SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method

Abstract

The invention provides a SiC / SiC pyrolytic carbon interface optimization method based on a neural network and a phase field fracture method. The SiC / SiC pyrolytic carbon interface optimization method comprises the following steps: firstly, constructing a representative volume element model of a SiC fiber reinforced composite material; performing transverse tensile simulation on the representative volume element model of the SiC fiber reinforced composite material by adopting a phase field fracture method to obtain transverse tensile strength; and finally, screening the optimal interface thickness by taking the transverse tensile strength as a criterion, and training a reverse neural network to realize prediction and optimization of the optimal interface thickness of the SiC fibers with different material parameters. According to the method, an accurate and efficient representative volume element model is used, a phase field fracture method is combined with a reverse design neural network, and the PyC interface parameters are systematically optimized by simulating crack behaviors and predicting the optimal interface thicknesses of fibers with different radiuses, elastic moduli, critical energy release rates and strengths, so that the performance of the PyC interface is improved. Therefore, the mechanical property and the reliability of the SiC / SiC composite material are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of silicon carbide composite materials, relates to the optimization design of silicon carbide composite materials, and particularly relates to a method for optimizing the SiC / SiC pyrolytic carbon interface based on neural network and phase field fracture method. Background Art

[0002] SiC / SiC composite material is a continuous silicon carbide fiber reinforced silicon carbide ceramic matrix composite material. Due to its low density, high strength, high temperature resistance and excellent oxidation resistance, it shows broad application prospects in the fields of advanced nuclear energy, aerospace and so on. In the SiC / SiC composite material, the pyrolytic carbon (PyC) interface layer is a key structure connecting the fiber and the matrix, and it is mainly deposited on the fiber surface through processes such as chemical vapor infiltration and precursor infiltration and pyrolysis. The PyC interface layer not only protects the fiber from damage during the preparation process, but also significantly affects the mechanical properties and fracture behavior of the composite material by adjusting the bonding strength between the fiber and the matrix. Research shows that an appropriate PyC interface thickness can improve the fracture toughness of the material and prevent brittle failure, and the optimization of interface parameters (such as thickness, microstructure) is an important technical challenge for improving the overall performance of SiC / SiC composite materials. However, traditional optimization methods mostly rely on experimental trial and error, and have the limitations of low efficiency and high cost.

[0003] In recent years, with the development of computational materials science, numerical simulation technology has provided a new path for interface optimization. The phase field fracture method is an advanced simulation method that can effectively describe the processes of crack initiation, propagation and coalescence in materials, and is particularly suitable for studying the complex fracture mechanisms of composite materials. On the other hand, as a powerful machine learning algorithm, neural network can quickly establish the mapping relationship between process parameters and material properties by learning experimental or simulation data, and realize efficient prediction and optimization design.

[0004] In the field of SiC / SiC composites, there is little research on the numerical simulation of cracks in the mesoscopic model of SiC / SiC composites. Currently, the crack simulation methods are mainly divided into two categories: one is the geometric description method, such as the element deletion method, the interface element method, the cohesive force model method, etc.; the other is the non-geometric description method, such as the extended finite element method, in which cracks can expand inside the mesh regardless of the mesh. For example, in the literature "Numerical Simulation of Tensile Mechanical Behavior of Defect-containing C / SiC Plain Weave Composites", the author based on the mesoscopic unit cell model, adopted the Linde failure criterion and Weibull distribution to simulate the random distribution of pore defects, and successfully predicted the tensile stress-strain behavior of plain weave composites, which was in good agreement with the experimental data; however, this study used a progressive damage model and did not consider the dynamic degradation of the interface phase (such as the pyrolytic carbon layer) and its influence on the crack path. Another example is the literature "Molecular Dynamics Study on the Tensile Behavior of SiC Nanofiber / C / SiC Composites", in which the author used the method of molecular dynamics simulation to numerically simulate the significant influence of the amorphous carbon (a-C) coating thickness of SiC nanofibers on the fracture mode and mechanical properties of composites; however, due to the small simulation scale of the molecular dynamics method, it is difficult to guide the optimization of macroscopic preparation process parameters.

