mixed mode according to the state variable at the current time ; the relative displacement at the current time includes: normal direction relative displacement , first shear direction relative displacement , second shear direction relative displacement ;

[0072] The calculation method of the displacement change amount in the mixed mode is:

[0073] (19);

[0074] C3: calculating the relative displacement at the next time according to the displacement change amount in the mixed mode, and further obtaining the displacement change amount at the next time in the mixed mode ;

[0075] (20);

[0076] wherein, is the normal direction relative displacement at the next time, is the first shear direction relative displacement at the next time, is the second shear direction relative displacement at the next time;

[0077] C4: calculating the initial critical displacement and failure displacement in the mixed mode according to the relative displacement at the next time, the critical displacement, the initial stiffness and the fracture energy;

[0078] (21);

[0079] (22);

[0080] (23);

[0081] wherein, is the mixed loading ratio, is the initial critical displacement in the mixed mode, is the failure displacement in the mixed mode;

[0082] C5: the displacement change amount under the mixed mode , the displacement change amount at the next time under the mixed mode , the initial critical displacement and the failure displacement under the mixed mode, the damage accumulation amount under the monotonic loading at the next time is calculated ;

[0083] (24);

[0084] wherein, the damage accumulation amount under the monotonic loading at the next time;

[0085] C6: the damage accumulation amount under the cyclic loading at the next time is calculated according to the damage amplification coefficient and the constant representing the stress threshold ;

[0086] (25);

[0087] wherein, the damage accumulation amount under the cyclic loading at the next time, the accumulated damage, the normal direction stress, the first shear direction stress, the second shear direction stress;

[0088] C7: the damage accumulation amount under the monotonic loading at the next time is added to the damage accumulation amount under the cyclic loading at the next time , to obtain the accumulated damage of the cohesive element at the next time ;

[0089] C8: the stress and the Jacobian matrix of the cohesive element after the accumulated damage of the cohesive element at the next time is updated are calculated;

[0090] (26);

[0091] (27);

[0092] wherein, the updated normal direction stress, the updated first shear direction stress, the updated second shear direction stress, the updated normal direction relative displacement, the updated first shear direction relative displacement, the updated second shear direction relative displacement;

[0093] C9: judge the accumulated damage of the current cohesive element whether the set threshold is reached, if yes, the stiffness of the cohesive element is degraded to 0, the cohesive element is failed, C10 is executed, and the fatigue damage evolution simulation process of one cohesive element is ended, otherwise, C2 is returned;

[0094] C10: the number of the current cohesive element is added by 1 and C2 is returned.

[0095] In another aspect, the present application also provides a hybrid driven fiber metal laminates fatigue crack propagation prediction system for implementing a hybrid driven fiber metal laminates fatigue crack propagation prediction method, comprising:

[0096] a finite element model construction module for establishing a finite element model of the fiber metal laminates;

[0097] a cohesive model construction module for constructing an improved bilinear cohesive model;

[0098] a Paris formula construction and criterion setting module for constructing an improved Paris formula and setting a criterion in the fatigue damage evolution simulation of the multi-damage behavior of the fiber metal laminates;

[0099] a simulation module for performing finite element simulation on the finite element model of the fiber metal laminates based on the criterion in the fatigue damage evolution simulation of the multi-damage behavior of the fiber metal laminates, using a double-position cohesive element combined with the improved bilinear cohesive model and the improved Paris formula, considering the fatigue crack propagation of the metal layer and the delamination propagation of the interlayer interface during the simulation process, and obtaining the crack propagation length and the delamination shape;

[0100] a data set construction module for setting different simulation conditions and obtaining a plurality of simulation data samples, simultaneously obtaining a plurality of real experiment samples, collectively constituting a sample set, dividing the sample set into a training set, a validation set and a test set, and preprocessing the training set, the validation set and the test set after the division;

[0101] a genetic algorithm module for improving the BP neural network based on the preprocessed training set using a genetic algorithm, and obtaining an improved BP neural network;

[0102] a fine-tuning module for training the improved BP neural network using the training set, adjusting the weights and thresholds, then adjusting the hyperparameters of the trained BP neural network using the validation set, and finally further evaluating and optimizing the calculation error of the BP neural network using the test set, verifying the generalization ability, and obtaining the final BP neural network;

[0103] a prediction module for obtaining the predicted crack propagation length using the final BP neural network.

[0104] In a third aspect, the present application provides an electronic device, comprising: one or more processors, and a memory for storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the fatigue crack propagation prediction method of the hybrid-driven fiber metal laminates.

[0105] In a fourth aspect, the present application provides a computer-readable storage medium storing executable instructions that, when executed, cause a processor to perform the fatigue crack propagation prediction method of the hybrid-driven fiber metal laminates.

[0106] In a fifth aspect, the present application provides a computer program product comprising a computer program or instructions that, when executed by a processor, implement the fatigue crack propagation prediction method of the hybrid-driven fiber metal laminates.

[0107] Compared with the prior art, the present application has the following beneficial effects:

[0108] (1) In view of the defects of the traditional cohesive force model, such as the lack of applicability of the constitutive framework under fatigue load conditions, and the inability of the energy dissipation mechanism to meet the representation requirements of damage evolution under cyclic load, the present application improves the traditional bilinear cohesive force model to realize the gradual stiffness degradation of the laminate multi-damage under constant amplitude cyclic load. By introducing a damage amplification factor, the equivalent reduction of the cycle number is realized, and the calculation efficiency is improved to several times of the traditional method while maintaining the accuracy of the physical mechanism.

[0109] (2) The core damage modes of fiber metal laminates are metal layer cracks and interfacial delamination, and the collaborative representation of the two is the key to research. The present application first introduces an improved Paris theory, and establishes a metal plastic damage evolution criterion based on the improved Paris theory with the energy dissipation principle; then, the initiation of interfacial delamination is judged by a quadratic nominal stress criterion, and a double-cohesive element is used to simulate multi-damage modes by combining an improved bilinear cohesive force model. Finally, a collaborative evolution framework is constructed to realize the collaborative representation of fatigue crack and interfacial delamination propagation.

[0110] (3) A fatigue crack propagation prediction model for fiber metal laminates based on genetic algorithm and BP neural network is constructed. The BP neural network accurately fits the complex nonlinear mapping relationship of the related parameters by avoiding the idealization of the input parameters of the mechanism model, the dispersion of the material performance, and other limitations through the powerful nonlinear fitting capability; and by means of the powerful global search capability of the genetic algorithm, combined with the cross variation operator to carry out multi-path parallel search in the weight threshold space, the loss of excellent solutions in the training process is avoided. BRIEF DESCRIPTION OF DRAWINGS

[0111] Figure 1A schematic diagram of a geometric model of a fiber metal laminate in an embodiment of the present application;

[0112] Figure 2 A schematic diagram of meshing of a fiber metal laminate in an embodiment of the present application;

[0113] Wherein, (a) is a meshing diagram as a whole; (b) is a central partial enlarged view;

[0114] Figure 3 A schematic diagram of a loading process of an improved bilinear cohesive force model in an embodiment of the present application;

[0115] Figure 4 A flowchart of a genetic algorithm in an embodiment of the present application;

[0116] Figure 5 A center crack tensile M(T) test piece of a laminate in an embodiment of the present application;

[0117] Figure 6 Crack propagation data of a fiber metal laminate at a stress ratio of 0.06 in an embodiment of the present application;

[0118] Figure 7 Crack propagation data of a fiber metal laminate at a stress ratio of -1 in an embodiment of the present application;

[0119] Figure 8 A delamination size curve of a fiber metal laminate at different crack lengths in an embodiment of the present application;

[0120] Wherein, (a) is a crack length of 16 mm; (b) is a crack length of 20 mm;

[0121] Figure 9 A fatigue delamination propagation diagram of a fiber metal laminate in an embodiment of the present application;

[0122] (a) DIC delamination image; (b) delamination form of a fiber metal laminate;

[0123] Figure 10 A stress cloud diagram under a 60 MPa load in an embodiment of the present application;

[0124] Wherein, (a) is a cycle of 60000 times; (b) is a cycle of 80000 times;