[0005] In summary, there is little research on the numerical simulation of mesoscopic cracks in SiC / SiC composites at home and abroad, and there are limitations in aspects such as the characterization of composite material defects and the simulation of interface fracture behavior, resulting in defects in the interface optimization design based on numerical simulation. Summary of the Invention

[0006] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide an optimization method for the SiC / SiC pyrolytic carbon interface based on neural network and phase field fracture method, so as to solve the technical problems that the existing technology has limitations in aspects such as the characterization of composite material defects and the simulation of interface fracture behavior, resulting in defects in the interface optimization design based on numerical simulation.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions to achieve:

[0008] An optimization method for the SiC / SiC pyrolytic carbon interface based on neural network and phase field fracture method, the method includes: first constructing a representative volume unit model of SiC fiber-reinforced composites; then using the phase field fracture method to perform transverse tensile simulation on the representative volume unit model of SiC fiber-reinforced composites to obtain the transverse tensile strength; finally, screening the optimal interface thickness based on the transverse tensile strength and training a reverse neural network to realize the prediction and optimization of the optimal interface thickness of SiC fibers with different material parameters.

[0009] The present invention also has the following technical features:

[0010] The method specifically includes the following steps:

[0011] Step 1, construct a representative volume element model of SiC fiber-reinforced composite material:

[0012] Step 1.1, set the cross-sectional size parameters of the SiC / SiC composite material:

[0013] Set the size of the cross-section of the SiC fiber-reinforced composite material. The cross-sectional size parameters of the SiC fiber-reinforced composite material include: the length of the SiC matrix, the width of the SiC matrix, the radius r of the SiC fiber, and the thickness h of the PyC boundary layer; establish a SiC / SiC composite material model according to the set size parameters. Specifically, in Step 1.1, the SiC / SiC composite material model is a transverse tensile model containing a single fiber; the length of the SiC matrix is 25 μm, the width of the SiC matrix is 25 μm, the radius r of the SiC fiber is 4 - 8 μm, and the thickness h of the PyC boundary layer is 50 - 550 nm.

[0014] Step 1.2, perform mesh division:

[0015] Use quadrilateral elements to perform mesh division on the SiC / SiC composite material model established in Step 1.1, and perform mesh refinement on the regions where cracks may grow to make the calculation results more accurate; thus, the construction of the representative volume element model of the SiC fiber-reinforced composite material is completed. Specifically, in Step 1.2, when performing mesh division, the mesh size is set to 1 / 100 of the width of the SiC matrix in Step 1.1; when performing mesh refinement, the mesh size is set to 1 / 500 - 1 / 400 of the width of the SiC matrix in Step 1.1.

[0016] Step 2, phase-field fracture simulation:

[0017] Step 2.1, set the material property parameters:

[0018] The material property parameters include: the Young's modulus E of the SiC fiber interface 1 , 150 GPa ≤ E 1 ≤ 430 GPa; the Young's modulus E of the SiC matrix interface 2 , 150 GPa ≤ E 2 ≤ 430 GPa; the Young's modulus E of the PyC interface 3 , 10 GPa ≤ E 3 ≤ 30 GPa; the strength Ft of the SiC fiber interface 1 , 300 MPa ≤ Ft 1 ≤ 600 MPa; the strength Ft of the SiC fiber interface 2 , 300 MPa ≤ Ft 2≤600 MPa; The strength Ft of the PyC interface 3 , 20 MPa ≤ Ft 3 ≤60 MPa; The critical energy release rate Gc of the SiC fiber interface 1 , 6 J / m 2 ≤ Gc 1 ≤10 J / m 2 ; The critical energy release rate Gc of the SiC matrix interface 2 , 6 J / m 2 ≤ Gc 2 ≤10 J / m 2 ; The critical energy release rate Gc of the PyC interface 3 , 0.4 J / m 2 ≤ Gc 3 ≤0.6 J / m 2 ; The above parameters are randomly selected within the value range.

[0019] The material property parameters described above also include: The Poisson's ratio v of the SiC fiber interface 1 , v 1 takes the value of 0.185; The Poisson's ratio v of the SiC matrix interface 2 , v 2 takes the value of 0.185; The Poisson's ratio v of the PyC interface 3 , v 3 takes the value of 0.18. After each material property parameter is selected, the maximum-minimum normalization is performed within its respective value range using Equation Ⅰ; The Equation Ⅰ is as follows:

[0020] Equation Ⅰ.

[0021] In the formula:

[0022] X represents the value after normalization.

[0023] A represents the current selected value.

[0024] A max represents the maximum value among all selected values.

[0025] A min represents the minimum value among all selected values.

[0026] Step 2.2, Set boundary conditions:

[0027] Set displacement constraints on the left and right sides of the model established in Step 1.1.

[0028] Step 2.3, Phase field fracture simulation:

[0029] Set the phase field cohesive force control equation and related parameters, set the partial differential equation solver to solve the phase field cohesive force control equation, and set the relevant parameter equations and relevant material parameters; the phase field cohesive force control equation is as shown in Equation II below:

[0030] Equation II.

[0031] Where:

[0032] Represents the phase field cohesive force control equation.

[0033] Represents the phase field value.

[0034] Represents the first derivative of the degradation function.

[0035] Represents the historical maximum value of the strain energy density at a certain point within time period.

[0036] Represents the critical energy release rate.

[0037] Represents the first derivative of the crack geometry function.

[0038] Represents the proportionality constant.

[0039] Represents the characteristic width of the phase field model.

[0040] Step 2.4, Output of the transverse tensile strength:

[0041] After the phase field fracture simulation is completed, extract the model stress-strain curve, and obtain the transverse tensile strength from the stress-strain curve.

[0042] Step Three, Construction and Training of the Neural Network:

[0043] Step 3.1, Data Set Collection:

[0044] Repeat Step Two N times to obtain N transverse tensile strengths, and form the forward neural network training set; N is an integer greater than or equal to 200 and less than or equal to 2000.

[0045] Step 3.2, Forward Neural Network Training:

[0046] For the forward neural network, its inputs are the thickness h of the PyC boundary layer, the radius r of the SiC fiber, and the Young's modulus E of the SiC fiber interface 1 , the Young's modulus E of the SiC matrix interface 2 , the Young's modulus E of the PyC interface3 、The strength Ft of the SiC fiber interface 1 、The strength Ft of the SiC fiber interface 2 、The strength Ft of the PyC interface 3 、The critical energy release rate Gc of the SiC fiber interface 1 、The critical energy release rate Gc of the SiC matrix interface 2 And the critical energy release rate Gc of the PyC interface 3 ; The output is the transverse tensile strength.

[0047] Specifically, in step 3.2, the input layer of the forward neural network has 11 parameters, the hidden layer has 5 layers, and each hidden layer contains 256 neurons; the output layer of the forward neural network has 1 parameter. Among them, the ReLU activation function is used between layers to maintain the nonlinearity of the model, and the Adam algorithm is used for model optimization. The initial learning rate of the Adam optimization algorithm is 1e-3, and the weight decay coefficient is set to 1e-5, and the number of training times is 1000 times.

[0048] Step 3.3, obtain the optimal solution:

[0049] Traverse the thickness h of the PyC boundary layer of the forward neural network for different SiC fiber radii r and different material property parameters, use the transverse tensile strength as the criterion and obtain the optimal PyC thickness, and the results are used for the dataset collection of the inverse neural network.

[0050] Step 3.4, inverse neural network training:

[0051] For the inverse neural network, its input is the SiC fiber radius r, the Young's modulus E of the SiC fiber interface 1 、The Young's modulus E of the SiC matrix interface 2 、The Young's modulus E of the PyC interface 3 、The strength Ft of the SiC fiber interface 1 、The strength Ft of the SiC fiber interface 2 、The strength Ft of the PyC interface 3 、The critical energy release rate Gc of the SiC fiber interface 1 、The critical energy release rate Gc of the SiC matrix interface 2 And the critical energy release rate Gc of the PyC interface 3 ; The output is the optimal PyC thickness; after training, the pyrolytic carbon interface optimization model in the SiC / SiC composite material is obtained.