[0125] Figure 11 A crack propagation diagram under a 60 MPa load in an embodiment of the present application;

[0126] Wherein, (a) is a cycle of 60000 times; (b) is a cycle of 80000 times;

[0127] Figure 12 A stress cloud diagram under a 70 MPa load in an embodiment of the present application;

[0128] Wherein, (a) is the figure of 40000 cycles; (b) is the figure of 60000 cycles;

[0129] Figure 13 It is the crack propagation figure under 70MPa load in the embodiment of the application;

[0130] Wherein, (a) is the figure of 40000 cycles; (b) is the figure of 60000 cycles;

[0131] Figure 14 The crack propagation simulation result and test data comparison figure;

[0132] Figure 15 It is the crack propagation process figure of initial crack length 10mm in the embodiment of the application;

[0133] Wherein, (a) is the figure of 50000 cycles; (b) is the figure of 100000 cycles; (c) is the figure of 150000 cycles; (d) is the figure of 200000 cycles;

[0134] Figure 16 It is the layered propagation process figure of initial crack length 10mm;

[0135] Wherein, (a) is the figure of 50000 cycles; (b) is the figure of 100000 cycles; (c) is the figure of 150000 cycles; (d) is the figure of 200000 cycles;

[0136] Figure 17 It is the simulation and test comparison figure of initial crack length 10mm in the embodiment of the application;

[0137] Figure 18 It is the simulation and test comparison figure of initial crack length 15mm in the embodiment of the application;

[0138] Figure 19 It is the layered size curve of fiber metal laminates under different crack lengths;

[0139] Wherein, (a) is the crack length 16mm; (b) is the crack length 20mm;

[0140] Figure 20 It is the BP neural network training iteration process and training regression curve figure;

[0141] Wherein, (a) is the iteration process figure; (b) is the training regression curve figure;

[0142] Figure 21 It is the GA-BP neural network training iteration process and training regression curve in the embodiment of the application;

[0143] Wherein, (a) is the iterative process diagram; (b) is the training regression curve diagram;

[0144] Figure 22 This is a comparison chart of the initial crack 10mm test, BP prediction, and GA-BP prediction results in an embodiment of the present invention;

[0145] Figure 23 This is a comparison chart of the initial crack 15mm test, BP prediction, and GA-BP prediction results in an embodiment of the present invention. Detailed Implementation

[0146] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0147] Example 1:

[0148] A hybrid-driven method for predicting fatigue crack propagation in fiber-reinforced metal laminates is proposed. First, based on the cohesive model, a bilinear cohesive model for characterizing damage behavior under fatigue loading is improved by incorporating the displacement memory effect and hysteretic damage evolution rule, and integrating the ROE cumulative damage theory. Then, an improved method based on the classical Paris theory of energy dissipation is introduced, using the improved Paris formula to establish a metal plastic damage evolution criterion. For the determination of interface damage initiation, the quadratic nominal stress criterion is adopted as the criterion for delamination damage initiation. Combining the improved bilinear cohesive model with simulation of delamination and crack propagation behavior, a co-evolutionary framework describing the interaction between fatigue crack propagation in metal layers and delamination propagation at the interlayer interface is constructed, achieving dynamic coupling simulation of the two damage modes under cyclic loading. Finally, a genetic algorithm and a backpropagation neural network are coupled to construct a fatigue crack propagation prediction model for fiber-reinforced metal laminates based on this coupling, significantly improving prediction accuracy and generalization performance. The specific steps include:

[0149] Step 1: Establish the finite element model of the fiber-reinforced metal laminate;

[0150] Step 1.1: Based on the shape and size of the fiber-metal laminate, establish a geometric model of the fiber-metal laminate in the finite element analysis software; the middle layer of the geometric model of the fiber-metal laminate is the fiber layer, and the upper and lower layers are metal layers;

[0151] The purpose of creating a geometric model of a fiber-reinforced metal laminate in ABAQUS is to provide a solid foundation for finite element analysis. By accurately constructing the shape and size of the research object (fiber-reinforced metal laminate), the accuracy of subsequent mesh generation, material definition, and boundary condition application is ensured, which directly affects the reliability of simulation results and computational efficiency.

[0152] Therefore, in the embodiment, a deformable cube with a length of 270 mm, a width of 75 mm and a thickness of 5.5 mm is established in ABAQUS. The cube is divided into three layers according to 2 mm, 1.5 mm and 2 mm, and the middle layer with a thickness of 1.5 mm is used as a fiber layer. Since the fiber layer hardly suffers damage during loading, the entire fiber layer is considered as a whole. The other two layers are used as metal layers. Figure 1 To establish the component diagram and structure division, the established components are assembled in ABAQUS.

[0153] Step 1.2: Create a cohesive layer on the geometric model of the fiber metal laminate.

[0154] Specifically, a zero-thickness cohesive layer is created on each of the two contact surfaces of the metal layer and the fiber layer, and two cohesive layers are created on the two metal layers as crack propagation paths.

[0155] Step 1.3: Set the cohesive parameters and assign the cohesive parameters to the cohesive layer. For the cohesive layers on the metal layers and the two contact surfaces of the metal layer and the fiber layer, the initial stiffness of the cohesive parameters includes the normal stiffness , the first shear direction stiffness and the second shear direction stiffness .

[0156] In this embodiment, the initial stiffness of the cohesive layer on the metal layer is defined as 7.24×104MPa / mm; the initial stiffness of the zero-thickness cohesive layer between the metal layer and the fiber layer is set as 1×105MPa / mm, which is used to determine the elastic response when there is no damage.

[0157] Step 1.4: Set the material parameters of the geometric model of the fiber metal laminate.

[0158] The function of setting the material parameters in ABAQUS is to define the mechanical behavior of the material, including elasticity, plasticity, damage and other characteristics, to ensure that the simulation accurately reflects the physical response of the real material and provides a reliable constitutive model basis for subsequent analysis.

[0159] Step 1.5: Mesh the metal layer, the fiber layer and the cohesive layer in the geometric model of the fiber metal laminate, wherein the metal layer and the fiber layer are divided into a plurality of three-dimensional solid elements, and the cohesive layer is divided into a plurality of cohesive elements, to complete the construction of the finite element model.

[0160] The meshing in this embodiment discretizes the geometric model into finite elements, and calculating a reasonable mesh density and type (such as structured / free mesh) can balance the solution accuracy and the calculation cost. In order to obtain a better simulation effect, the mesh of the central part of the fiber metal plate is appropriately encrypted, and the mesh on the crack propagation path and the delamination expansion area is encrypted, and a relatively coarse mesh is selected at other positions in order to improve the calculation efficiency. The metal layer and the fiber layer adopt the three-dimensional solid element C3D8R in ABAQUS, and the cohesive layer adopts COH3D8. A total of 107534 nodes and 87798 units are arranged, including 52542 C3D8R units and 35256 COH3D8 units. The meshing is as shown in Figure 2 .

[0161] Step 2: The traditional cohesive zone model cannot accurately represent the energy damage accumulation behavior under fatigue load conditions, and the fatigue failure mechanism of the fiber metal laminate is analyzed. The traditional bilinear cohesive zone model is improved by introducing a cumulative damage formula to calculate the cumulative damage of each part of the fiber metal laminate, so as to realize the fatigue damage evolution simulation of the multi-damage behavior of the fiber metal laminate. The multi-damage behavior includes crack propagation in the metal layer and delamination at the interface between the layers.

[0162] In order to improve the calculation efficiency of the cohesive zone model in simulation, the fatigue damage accumulation is realized through the displacement memory effect and the hysteretic damage rule of the cohesive zone model itself and the cumulative damage evolution equation. The damage amplification coefficient is introduced in the damage evolution equation to realize the equivalent reduction of the cycle number, and the calculation efficiency is improved to several orders of magnitude of the traditional method while maintaining the accuracy of the physical mechanism. The fatigue damage calculation formula under cyclic loading and the softening damage calculation formula under monotonic loading in the cumulative damage formula, and the cohesive zone loading process are as shown in Figure 3 .