[0052] Specifically, in step 3.4, the input layer of the reverse neural network has 10 parameters, the hidden layer has 5 layers, each hidden layer contains 256 neurons, and the output layer has 1 parameter. The sigmoid function is used as the activation function between layers to maintain the non-linearity of the model. The Adam algorithm is used to optimize the model, and the initial learning rate is set to 5e-3, and the number of training times is 1000 times.

[0053] The beneficial technical effects of the present invention compared with the prior art:

[0054] (Ⅰ) The present invention uses an accurate and efficient representative volume element model, combines the phase field fracture method with the reverse design neural network, predicts the optimal interface thickness of fibers with different radii, elastic moduli, critical energy release rates, and strengths by simulating crack behavior, and systematically optimizes the PyC interface parameters, thereby improving the mechanical properties and reliability of SiC / SiC composites.

[0055] (Ⅱ) The phase field fracture method used in the present invention is a method that can simulate fracture in a continuous medium, can accurately capture the entire process of material damage including crack nucleation, crack propagation, and final fracture, and can solve problems of complex fracture modes and multi-physical field coupling.

[0056] (Ⅲ) The whole process of the present invention is carried out on a computer, reducing unnecessary experimental trial and error and experimental dependence, and improving the interface optimization design efficiency of SiC / SiC composites.

[0057] (Ⅳ) The reverse optimization design neural network used in the present invention is composed of a forward neural network and a reverse neural network connected in series. It can not only be applied to the interface optimization design of various SiC / SiC composites involving bundled SiC fibers, PyC interfaces, and SiC matrices, but also be extended to other brittle composites with interfaces, that is, it has wide applicability. Description of the Drawings

[0058] Figure 1 It is a flowchart of the SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method.

[0059] Figure 2 It is a schematic diagram of the neural network structure obtained in the embodiment of the present invention.

[0060] Figure 3 It is the prediction result of the interface optimization model obtained in the embodiment of the present invention for the optimal interface thickness of certain fibers.

[0061] The technical solution of the present invention will be further described below in conjunction with embodiments. Specific Embodiments

[0062] In the present invention: SiC refers to silicon carbide, and the two SiCs in "SiC / SiC" respectively refer to SiC fibers and SiC matrix. PyC refers to pyrolytic carbon.

[0063] It should be noted that all the software, algorithms, and functions used in the present invention, without special instructions, are the software, algorithms, and functions known in the art. For example:

[0064] The software for modeling and phase-field fracture simulation uses the commonly used Comsol 6.0 software known in the prior art.

[0065] The Sigmoid function, also known as the S-shaped growth curve, is a commonly used activation function, generally existing between the input layer and the output layer of a neural network, and its role is to add some non-linear factors to the neural network.

[0066] The Adam (Adaptive Moment Estimation) algorithm is an optimization algorithm that combines the momentum method and the adaptive learning rate, proposed by D.P. Kingma and J. Ba in 2014.

[0067] represents the degradation function, and the degradation function is generally selected as: ; where is the phase-field value to be solved.

[0068] Following the above technical solutions, the specific embodiments of the present invention are given below. It should be noted that the present invention is not limited to the following specific embodiments, and all equivalent transformations made on the basis of the technical solutions of this application fall within the protection scope of the present invention.

[0069] Embodiment:

[0070] This embodiment provides an optimization method for the SiC / SiC pyrolytic carbon interface based on a neural network and the phase-field fracture method; as Figure 1 shown, the method specifically includes the following steps:

[0071] Step 1: Construct a representative volume element model of the SiC fiber-reinforced composite material:

[0072] Step 1.1: Set the cross-sectional dimension parameters of the SiC / SiC composite material:

[0073] To simulate the transverse tensile strength of SiC fiber-reinforced composites, it is necessary to set the dimensions of the cross-section of the SiC fiber-reinforced composites. The cross-sectional dimension parameters of the SiC fiber-reinforced composites include: the length of the SiC matrix is 25 μm, the width of the SiC matrix is 25 μm, the radius r of the SiC fiber is 4 - 8 μm, and the thickness h of the PyC boundary layer is 50 - 550 nm. A SiC / SiC composite model (this SiC / SiC composite model is a transverse tensile model containing a single SiC fiber) is established according to the set dimension parameters.

[0074] Step 1.2, perform mesh generation:

[0075] The SiC / SiC composite model established in Step 1.1 is meshed using quadrilateral elements. The mesh size is set to 1 / 100 of the SiC matrix width in Step 1.1, and the area where cracks may grow is meshed with a finer mesh size, which is set to 1 / 500 of the SiC matrix width in Step 1.1 to make the calculation results more accurate. Thus, the construction of the representative volume element model of the SiC fiber-reinforced composites is completed.

[0076] Step Two, phase-field fracture simulation:

[0077] Step 2.1, set the material property parameters:

[0078] The specific values of the material property parameters are as follows: The Young's modulus E of the SiC fiber interface 1 , 150 GPa ≤ E 1 ≤ 430 GPa; The Young's modulus E of the SiC matrix interface 2 , 150 GPa ≤ E 2 ≤ 430 GPa; The Young's modulus E of the PyC interface 3 , 10 GPa ≤ E 3 ≤ 30 GPa; The strength Ft of the SiC fiber interface 1 , 300 MPa ≤ Ft 1 ≤ 600 MPa; The strength Ft of the SiC fiber interface 2 , 300 MPa ≤ Ft 2 ≤ 600 MPa; The strength Ft of the PyC interface 3 , 20 MPa ≤ Ft 3 ≤ 60 MPa; The critical energy release rate Gc of the SiC fiber interface 1 , 6 J / m 2 ≤ Gc 1 ≤ 10 J / m 2 ; The critical energy release rate Gc of the SiC matrix interface 2 , 6 J / m 2 ≤ Gc2 ≤10 J / m 2 ; The critical energy release rate Gc of the PyC interface 3 , 0.4 J / m 2 ≤Gc 3 ≤0.6 J / m 2 ; Randomly assign values to each material property parameter within the above value range. In addition, the Poisson's ratio v of the SiC fiber interface 1 , v 1 is assigned a value of 0.185; the Poisson's ratio v of the SiC matrix interface 2 , v 2 is assigned a value of 0.185; the Poisson's ratio v of the PyC interface 3 , v 3 is assigned a value of 0.18.

[0079] The material property parameters need to be normalized separately using Equation Ⅰ, and Equation Ⅰ is as follows:

[0080] Equation Ⅰ.

[0081] In the formula:

[0082] X represents the value after normalization.

[0083] A represents the current value.

[0084] A max represents the maximum value among all values.

[0085] A min represents the minimum value among all values.

[0086] Step 2.2, set boundary conditions: To simulate the transverse tensile strength of the SiC / SiC composite material, set displacement constraints on the left and right sides of the model established in Step 1.1.

[0087] Step 2.3, phase field fracture simulation: Set the phase field cohesive force control equation and related parameters, and set the partial differential equation solver to solve the phase field cohesive force control equation, and set the relevant parameter equations and related material parameters; the phase field cohesive force control equation is as follows in Equation Ⅱ:

[0088] Equation Ⅱ.

[0089] In the formula:

[0090] represents the phase field cohesive force control equation.

[0091] represents the phase field value, to be solved, used to describe the fracture situation; 0 represents completely intact, and 1 represents completely fractured.

[0092] Represents the first derivative of the degradation function, which is used to reflect the reduction of strain energy caused by fracture damage.

[0093] Represents the historical maximum value of the strain energy density at a certain point within the time period.

[0094] Represents the critical energy release rate, which is related to the material properties and is set in Step 2.