[0163] The fatigue damage calculation formula under cyclic loading is:

[0164] (1);

[0165] In the formula, is the fatigue damage increment, is the fatigue damage variable, is the damage amplification coefficient, is the separation displacement value, is the separation displacement, is the separation displacement increment, is the initial critical displacement, is a material constant, which is usually set to several times of , is the current equivalent stress, is the Heaviside function, The maximum cohesive force is given by Equation 2. The constant characterizing the stress threshold is given by Equation 3;

[0166] (2);

[0167] (3);

[0168] in, The maximum stress in the mixed mode, As a variable representing overall damage, The critical stress at which failure occurs;

[0169] In formula (1), the Heaviside function That is, when hour ;

[0170] Equivalent separation displacement and its increment The formula is given by equations (4) and (5) as follows:

[0171] (4);

[0172] (5);

[0173] in, For the normal separation displacement components, To separate the tangential displacement components, Let be the cumulative separation displacement vector at time T. For a moment The cumulative separation displacement vector at time t, For time intervals;

[0174] The formula for calculating softening damage under monotonic load is:

[0175] (6);

[0176] In the formula, This represents the softening damage increment under monotonic loading. and These are the critical displacement and the failure displacement, respectively.

[0177] Roe and Siegmund further clarified two damage mechanisms—fatigue damage and softening damage under monotonic loading—and cumulative damage. The relationship between them is as follows:

[0178] (7);

[0179] where, is time; the model takes the cohesive element damage in the first loading stage as the linear superposition of softening damage increment and fatigue damage increment when describing the evolution of cumulative damage, and then ignores the material softening effect and only considers the cumulative of fatigue damage component.

[0180] In this embodiment, an improved bilinear cohesive force model is constructed by using the user subroutine development framework and Fortran language, and the change of the cohesive force element in the fiber metal laminate model is analyzed, thereby providing an efficient numerical tool for the subsequent study on the fatigue analysis of the fiber metal laminate structure.

[0181] Step 3: By introducing the improved Paris formula, the crack tip energy release rate amplitude is combined with the traditional Paris formula, and the criteria in the fatigue damage evolution simulation of the multi-damage behavior of the fiber metal laminate are set, including the crack propagation initiation criterion, the interface delamination damage initiation criterion and the interface delamination damage failure criterion, so as to cooperatively characterize the crack propagation and delamination propagation.

[0182] In the study of the fatigue crack propagation of the fiber metal laminate, the crack propagation of the metal layer is affected by the cyclic plasticity accumulation and presents a discontinuous characteristic, and the interface delamination behavior depends on the interface toughness and the interlaminar stress state, and the evolution mechanisms of the two are essentially different. Therefore, there is a problem that the two damage modes of the metal layer crack and the interlaminar interface delamination in the fatigue crack propagation of the fiber metal laminate are difficult to cooperatively characterize.

[0183] To solve the problem, the improved Paris formula is introduced to establish the metal plastic damage evolution criterion to characterize the damage evolution characteristics of the laminate metal, to judge whether the metal layer crack propagates or not by using the fatigue failure index, and to judge whether the delamination evolves or not by using the secondary nominal stress criterion as the damage initiation criterion of the interlaminar delamination.

[0184] The improved Paris formula combines the crack tip energy release rate amplitude with the traditional Paris formula, thereby forming an improved Paris formula system based on the energy dissipation principle:

[0185] (8) ;

[0186] In the formula, is the crack length, is the cycle number, and are material parameters, is the energy release rate amplitude of the crack tip; in order to apply the material parameters to the Paris formula based on the energy release rate, the metal layer material parameters and The conversion is made to obtain and It can be obtained by the following formula:

[0187] (9);

[0188] (10);

[0189] wherein, is the elastic modulus of the material, is the stress ratio of the load;

[0190] The crack propagation initiation criterion is:

[0191] (11);

[0192] wherein, is the fatigue failure index, , is the fatigue crack propagation coefficient;

[0193] The quadratic nominal stress criterion is used as the interface delamination damage initiation criterion:

[0194] (12);

[0195] In the formula, , and are the normal cohesive force, the cohesive force in the first shear direction and the cohesive force in the second shear direction, respectively; , , are the normal damage initiation critical stress, the damage initiation critical stress in the first shear direction and the damage initiation critical stress in the second shear direction, respectively, and the calculation method is as follows:

[0196] (13);

[0197] (14);

[0198] (15);

[0199] wherein, is the normal cohesive strength, is the tangential cohesive strength;

[0200] The B-K criterion is used as the interface delamination damage failure criterion;

[0201] (16);

[0202] In the formula, is the critical fracture energy, is the normal fracture energy, is the shear fracture energy, is the sum of tangential fracture energy, G s is the sliding shear energy release rate, G t is the tearing shear energy release rate, G T is the total energy release rate; is a coefficient related to the material; when , the B-K criterion is triggered;

[0203] Step 4: Based on the criterion in the fatigue damage evolution simulation of the multi-damage behavior of fiber metal laminates, a double-position cohesive element is used in combination with an improved bilinear cohesive model and an improved Paris formula to perform finite element simulation on the finite element model of the fiber metal laminate. In the simulation process, the fatigue crack propagation of the metal layer and the delamination propagation of the interfacial layer are considered simultaneously to obtain the crack propagation length and delamination shape.

[0204] On the basis of the improved Paris formula, a double-position cohesive element is used in combination with an improved cohesive model to construct a co-evolution framework to realize the multi-damage coupling representation of the laminate and the unified description and prediction of the coupling behavior of the two key damage processes of fatigue crack propagation and interfacial delamination propagation, effectively solving the challenge of multi-damage co-representation.

[0205] Specifically, the following steps are included:

[0206] Step 4.1: Defining the analysis step for finite element simulation of the finite element model of the fiber metal laminate;

[0207] In finite element analysis, the core function of defining the analysis step is to divide a complex physical process into a series of sequentially executed and controllable calculation stages. This step-by-step processing is the key to accurately simulating path-dependent behavior (such as material yield or contact separation) and effectively controlling the accuracy and efficiency of the solution.

[0208] The defined analysis step is specifically: creating an analysis step, determining the increment step size, the maximum increment step number and the maximum cycle number to determine the evolution process of the load, boundary conditions and the like over time;

[0209] Step 4.2: Setting the load of the finite element model of the fiber metal laminate;

[0210] In ABAQUS, load setting is used to simulate the mechanical environment (such as force, pressure, displacement boundary, etc.) under real working conditions, ensuring that the stress state of the model is consistent with the actual situation; therefore, the load setting is fixed at one end and all constraints are applied, and a cyclic load is applied at the other end. The load is a sinusoidal curve with a frequency of 10HZ;

[0211] Step 4.3: Set the material state variable, critical stress, fracture energy, B-K criterion coefficient, damage amplification factor and constant of stress threshold;

[0212] The material state variable is used to describe the internal state change; the critical stress is used to set the damage initiation condition; the fracture energy is used to control the crack propagation energy dissipation to avoid mesh sensitivity; the B-K criterion coefficient is used to adjust the I / II type damage ratio in mixed mode fracture; the damage amplification factor is used to accelerate or slow down the damage evolution to optimize the calculation efficiency; and the constant of stress threshold is used to determine the damage localization starting position to improve the failure prediction accuracy. These parameters together ensure the physical authenticity and numerical stability of material damage evolution, and are suitable for complex simulations such as composite failure and dynamic fracture.