[0095] Represents the first derivative of the crack geometry function.

[0096] Represents the proportionality constant; generally takes a value of .

[0097] Represents the characteristic width of the phase field model, which is used to control the degree of crack dispersion; generally takes a value of 1 / 4 of the grid size.

[0098] Step 2.4, Output of the transverse tensile strength: After the phase field fracture simulation is completed, extract the stress-strain curve of the model. The horizontal axis is the strain and the vertical axis is the stress. Obtain the value (maximum value) at the highest point from the stress-strain curve, which is the transverse tensile strength.

[0099] Step Three, Construction and Training of the Neural Network:

[0100] Step 3.1, Dataset Collection: Repeat Step Two 1000 times to obtain 1000 transverse tensile strengths, which form the forward neural network training set.

[0101] Step 3.2, Forward Neural Network Training: For the forward neural network, its inputs are the thickness h of the PyC boundary layer, the radius r of the SiC fiber, the Young's modulus E 1 of the SiC fiber interface, the Young's modulus E 2 of the SiC matrix interface, the Young's modulus E 3 of the PyC interface, the strength Ft 1 of the SiC fiber interface, the strength Ft 2 of the PyC interface, the strength Ft 3 of the SiC fiber interface, the critical energy release rate Gc 1 of the SiC matrix interface, the critical energy release rate Gc 2 of the SiC matrix interface, and the critical energy release rate Gc 3 of the PyC interface; the output is the transverse tensile strength.

[0102] As a specific solution of this embodiment, the input layer of the forward neural network has 11 parameters, the hidden layer has 5 layers, and each hidden layer contains 256 neurons; the output layer of the forward neural network has 1 parameter. Among them, the ReLU activation function is used between layers to maintain the nonlinearity of the model, and the Adam algorithm is used for model optimization. The initial learning rate of the Adam optimization algorithm is 1e-3, the weight decay coefficient is set to 1e-5, and the number of training times is 1000 times.

[0103] Step 3.3, obtaining the optimal solution: For the forward neural network, traverse the thickness h of the PyC boundary layer at different SiC fiber radii r and different material property parameters (including the Young's modulus E of the SiC fiber interface 1 , the Young's modulus E of the SiC matrix interface 2 , the Young's modulus E of the PyC interface 3 , the strength Ft of the SiC fiber interface 1 , the strength Ft of the SiC fiber interface 2 , the strength Ft of the PyC interface 3 , the critical energy release rate Gc of the SiC fiber interface 1 , the critical energy release rate Gc of the SiC matrix interface 2 and the critical energy release rate Gc of the PyC interface 3 ), and use the transverse tensile strength as the criterion to obtain the optimal PyC thickness. The results are used for dataset collection of the inverse neural network.

[0104] Step 3.4, inverse neural network training: For the inverse neural network, its inputs are the SiC fiber radius r, the elastic modulus E of the SiC fiber interface 1 , the Young's modulus E of the SiC matrix interface 2 , the Young's modulus E of the PyC interface 3 , the strength Ft of the SiC fiber interface 1 , the strength Ft of the SiC fiber interface 2 , the strength Ft of the PyC interface 3 , the critical energy release rate Gc of the SiC fiber interface 1 , the critical energy release rate Gc of the SiC matrix interface 2 and the critical energy release rate Gc of the PyC interface 3 ; the output is the optimal PyC thickness. After training, a pyrolytic carbon interface optimization model in the SiC / SiC composite material is obtained (as shown in Figure 3 ). The obtained neural network structure is as shown in Figure 2 .

[0105] As a specific solution of this embodiment, the input layer of the reverse neural network has 10 parameters, the hidden layer has 5 layers, each hidden layer contains 256 neurons, and the output layer has 1 parameter. Among them, the sigmoid function is used as the activation function between layers to maintain the non-linearity of the model. The Adam algorithm is used to optimize the model, the initial learning rate is set to 5e-3, and the number of training times is 1000 times.