[0213] In this embodiment, the following are defined in ABAQUS: the critical stress includes the critical stress in the normal direction , the critical stress in the first shear direction , and the critical stress in the second shear direction ; the fracture energy includes the normal fracture energy , the first shear direction fracture energy , and the second shear direction fracture energy ; the B-K criterion coefficient is ; the damage amplification factor is ; and the constant of stress threshold is ;

[0214] Step 4.4: Set the number of the current cohesive element b = 0;

[0215] Step 4.5: When the current cohesive element satisfies the interface delamination damage initiation criterion, use the improved bilinear cohesive model for damage evolution simulation, and real-time monitor whether the crack propagation initiation criterion or the interface delamination damage failure criterion is reached. If the crack propagation initiation criterion is reached, execute step 4.6; if the interface delamination damage failure criterion is reached, delete the current cohesive element and continue to use the improved bilinear cohesive model for damage evolution simulation;

[0216] Step 4.5.1: For the current cohesive element, calculate the critical displacement and the failure displacement according to the set critical stress, initial stiffness and fracture energy;

[0217] (17) ;

[0218] wherein, is the critical displacement in the normal direction, is the critical displacement in the first shear direction, is the critical displacement in the second shear direction;

[0219] (18);

[0220] wherein, is the normal direction failure displacement, is the first shear direction failure displacement, is the second shear direction failure displacement;

[0221] Step 4.5.2: Obtain the relative displacement of the cohesive element at the current time, and calculate the displacement change amount in the mixed mode according to the state variable at the current time ; the relative displacement at the current time includes: normal direction relative displacement , first shear direction relative displacement , second shear direction relative displacement ;

[0222] The calculation method of the displacement change amount in the mixed mode is:

[0223] (19);

[0224] Step 4.5.3: Calculate the relative displacement at the next time according to the displacement change amount in the mixed mode, and then obtain the displacement change amount at the next time in the mixed mode ;

[0225] (20);

[0226] wherein, is the normal direction relative displacement at the next time, is the first shear direction relative displacement at the next time, is the second shear direction relative displacement at the next time;

[0227] Step 4.5.4: Calculate the initial critical displacement and failure displacement in the mixed mode according to the relative displacement at the next time, the critical displacement, the initial stiffness and the fracture energy;

[0228] (21);

[0229] (22);

[0230] (23);

[0231] wherein, is the mixed loading ratio, is the initial critical displacement in the mixed mode, is the failure displacement in the mixed mode;

[0232] Step 4.5.5: Damage increment is able to quantify the damage evolution degree of each calculation step, drive the material stiffness degradation step by step, ensure the continuity and stability of the damage process, and avoid numerical mutation. Therefore, according to the displacement change amount of the mixed mode , the displacement change amount of the next time of the mixed mode , the initial critical displacement and the failure displacement of the mixed mode, the damage accumulation of the next time under monotonic loading is calculated ;

[0233] (24);

[0234] wherein, is the damage accumulation of the next time under monotonic loading;

[0235] Step 4.5.6: According to the damage amplification coefficient and the constant representing the stress threshold, the damage accumulation of the next time under cyclic loading is calculated ;

[0236] (25);

[0237] wherein, is the damage accumulation of the next time under cyclic loading, is the accumulated damage, is the normal direction stress, is the first shear direction stress, is the second shear direction stress;

[0238] Step 4.5.7: The damage accumulation of the next time under monotonic loading is added to the damage accumulation of the next time under cyclic loading to obtain the accumulated damage of the cohesive element in the next time ;

[0239] Step 4.5.8: The stress and Jacobian matrix of the cohesive element after updating the accumulated damage of the cohesive element in the next time become effective tools for calculating the fatigue crack propagation prediction of fiber metal laminates;

[0240] (26);

[0241] (27);

[0242] wherein, is the updated normal direction stress, is the updated first shear direction stress, is the updated second shear direction stress, the updated normal direction relative displacement, the updated first shear direction relative displacement, the updated second shear direction relative displacement;

[0243] In the ABAQUS subroutine, the updated state variables (normal direction relative displacement, first shear direction relative displacement, second shear direction relative displacement, normal direction stress, first shear direction stress, second shear direction stress and damage factor) are used to record the evolution history of material nonlinear behavior (such as damage, plasticity, fracture, etc.) in real time, ensure path dependence, drive improved cohesive zone model updating, and provide key criterion data for failure analysis, while supporting multi-field coupling and post-processing visualization;

[0244] Step 4.5.9: Determine whether the cumulative damage of the current cohesive element reaches the set threshold value If yes, the stiffness of the cohesive element is degraded to 0, the cohesive element fails, step 4.5.10 is executed, and the fatigue damage evolution simulation process of one cohesive element is ended, otherwise, return to step 4.5.2;

[0245] Step 4.5.10: The number of the current cohesive element is incremented by 1 and returns to step 4.5.2;

[0246] Step 4.6: After reaching the crack propagation initiation criterion, the improved Paris formula is used to simulate the fatigue crack propagation of the metal layer, and finally the crack propagation length and delamination shape are obtained;

[0247] Step 5: Set different simulation conditions for the finite element model of the fiber metal laminates and obtain several simulation data samples according to the method of step 4, and obtain several real experiment samples, which together form a sample set. After dividing the sample set into training set, validation set and test set, pre-processing is performed to obtain pre-processed training set, validation set and test set; each simulation data sample or real experiment sample includes initial crack length, stress ratio, cycle number, load and crack propagation length;

[0248] In this embodiment, the improved cohesive zone model is used for simulation to obtain the crack propagation length and delamination shape of the fiber metal laminates under different stress levels of 50, 60, 70, 80 and 90, with initial crack lengths of 10 mm and 15 mm, and stress ratios of 0, 0.06, 0.1, 0.3 and 0.5 for each crack length. The test data has two initial crack lengths of 10 mm and 15 mm, a stress ratio of 0.06, and two stress levels of 60 and 70;

[0249] The sample set is divided to accurately evaluate the individual fitness, prevent overfitting, and ensure the generalization ability of the final model. The validation set guides the selection of the genetic algorithm, and the test set independently verifies the optimal performance.

[0250] The preprocessing is normalization:

[0251] Accelerate the convergence of neural network training, avoid certain features due to dimensional differences leading to genetic algorithm fitness evaluation;

[0252] (28);

[0253] Wherein, x is the original input data value, x min is the minimum value in the whole sample set, x max is the maximum value in the whole sample set, and x' is the output value.

[0254] Step 6: Based on the preprocessed training set, the genetic algorithm is used to improve the BP neural network, and the improved BP neural network is obtained.

[0255] In the modeling research of fatigue crack propagation behavior of fiber metal laminates, the traditional finite element simulation method still has inherent limitations in the research of fatigue crack propagation of fiber metal laminates. The specific performance is: the calculation results discreteness caused by grid dependence, the difference between the material constitutive model and the actual damage dispersion, the energy dissipation representation error caused by the simplification of boundary conditions, and the calculation efficiency bottleneck under complex cyclic load.

[0256] In view of the above challenges, the present application proposes an intelligent modeling framework based on artificial neural network, which realizes the accurate prediction of fatigue damage evolution by fusing the deep law of simulation data and test data. A fatigue crack propagation model for predicting fiber metal laminates is established based on the genetic algorithm and back propagation coupled optimization (GA-BP) framework. The present application uses the powerful nonlinear fitting capability of BP neural network to avoid the idealization of mechanism model input parameters, material performance dispersion and other limitations, and accurately fits the complex nonlinear mapping relationship of related parameters; genetic algorithm is selected as the improvement method, which is essentially to realize multi-path parallel search of parameter space by simulating biological evolution mechanism, and has global exploration ability and adaptive search strategy. By combining genetic algorithm and BP neural network to construct a hybrid optimization framework, the purpose is to break through the local optimal limit of single gradient descent algorithm, and significantly improve the prediction accuracy and generalization performance of the model. The genetic algorithm flow chart is shown in Figure 4 ;

[0257] Step 6.1: Determine the number of hidden layer nodes, balance the learning ability and generalization performance of the network, avoid overfitting or underfitting, and make the genetic algorithm can efficiently optimize the best network structure to adapt to the task requirements. The formula is shown in formula 29:

[0258] (29);

[0259] wherein, , , are the number of nodes of the hidden layer, input layer and output layer respectively, and constant is in the range of 1 to 10;

[0260] Step 6.2: generate a number of chromosomes to form an initial population, wherein each chromosome represents a set of weights and thresholds of a BP neural network, and each gene is a single weight or threshold of the BP neural network;

[0261] Step 6.3: encode the chromosomes by using a real number coding strategy;

[0262] The calculation formula of the length of the encoded chromosome is:

[0263] S=pk+kn+k+n (30);

[0264] wherein, S is the length of the encoded chromosome, and p, k, n are the number of neurons of the input layer, hidden layer and output layer respectively, which are 4, 10 and 1 respectively in this embodiment;