[0106] The above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method, characterized in that: The method comprises: firstly constructing a representative volume unit model of SiC fiber reinforced composite materials; then adopting the phase field fracture method to simulate the transverse tensile strength of the representative volume unit model of SiC fiber reinforced composite materials to obtain the transverse tensile strength; finally, the optimal interface thickness is screened and the inverse neural network is trained based on the transverse tensile strength to realize the prediction and optimization of the optimal interface thickness of SiC fibers with different material parameters.

2. The SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method according to claim 1, characterized in that: The method specifically comprises the following steps: Step 1: Construct a representative volume unit model of SiC fiber reinforced composite materials: Step 1.1, set the cross-sectional dimension parameters of SiC / SiC composite materials: The size of the cross section of the SiC fiber reinforced composite material is set. The size parameters of the cross section of the SiC fiber reinforced composite material include: SiC matrix length, SiC matrix width, SiC fiber radius r and PyC boundary layer thickness h; a SiC / SiC composite material model is established according to the set size parameters; Step 1.2, mesh division: Use quadrilateral elements to mesh the SiC / SiC composite material model built in step 1.1, and encrypt the mesh in the area where cracks may grow to make the calculation results more accurate; thus, the construction of the representative volume unit model of SiC fiber reinforced composite materials is completed; Step 2: Phase field fracture simulation: Step 2.1, set material property parameters: The material property parameters include: Young's modulus E1 of SiC fiber interface, Young's modulus E2 of SiC matrix interface, Young's modulus E3 of PyC interface, strength Ft1 of SiC fiber interface, strength Ft2 of SiC fiber interface, strength Ft3 of PyC interface, critical energy release rate Gc1 of SiC fiber interface, critical energy release rate Gc2 of SiC matrix interface and critical energy release rate Gc3 of PyC interface, and the above parameters are randomly selected within the value range; The material property parameters also include: Poisson's ratio v1 of the SiC fiber interface, Poisson's ratio v2 of the SiC matrix interface and Poisson's ratio v3 of the PyC interface, and the Poisson's ratio is directly taken; After taking the values ​​of each material property parameter, the maximum and minimum normalization processing is performed within the respective value range; Step 2.2, set boundary conditions: Set displacement constraints on the left and right sides of the model established in step 1.1; Step 2.3, phase field fracture simulation: Set the phase field cohesion control equation and related parameters, set the partial differential equation solver to solve the phase field cohesion control equation, and set related parametric equations and related material parameters; Step 2.4, output of transverse tensile strength: After the phase field fracture simulation is completed, the model stress-strain curve is extracted, and the transverse tensile strength is obtained from the stress-strain curve; Step 3: Construction and training of neural network: Step 3.1, dataset collection: Repeat step 2 N times to obtain N transverse tensile strengths to form a forward neural network training set; N is an integer greater than or equal to 200 and less than or equal to 2000; Step 3.2, forward neural network training: For the forward neural network, its input is the thickness h of the PyC boundary layer, the radius r of the SiC fiber, the Young's modulus E1 of the SiC fiber interface, the Young's modulus E2 of the SiC matrix interface, the Young's modulus E3 of the PyC interface, the strength Ft1 of the SiC fiber interface, the strength Ft2 of the SiC fiber interface, the strength Ft3 of the PyC interface, the critical energy release rate Gc1 of the SiC fiber interface, the critical energy release rate Gc2 of the SiC matrix interface, and the critical energy release rate Gc3 of the PyC interface; the output is the transverse tensile strength; Step 3.3, get the optimal solution: The thickness h of the PyC boundary layer is traversed by the forward neural network at different SiC fiber radii r and different material property parameters, and the optimal PyC thickness is obtained based on the transverse tensile strength. The results are used to collect data sets for the reverse neural network. Step 3.4, reverse neural network training: For the inverse neural network, its input is the SiC fiber radius r, Young's modulus E1 of the SiC fiber interface, Young's modulus E2 of the SiC matrix interface, Young's modulus E3 of the PyC interface, strength Ft1 of the SiC fiber interface, strength Ft2 of the SiC fiber interface, strength Ft3 of the PyC interface, critical energy release rate Gc1 of the SiC fiber interface, critical energy release rate Gc2 of the SiC matrix interface and critical energy release rate Gc3 of the PyC interface; the output is the optimal PyC thickness; after the training is completed, the optimization model of the pyrolytic carbon interface in the SiC / SiC composite material is obtained.