[0265] Step 6.4: set the current iteration number to 0;

[0266] Step 6.5: based on the multi-objective optimization framework, quantitatively evaluate the candidate solutions by defining the fitness index, train the BP neural network corresponding to each chromosome by using the training set, and obtain the root mean square error as the value of the fitness function;

[0267] Step 6.6: according to the value of the fitness function of the chromosome, perform a selection operation on the current chromosome, and retain the chromosomes with a fitness function value higher than a set threshold through a competitive screening mechanism, and then directly retain the selected excellent chromosomes to the next generation population;

[0268] Step 6.7: perform a crossover operation on the non-excellent chromosomes through a linear combination mechanism to obtain the chromosomes after the crossover;

[0269] (31);

[0270] (32);

[0271] wherein, is the iteration number, and are the numbers of chromosomes, and are the chromosomes before the crossover, and the chromosomes after the crossover, a crossover weight factor;

[0272] Step 6.8: the chromosomes after the crossover are subjected to mutation operation by using a site-independent mutation strategy to obtain mutated chromosomes;

[0273] In this embodiment, each gene site is randomly triggered by a preset mutation probability to select a value to replace the original value;

[0274] Step 6.9: the mutated chromosomes and the excellent chromosomes are used to form a new population;

[0275] Step 6.10: it is judged whether the maximum iteration number L is reached, if yes, the chromosome with the minimum value of the fitness function is output, otherwise, the iteration number is increased by 1 and the step 6.5 is returned;

[0276] The iteration mechanism promotes the global search process while maintaining population diversity; after completing the generation evolution, the population is subjected to fitness evaluation, the optimal parameter combination is selected, and the optimal individual is output. The iteration evolution of the population is realized through selection, crossover, mutation and other basic operations, and the candidate solution is gradually approximated to the optimal solution space based on the gene recombination mechanism. After the genetic algorithm optimization, the local optimal limit of the single gradient descent algorithm is broken through;

[0277] Step 6.11: the chromosome with the minimum value of the fitness function is decoded to obtain an optimal set of weights and thresholds of the BP neural network;

[0278] Step 6.12: the optimal set of weights and thresholds is assigned to the BP neural network to obtain an improved BP neural network;

[0279] Step 7: the improved BP neural network is trained by using the training set to adjust the weights and thresholds, then the hyperparameters of the trained BP neural network are adjusted by using the validation set, finally the BP neural network is further evaluated and optimized to calculate the error by using the test set, the generalization ability is verified, the final BP neural network is obtained, and the predicted crack propagation length is obtained by using the final BP neural network;

[0280] The forward propagation process of the BP neural network is as follows:

[0281] (33) ;

[0282] (34) ;

[0283] (35) ;

[0284] wherein, where, w1 is the input layer-hidden layer weight, w2 is the output layer-hidden layer weight, b1 is the hidden layer threshold (bias), b2 is the output layer threshold (bias) for the output of the BP neural network, is the input feature, is the hyperbolic tangent function, is the exponential function with base e, is the linear activation function, is the hidden layer activation function, is the output layer activation function.

[0285] Since the crack propagation rate is not linear with the stress intensity factor, the hyperbolic tangent function is applied to the hidden layer to enable the network to learn complex nonlinear relationships. The linear activation function applied to the output layer can avoid nonlinear distortion and speed up convergence.

[0286] The method for updating the weights and thresholds of the BP neural network by gradient descent is:

[0287] Calculate the mean square error, quantify the neural network prediction error, and then pass it back according to the mean square error. The BP neural network updates the weights and thresholds by gradient descent:

[0288] (36) ;

[0289] (37) ;

[0290] where, and are the updated weights and thresholds, respectively, is the weight before updating, is the threshold before updating, is the weight gradient, is the threshold gradient, is the learning rate.

[0291] The updating feedback is performed in a loop until the training requirements are met, and the fiber metal laminated plate fatigue crack propagation prediction result is output.

[0292] Experimental verification:

[0293] Dataset introduction:

[0294] The verification dataset of the present application is derived from the experimental data obtained from the fatigue crack propagation rate test and the fatigue delamination expansion experiment conducted on the fiber metal laminated plate. The fatigue crack propagation experiment and delamination expansion experiment methods are as follows:

[0295] 1. Experimental material

[0296] The alloy in the fiber metal laminate in the present application is a metal material prepared by taking metal lithium as an alloying element and an aluminum matrix. This alloying effect greatly improves the elastic modulus while significantly reducing the mass density of the material, and the strengthening mechanism can not only reduce the density by 3% for each addition of 1% lithium element, but also maintain the high strength characteristics (comparable to high-strength aluminum alloys such as 2024 and 7075), excellent fatigue resistance and plastic processing capability of the material. Therefore, the test alloy is an 2060-T8E30 aluminum-lithium alloy, with a nominal thickness of 2 mm and an actual thickness of 1.98 mm. According to the “Metal Tensile Test Method at Room Temperature” (HB5143-1996), the tensile properties of the aluminum-lithium alloy plate are tested (rolling direction). The material performance test results are shown in Table 1 below.

[0297] The glass fiber prepreg layer is composed of SY-24 / S4C9-1200 material, which includes high-strength S4 fiber and SY-24 adhesive, and the nominal thickness of the glass fiber prepreg layer is 1.5 mm. For S4 fiber, the average tensile breaking strength of the yarn is ≥780 MPa, and the average strength of the impregnated yarn is ≥3000 MPa; for SY-24 adhesive, the average tensile shear strength is ≥30 MPa, and the average peel strength is ≥6 kN / m. According to the “Tensile Properties Test Method for Directional Fiber Reinforced Plastics” (GB3354-2014), the tensile properties of the unidirectional glass fiber composite material are tested (fiber direction). The material performance test results are shown in Table 1.

[0298] Table 1: Mechanical property data table of materials

[0299]

[0300] The fiber metal laminate used in the test is a 2 / 1 layup structure, which is composed of two layers of aluminum-lithium alloy plates on the outside of the laminate and a fiber prepreg layer in the middle layer.

[0301] 2. Sample form, equipment and scheme

[0302] The constant amplitude fatigue crack propagation test of the fiber metal laminate selects a center crack tension (M(T)) sample. The sample length and width are 270 mm and 75 mm, the center circular hole diameter is 3 mm, and the initial notch is 10 mm and 15 mm. The sample form is as shown in Figure 5 .

[0303] The test strictly follows the standard of "Metallic Materials Fatigue Crack Propagation Rate Test Method" (GB / T 6398-2000), and the research object is 2 / 1 layer plate of fiber reinforced aluminum lithium alloy. The fatigue crack propagation test system is composed of SHIMADZU electro-hydraulic servo control low frequency fatigue testing machine and JDX-B mobile microscope (20 times magnification); the fatigue delamination propagation test system is composed of SHIMADZU electro-hydraulic servo control low frequency fatigue testing machine and digital optical strain measuring instrument. Among them, the fatigue testing machine can realize constant amplitude loading and conventional variable amplitude loading, and the microscope has a measurement accuracy of 0.01 mm.

[0304] The test is carried out in room temperature air environment, using sinusoidal wave loading mode, and the basic frequency is set to 10 Hz. The frequency parameter is adjusted according to the dynamic response characteristics of the test piece to ensure that the measured load peak and valley value is synchronized with the set value. When the system appears vibration instability, the loading frequency is appropriately reduced. In the crack propagation test of constant amplitude load control (stress ratio R=0.06), the crack length is measured by microscope at an interval of 0.5±0.1 mm according to the specification requirements.

[0305] 3. Test data

[0306] The fiber metal laminate fatigue crack propagation test sets two stress ratios, 0.06 and -1 respectively, and carries out fatigue crack propagation test on the initial crack size of 10mm and 15mm of the laminate by applying two constant amplitude loads of 60MPa and 70MPa. According to the test results obtained from the above test, the crack propagation length corresponding to the load cycle number of the fiber metal laminate under the stress ratio of 0.06 and the load of 60Mpa and 70Mpa is shown in Figure 6 , and the crack propagation length corresponding to the load cycle number of the fiber metal laminate under the stress ratio of -1 is shown in Figure 7 .

[0307] The delamination propagation test is carried out on the fiber metal laminate by using stress ratio of 0.06 and constant amplitude load of 60MPa, and the test data obtained when the crack propagation is to 16mm and 20mm is collected by using DIC system, then the delamination size curve of the fiber metal laminate under different crack lengths is drawn according to the test results, as shown in Figure 8 .

[0308] Figure 9 DIC delamination image of fiber metal laminate. In the DIC image, different colors represent different surface strain sizes, and the dark color on the right side of the laminate represents the delamination area, and the light color part is the fatigue crack.

[0309] Comparison and verification of cohesive module:

[0310] High-fidelity modeling of crack propagation of FML under fatigue loading is conducted using ABAQUS. Figure 10 With Figure 11 The system presents the evolution of stress contours and crack propagation under 60MPa loading conditions, Figure 12 With Figure 13 The system presents the evolution of stress contours and crack propagation under 70MPa loading conditions.

[0311] The simulation results show that the pre-crack always expands along the initial crack direction under fatigue loading, and significant stress concentration phenomenon occurs in the crack tip region. Notably, the crack propagation rate is directly related to the load level: as the number of fatigue loading cycles accumulates, the crack length shows a continuous growth trend, and the crack accelerates under high load conditions. This phenomenon is consistent with the Paris fatigue crack propagation theory.

[0312] According to the simulation results, the relationship between the number of load cycles and the corresponding crack length is plotted as follows Figure 14 .

[0313] The data shows that under the same number of cycles, the crack length is significantly positively correlated with the load amplitude. Specifically, the 60MPa load amplitude needs to experience 100000 cycles to reach a crack length of 25.28mm, while the 70MPa load amplitude needs to experience 70000 cycles to reach a crack length of 22.88mm. According to the experimental data of crack propagation under 60MPa and 70MPa loads, the simulation results are compared with the experimental data, and it is found that the data and trends are very close, verifying the accuracy of the fatigue crack propagation simulation model based on the improved cohesive zone model.

[0314] Comparison and verification of fatigue multi-damage coupling module:

[0315] Dynamic coupling simulation of crack propagation and delamination expansion of FML under fatigue loading is conducted using ABAQUS. The initial crack length is considered to be 10mm and 15mm, and the stress ratio is considered to be 0, 0.06, 0.1, 0.3, and 0.5 for each initial crack length. Different stress levels such as 50, 60, 70, 80, and 90 are simulated under different stress ratios. Take the initial crack length of 10mm, stress ratio of 0.06, and stress level of 60MPa as an example. The crack propagation process during simulation is shown in , and the delamination expansion process is shown in Figure 15 . Figure 16 .

[0316] In the simulation process of the fiber metal laminates, delamination occurs at the interface between the metal layer and the fiber layer, and crack propagation also occurs in the metal layer along the initial crack direction. With the increase of the number of cycles, the delamination area of the metal layer and the fiber layer expands continuously, and the shape of the delamination is approximately elliptical. During the crack propagation process of the metal layer, the stress tip moves along the direction of the initial crack to both ends, and the crack also expands to both ends along the direction of the initial crack.

[0317] The initial crack length is 10 mm, and the stress ratio R = 0.06. According to the test data of crack propagation under the loads of 60 MPa and 70 MPa, the comparison results between the simulation results and the test data are shown in Figure 17 . The initial crack length is 15 mm, and the stress ratio R = 0.06. According to the test data of crack propagation under the loads of 60 MPa and 70 MPa, the comparison results between the simulation results and the test data are shown in Figure 18 . According to the comparison between the existing simulation data and the test data, it is found that they are relatively close, which indicates that the established model can simulate the fatigue crack propagation of the fiber metal laminates, and the error is within an acceptable range. The comparison of delamination of fiber metal laminates under different crack lengths is shown in Figure 19 . The simulation results have certain error compared with the corrosion delamination method, but the results are better than the DIC results, which further reflects the advantages of the simulation.

[0318] GA-BP hybrid module comparison and verification:

[0319] (1) Model training and verification analysis

[0320] The training iteration process and training regression curve of the BP neural network and the GA-BP model are shown in Figure 20 and Figure 21 During the neural network training process, the BP algorithm uses the mean square error as the loss function to quantify the prediction deviation. With the increase of the number of iterations, the error curve presents a typical convergence trend until it tends to be stable. The dynamic monitoring of the training state is realized through the gradient descent rate, the model accuracy index and the performance parameters of the verification set. The predicted value and the true value have a high linear correlation.

[0321] The GA-BP model based on genetic algorithm optimization shows excellent prediction ability. The test verification shows that the prediction error curves of the training set and the test set of the model present strong synchronous convergence characteristics, and the generalization performance is significantly better than that of the traditional BP architecture. The root mean square error of the prediction results of the GA-BP model is less than that of the basic BP model. The research results reveal the key role of genetic algorithm in breaking through the local optimal solution limit of BP network and improving the modeling accuracy of nonlinear systems from the algorithm optimization mechanism level, and provide an intelligent solution with more engineering practical value for high-precision fatigue crack propagation prediction.

[0322] The performance difference between the two neural network architectures of BP and GA-BP is accurately characterized in quantitative indicators through error analysis of the prediction model. As shown in Table 2.

[0323] Table 2 Comparison of prediction error of test set;

[0324]

[0325] The data shows that the root mean square error RMSE = 0.62 and the mean absolute error MAE = 0.45 of the BP model test set are significantly higher than the RMSE = 0.58 and the MAE = 0.42 of the GA-BP model, and the average absolute percentage error indicators of the two are of the same order of magnitude. This error distribution characteristic reveals that although the two models perform similarly in the relative error dimension, the GA-BP model realizes systematic improvement in the absolute error indicators through global optimization of the network initial parameters by the genetic algorithm, and the fluctuation range of the prediction value tends to be closer to the test. This comparison result confirms from the algorithm optimization level that the introduction of the genetic algorithm mechanism effectively improves the numerical stability and generalization ability of the BP neural network in the fatigue crack propagation prediction task.

[0326] (2) Model prediction application

[0327] According to the established GA-BP model, the fatigue crack propagation of the fiber metal laminates is predicted. The input is: the crack propagation length corresponding to different cycle times under the conditions of initial crack length of 10 mm, stress ratio of 0.06, load of 60 MPa and 70 MPa, and the crack propagation length corresponding to different cycle times under the conditions of initial crack length of 15 mm, stress ratio of 0.06, load of 60 MPa and 70 MPa. The comparison results of the test results, BP prediction results and GA-BP model prediction results are shown in Figure 22 and Figure 23 .

[0328] From the absolute error indicators corresponding to the new prediction values and actual values of the model, the root mean square error of the GA-BP model under the conditions of 60 MPa and 70 MPa for the initial crack of 10 mm is 0.56 and 1.09, the mean absolute error is 0.43 and 0.92, and the average absolute percentage error is 3.67% and 8.55%. The error comparison of BP prediction results and GA-BP model prediction results is shown in Table 3. In addition, it can be found that the error of BP prediction results and GA-BP model prediction results under 70 MPa is larger, and combined with the test data, it is speculated that it may be caused by the dispersion of the test data.

[0329] Table 3 Comparison of BP prediction error and GA-BP prediction error (initial crack 10 mm);

[0330]

[0331] The root mean square error of the BP neural network model under the condition of 60 MPa and 70 MPa with an initial crack of 15 mm is 0.49 and 0.59, the average absolute error is 0.41 and 0.51, and the average absolute percentage error is 5.29% and 5.54%. The error comparison between the BP prediction result and the GA-BP model prediction result is shown in Table 4.

[0332] Table 4. Error comparison between BP prediction result and GA-BP prediction result (initial crack 15 mm);

[0333]

[0334] Compared with the BP neural network, the GA-BP model reduces the RMSE by 16.15% to 34.4%, the MAE by 11.5% to 35.4%, and the MAPE by 5% to 26.6%. This error distribution characteristic reveals that although the two models perform similarly in the relative error dimension, the GA-BP model achieves systematic improvement in absolute error indicators through global optimization of network initial parameters by genetic algorithm, and the fluctuation range of the prediction value is closer to the test. The comparison result confirms from the algorithm optimization level that the introduction of genetic algorithm mechanism effectively improves the numerical stability and generalization ability of the BP neural network in the fatigue crack propagation prediction task.