3. The SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method according to claim 2, characterized in that: In step 1.1, the SiC / SiC composite material model is a transverse tensile model containing a single fiber.

4. The SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method according to claim 2, characterized in that: In step 1.1, the length of the SiC substrate is 25 μm, the width of the SiC substrate is 25 μm, the radius r of the SiC fiber is 4 to 8 μm, and the thickness h of the PyC boundary layer is 50 to 550 nm.

5. The SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method according to claim 2, characterized in that: In step 1.2, when meshing, the mesh size is set to 1 / 100 of the width of the SiC substrate in step 1.1; when meshing, the mesh size is set to 1 / 500 to 1 / 400 of the width of the SiC substrate in step 1.

1.

6. The SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method according to claim 2, characterized in that: In step 2.1, the range and value of the material property parameters are: Young's modulus E1 of SiC fiber interface, 150 GPa≤E1≤430 GPa; Young's modulus E2 of SiC matrix interface, 150 GPa≤E2≤430 GPa; Young's modulus E3 of the PyC interface, 10 GPa≤E3≤30 GPa; The strength of SiC fiber interface Ft1, 300 MPa≤Ft1≤600 MPa; The strength of SiC fiber interface Ft2, 300 MPa≤Ft2≤600 MPa; The strength of the PyC interface is Ft3, 20 MPa≤Ft3≤60 MPa; Critical energy release rate of SiC fiber interface Gc1,6 J / m 2 ≤Gc1≤10 J / m 2 ; Critical energy release rate of SiC matrix interface Gc2,6 J / m 2 ≤Gc2≤10 J / m 2 ; Critical energy release rate of PyC interface Gc3, 0.4 J / m 2 ≤Gc3≤0.6 J / m 2 ; The Poisson’s ratio v1 of the SiC fiber interface is 0.185; The Poisson’s ratio v2 of the SiC matrix interface is 0.185; The Poisson's ratio v3 of the PyC interface is 0.

18.

7. The SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method according to claim 2, characterized in that: In step 2.1, the material property parameters are normalized using formula I, which is as follows: Formula I; Where: X represents the normalized value; A represents the current value; A max Indicates the maximum value among all values; A min Indicates the minimum value among all values.

8. The SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method according to claim 2, characterized in that: In step 2.3, the phase field cohesion control equation is as follows: Formula II; Where: represents the governing equation of phase field cohesion; represents the phase field value; represents the first derivative of the degradation function; The strain energy density at a point is The historical maximum value in the time period; represents the critical energy release rate; represents the first derivative of the crack geometry function; represents the proportionality constant; Represents the characteristic width of the phase field model.

9. The SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method according to claim 2, characterized in that: In step 3.2, the input layer of the forward neural network has 11 parameters, the hidden layer has 5 layers, and each hidden layer contains 256 neurons; the output layer of the forward neural network has 1 parameter, and the ReLU activation function is used between layers to maintain the nonlinearity of the model. The Adam algorithm is used for model optimization. The initial learning rate of the Adam optimization algorithm is 1e-3, and the weight decay coefficient is set to 1e-5. The number of training times is 1000.

10. The SiC / SiC pyrolytic carbon interface optimization method based on neural network and phase field fracture method according to claim 2, characterized in that: In step 3.4, the input layer of the inverse neural network has 10 parameters, the hidden layer has 5 layers, each hidden layer contains 256 neurons, and the output layer has 1 parameter. The Sigmoid function is used as the activation function between layers and the nonlinearity of the model is maintained. The Adam algorithm is used for model optimization, and the initial learning rate is set to 5e-3, and the number of training times is 1000.

Images