[0335] Embodiment 2:

[0336] A hybrid-driven fiber metal laminated fatigue crack propagation prediction system for implementing a hybrid-driven fiber metal laminated fatigue crack propagation prediction method, comprising:

[0337] A finite element model construction module for establishing a finite element model of the fiber metal laminate;

[0338] A cohesive zone model construction module for constructing an improved bilinear cohesive zone model;

[0339] A Paris formula construction and criterion setting module for constructing an improved Paris formula and setting criteria in the fatigue damage evolution simulation of the fiber metal laminate multi-damage behavior;

[0340] A simulation module for performing finite element simulation on the finite element model of the fiber metal laminate based on the criteria in the fatigue damage evolution simulation of the fiber metal laminate multi-damage behavior, using a double-position cohesive zone unit combined with the improved bilinear cohesive zone model and the improved Paris formula. The simulation process simultaneously considers the fatigue crack propagation of the metal layer and the delamination propagation of the interfacial layer to obtain the crack propagation length and the delamination shape.

[0341] The dataset construction module is used to set different simulation conditions and obtain several simulation data samples, and at the same time obtain several real experimental samples to form a sample set. After dividing the sample set into training set, validation set and test set, it is preprocessed to obtain preprocessed training set, validation set and test set.

[0342] The genetic algorithm module is used to improve the BP neural network based on the preprocessed training set using a genetic algorithm, resulting in an improved BP neural network.

[0343] The fine-tuning module trains the improved BP neural network using the training set, adjusts the weights and thresholds, then uses the validation set to adjust the hyperparameters of the trained BP neural network, and finally uses the test set to further evaluate and optimize the computational error of the BP neural network, verify its generalization ability, and obtain the final BP neural network.

[0344] The prediction module uses the final BP neural network to obtain the predicted crack propagation length.

[0345] Example 3:

[0346] This embodiment proposes an electronic device, including: one or more processors, and a memory for storing instructions, which, when executed by the one or more processors, cause the one or more processors to execute the hybrid-driven fiber-reinforced metal laminate fatigue crack propagation prediction method.

[0347] The electronic device may be a mobile phone, computer, or tablet computer, etc., and includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates as described in the embodiments. It is understood that the electronic device may also include an input / output (I / O) interface and communication components.

[0348] The processor is used to execute all or part of the steps in the hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates as described in the above embodiments. The memory is used to store various types of data, which may include, for example, instructions for any application or method in the electronic device, as well as application-related data.

[0349] The processor can be an Application Specific Integrated Cricuit (ASIC), a Digital Signal Processor (DSP), a Programmable Logic Device (PLD), a Field Programmable Gate Array (FPGA), a controller, a microcontroller, a microprocessor, or other electronic elements, which are used to execute the hybrid-driven fiber metal laminates fatigue crack propagation prediction method described in the above embodiments.

[0350] Embodiment 4:

[0351] The embodiment provides a computer readable storage medium storing executable instructions, which, when executed, can be stored in one computer readable storage medium if implemented in the form of a software function unit and sold or used as an independent product.

[0352] The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the hybrid-driven fiber metal laminates fatigue crack propagation prediction method described in various embodiments of the present application.

[0353] The aforementioned storage medium includes a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD (Secure Digital Memory Card) or a DX (an abbreviation of Memory Data Register, MDR) memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, a server, an APP (an abbreviation of Application) application market, etc., which can store a program check code, and stores a computer program thereon, which, when executed by a processor, can implement each step of the aforementioned hybrid-driven fiber metal laminate fatigue crack propagation prediction method.

[0354] Example 5:

[0355] The present embodiment proposes a computer program product including a computer program or instructions, which, when executed by a processor, implements the aforementioned hybrid-driven fiber metal laminate fatigue crack propagation prediction method.

Claims

1. A hybrid-driven method for predicting fatigue crack propagation in fiber-reinforced metal laminates, characterized in that, Includes the following steps: A finite element model of a fiber-metal laminate is established. During the modeling process, a dual-position cohesive layer is set and cohesive elements are divided to obtain dual-position cohesive elements. The dual-position cohesive layer includes a zero-thickness cohesive layer created on each of the two contact surfaces of the metal layer and the fiber layer, and two cohesive layers created on each of the two metal layers, which serve as crack propagation paths. By introducing a cumulative damage formula to improve the traditional bilinear cohesive model, an improved bilinear cohesive model is obtained, which is used to calculate the cumulative damage of each part of the fiber-metal laminate, so as to realize the fatigue damage evolution simulation of the multi-damage behavior of the fiber-metal laminate; the multi-damage behavior includes metal layer crack propagation and interlayer interface delamination. An improved Paris formula is introduced to establish a criterion for characterizing the damage evolution of laminated metals. At the same time, criteria for fatigue damage evolution simulation of multiple damage behaviors of fiber-reinforced metal laminates are set, including crack propagation initiation criteria, interface delamination damage initiation criteria, and interface delamination damage failure criteria. Based on the criteria for fatigue damage evolution simulation of multi-damage behavior of fiber-metal laminates, a finite element model of fiber-metal laminates is simulated by using a dual-position cohesive element combined with an improved bilinear cohesive model and an improved Paris formula. During the simulation, fatigue crack propagation of the metal layer and delamination propagation at the interlayer interface are considered simultaneously to obtain the crack propagation length and delamination shape. For the finite element model of fiber-reinforced metal laminate, several simulation data samples were obtained by setting different simulation conditions. At the same time, several real experimental samples were also obtained to form a sample set. The sample set was divided into training set, validation set and test set and then preprocessed to obtain preprocessed training set, validation set and test set. Each simulation data sample or real experimental sample includes initial crack length, stress ratio, number of cycles, load and crack propagation length. Based on the preprocessed training set, the weights and thresholds of the BP neural network are improved using a genetic algorithm to obtain an improved BP neural network. The improved BP neural network is trained using the training set, and the weights and thresholds are adjusted. Then, the hyperparameters of the trained BP neural network are adjusted using the validation set. Finally, the BP neural network is further evaluated and the computational error is optimized using the test set to verify its generalization ability. The final BP neural network is obtained and used to predict the crack propagation length.

2. The hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates according to claim 1, characterized in that, The establishment of the finite element model of the fiber-reinforced metal laminate specifically includes: A1: Based on the shape and size of the fiber-metal laminate, a geometric model of the fiber-metal laminate is established in the finite element analysis software; the middle layer of the geometric model of the fiber-metal laminate is the fiber layer, and the upper and lower layers are metal layers; A2: Create a cohesive layer on the geometric model of the fiber-reinforced metal laminate; A3: Set the cohesive force parameter and assign the cohesive force parameter to the cohesive layer; for the cohesive layer on the metal layer and the cohesive layer on the two contact surfaces of the metal layer and the fiber layer, the cohesive force parameter is the initial stiffness, including the normal stiffness. Stiffness in the first shear direction Stiffness in the second shear direction ; A4: Set the material parameters for the geometric model of the fiber-reinforced metal laminate; A5: Mesh the metal layer, fiber layer, and cohesive layer in the geometric model of the fiber-metal laminate. The metal layer and fiber layer are divided into several three-dimensional solid elements, and the cohesive layer is divided into several cohesive elements, thus completing the finite element model construction.

3. The hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates according to claim 1, characterized in that, The cumulative damage formula introduced in the improved bilinear cohesive model includes the fatigue damage calculation formula under cyclic loading and the softening damage calculation formula under monotonic loading, specifically: The formula for calculating fatigue damage under cyclic loading is: （1）； In the formula, This represents the increase in fatigue damage. For fatigue damage variables, This is the damage amplification factor. For the separation displacement value, For separation displacement, To separate displacement increments, This is the initial critical displacement. For material constants, For the current equivalent stress, For the Heaviside function, For maximum cohesion, It is a constant characterizing the stress threshold; （2）； （3）； in, The maximum stress in the mixed mode, As a variable representing overall damage, The critical stress at which failure occurs; In formula (1), the Heaviside function That is, when hour ; Equivalent separation displacement and its increment The formula is given by equations (4) and (5) as follows: （4）； （5）； in, For the normal separation displacement components, To separate the tangential displacement components, Let be the cumulative separation displacement vector at time T. For a moment The cumulative separation displacement vector at time t, For time intervals; The formula for calculating softening damage under monotonic load is: （6）； In the formula, This represents the softening damage increment under monotonic loading. and These are the critical displacement and the failure displacement, respectively. Two damage mechanisms and cumulative damage: fatigue damage under cyclic loading and softening damage under monotonic loading. The relationship between them is: （7）； in, For time.

4. The hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates according to claim 1, characterized in that, The improved Paris formula used to establish the criterion for characterizing the damage evolution of laminated metals is as follows: （8）； In the formula, The length of the crack. The number of loops. and For material parameters, It is the amplitude of the energy release rate at the crack tip; （9）； （10）； in, The elastic modulus of the material, The stress ratio of the load; The crack propagation initiation criterion is: （11）； in, As a fatigue failure indicator, , The fatigue crack propagation coefficient; The second nominal stress criterion is used as the initiation criterion for interfacial delamination damage: （12）； In the formula, , and These are the cohesive forces in the normal direction, the cohesive forces in the first shear direction, and the cohesive forces in the second shear direction, respectively. , , The critical stresses for damage initiation in the normal direction, the first shear direction, and the second shear direction are respectively, and their calculation methods are as follows: （13）； （14）； （15）； in, Normal cohesive strength, It is the tangential cohesive strength; The BK criterion is adopted as the failure criterion for interface delamination damage. （16）； In the formula, The critical fracture energy, Normal fracture energy, Shear fracture energy, G is the sum of tangential fracture energies. s G represents the sliding shear energy release rate. t To achieve the shear energy release rate, G T This refers to the total energy release rate; For material-related coefficients; when The BK criterion is triggered.

5. The hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates according to claim 1, characterized in that, The criterion in the fatigue damage evolution simulation based on the multi-damage behavior of fiber-reinforced metal laminates utilizes an improved bilinear cohesive force model and a modified Paris formula to perform finite element simulations on the finite element model of the fiber-reinforced metal laminates. Specifically: B1: Defines the analysis step for finite element simulation of the finite element model of the fiber-reinforced metal laminate; B2: Set the loads for the finite element model of the fiber-reinforced metal laminate; B3: Sets the material state variables, critical stress, fracture energy, BK criterion coefficient, damage amplification factor, and constants characterizing the stress threshold; B4: Set the number of the current cohesive unit to b=0; B5: When the current cohesive unit satisfies the interface delamination damage initiation criterion, damage evolution simulation is performed using the improved bilinear cohesive model. During the process, it is monitored in real time whether the crack propagation initiation criterion or the interface delamination damage failure criterion is reached. If the crack propagation initiation criterion is reached, B6 is executed. If the interface delamination damage failure criterion is reached, the current cohesive unit is deleted and the improved bilinear cohesive model is used to continue the damage evolution simulation. B6: After the crack propagation initiation criterion is met, the improved Paris formula is used to simulate the fatigue crack propagation of the metal layer, and finally the crack propagation length and delamination shape are obtained.

6. The hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates according to claim 5, characterized in that, The damage evolution simulation using the improved bilinear cohesion model specifically includes: C1: For the current cohesive element, calculate the critical displacement and failure displacement based on the set critical stress, initial stiffness and fracture energy; （17）； in, This represents the critical displacement in the normal direction. This represents the critical displacement in the first shear direction. This represents the critical displacement in the second shear direction. （18）； in, The displacement is the failure displacement in the normal direction. The failure displacement is in the first shear direction. The failure displacement is in the second shear direction; C2: Obtain the relative displacement of the cohesive unit at the current moment, and calculate the displacement change in the mixing mode based on the state variables at the current moment. The relative displacement at the current moment includes: relative displacement in the normal direction. Relative displacement in the first shear direction Relative displacement in the second shear direction ; Displacement change in the hybrid mode The calculation method is as follows: （19）； C3: Based on the displacement change in the hybrid mode Calculate the relative displacement at the next moment, and then obtain the displacement change at the next moment in the hybrid mode. ; （20）； in, The relative displacement in the normal direction at the next moment. This represents the relative displacement in the first shear direction at the next moment. This represents the relative displacement in the second shear direction at the next moment; C4: Calculate the initial critical displacement and failure displacement under the hybrid mode based on the relative displacement, critical displacement, initial stiffness and fracture energy at the next moment; （21）； （22）； （23）； in, For mixed loading ratio, This represents the initial critical displacement in the hybrid mode. This refers to the destructive displacement under hybrid mode; C5: Based on the displacement change in the hybrid mode The displacement change at the next moment in the hybrid mode Calculate the initial critical displacement and failure displacement under mixed mode, and calculate the cumulative damage under monotonic loading at the next moment. ; （24）； in, This represents the cumulative damage under monotonic loading at the next time step. C6: Calculate the cumulative damage under cyclic loading at the next time step based on the damage amplification factor and the constant characterizing the stress threshold. ; （25）； in, This represents the cumulative damage amount under the next cyclic loading. For cumulative damage, Stress in the normal direction The stress is in the first shear direction. The stress is in the second shear direction; C7: The cumulative damage under monotonic loading at the next moment. The cumulative damage under the next cyclic loading time step Adding them together yields the cumulative damage of the cohesive unit at the next moment. ; C8: Calculate and update the cumulative damage of the cohesive elements at the next time step. The stress and Jacobian matrix of the cohesive element after cohesion; （26）； （27）； in, For the updated normal stress, The updated first shear direction stress, For the updated second shear direction stress, This represents the relative displacement in the updated normal direction. This represents the relative displacement in the first shear direction after the update. This represents the updated relative displacement in the second shear direction. C9: Determine the cumulative damage of the current cohesive element. If the set threshold is reached, the stiffness of the cohesive element degrades to 0, the cohesive element fails, and C10 is executed to end the fatigue damage evolution simulation process of a cohesive element; otherwise, return to C2. C10: Increment the number of the current cohesive element by 1 and return to C2.

7. A hybrid-driven fatigue crack propagation prediction system for fiber-reinforced metal laminates, used to implement the hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates according to any one of claims 1-6, characterized in that, include: The finite element model building module is used to create finite element models of fiber-reinforced metal laminates. The cohesion model building module is used to build an improved bilinear cohesion model. The Paris formula construction and criterion setting module is used to construct an improved Paris formula and set the criteria in the fatigue damage evolution simulation of multi-damage behavior of fiber-metal laminates. The simulation module is used as a criterion in the fatigue damage evolution simulation based on the multi-damage behavior of fiber-metal laminates. It uses a dual-position cohesive element combined with an improved bilinear cohesive model and an improved Paris formula to perform finite element simulation on the finite element model of the fiber-metal laminate. During the simulation, fatigue crack propagation of the metal layer and delamination propagation at the interlayer interface are considered simultaneously to obtain the crack propagation length and delamination shape. The dataset construction module is used to set different simulation conditions and obtain several simulation data samples, and at the same time obtain several real experimental samples to form a sample set. After dividing the sample set into training set, validation set and test set, it is preprocessed to obtain preprocessed training set, validation set and test set. The genetic algorithm module is used to improve the BP neural network based on the preprocessed training set using a genetic algorithm, resulting in an improved BP neural network. The fine-tuning module trains the improved BP neural network using the training set, adjusts the weights and thresholds, then uses the validation set to adjust the hyperparameters of the trained BP neural network, and finally uses the test set to further evaluate and optimize the computational error of the BP neural network, verify its generalization ability, and obtain the final BP neural network. The prediction module uses the final BP neural network to obtain the predicted crack propagation length.

8. An electronic device, characterized in that, include: One or more processors, and a memory for storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed, cause the processor to perform the hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates according to any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program or instructions that, when executed by a processor, implement the hybrid-driven fatigue crack propagation prediction method for fiber-reinforced metal laminates as described in any one of claims 1-6.

Images