A method for evaluating the reliability of a turbine blade thermal barrier coating force-thermal coupling

Abstract

The application provides a turbine blade thermal barrier coating force-thermal coupling reliability evaluation method, which comprises the following steps: measuring and constructing a density distribution function of a thermal barrier coating microstructure and thermodynamic parameters, completing macroscopic complex turbine blade thermal barrier coating temperature field and strain field simulation, constructing a Monte Carlo sample space of random parameters at different positions of the blade, establishing a micro-scale force-thermal coupling finite element model and obtaining a life data set, building and training a deep learning agent model, and completing turbine blade thermal barrier coating reliability evaluation and sensitivity analysis. The reliability evaluation method provided by the application comprehensively considers the influence of the thermal barrier coating microstructure factor and the force-thermal coupling failure mechanism, solves the problems of high dimension, high nonlinear calculation cost and the like in the reliability evaluation of the complex blade thermal barrier coating, improves the accuracy and efficiency of the reliability evaluation, and provides an evaluation means for the safe application and optimized design of the thermal barrier coating.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of thermal barrier coating reliability evaluation system in high-performance aero-engine, and particularly relates to a turbine blade thermal barrier coating force-thermal coupling reliability evaluation method. BACKGROUND

[0002] Thermal barrier coating (TBCs for short) is a ceramic coating deposited on the surface of high-temperature-resistant metal or superalloy, which is used to reduce the substrate temperature so that the engine turbine blade can safely operate at high temperature. Thermal barrier coating has the characteristics of high melting point, low thermal conductivity, corrosion resistance and thermal shock resistance, and is considered as the most practical method to significantly improve the working temperature of aero-engine, and is widely used in the fields of aviation, chemical industry, metallurgy and energy.

[0003] However, due to the mismatch of the performance parameters of each layer of material constituting the thermal barrier coating, the complex interface structure, and the extremely complex geometry, microstructure and service environment of the thermal barrier coating, the coating may crack and peel off under unpredictable conditions. In engineering practice, the thermal barrier coating material parameters have dispersion, such as microstructure parameters such as porosity, columnar structure, elastic modulus, interface bonding strength, coating fracture strength, which are difficult to accurately control in the preparation process and have dispersion, ultimately leading to the dispersion of the service life of the thermal barrier coating which is difficult to predict. The use of reliability analysis can quantify the influence of the uncertainty of random parameters on the life. Therefore, the structural reliability evaluation method is a feasible solution to solve the dispersion failure of the thermal barrier coating and the difficulty in predicting its safety in the field of aero-engine.

[0004] The failure mechanism of thermal barrier coating is complex and diverse, mainly including high-temperature oxidation, CMAS (calcium magnesium aluminum silicon metal oxide) corrosion and other force-thermal coupling failure mechanisms, and the thermal barrier coating is a multi-layer and porous structure, which also affects the dispersion failure life of the thermal barrier coating. Therefore, how to consider the microstructure of the thermal barrier coating and the failure mechanism of force-thermal coupling, and apply it to the reliability evaluation of complex turbine blade thermal barrier coating is an urgent demand in practical engineering application. However, due to the difficulties of high nonlinearity, multi-scale, multi-parameter and force-thermal coupling, the traditional structural reliability evaluation method is difficult to apply to the turbine blade thermal barrier coating. Therefore, it is of great engineering significance to develop the corresponding reliability method. SUMMARY

[0005] (I) Invention purpose

[0006] The purpose of the present application is to provide a turbine blade thermal barrier coating force-thermal coupling reliability evaluation method to improve the accuracy and efficiency of reliability evaluation, and to provide an evaluation means for the safe application and optimization design of turbine blade thermal barrier coating.

[0007] (II) Technical Solution

[0008] To solve the above problems, the application provides a turbine blade thermal barrier coating force-thermal coupling reliability evaluation method, comprising the following steps:

[0009] Step one, measure the thermal, mechanical and microstructure parameters of the turbine blade thermal barrier coating, and calculate the mean and standard deviation of each parameter; use the random parameter statistical distribution function to calculate the mean and standard deviation of each parameter, and obtain M density distribution functions describing the random characteristics of the thermal, mechanical and microstructure of the turbine blade thermal barrier coating;

[0010] Step two, establish a macroscopic complex turbine blade thermal barrier coating geometric model and its calculation grid, including: using numerical simulation to obtain the temperature field and strain field of the turbine blade thermal barrier coating, and exporting the coordinates, temperature and 3-direction strain of the data points of the calculation grid to an excel file;

[0011] Step three, use the loop statement in Python to read the temperature and 3-direction strain of each coordinate point in the excel file to form a 4×N array; then, based on the M parameter density distribution functions measured in step one, use the Monte Carlo method to take out N values respectively, thereby forming a (M+4)×N-dimensional array S i , i represents the i-th coordinate point;

[0012] Step four, select a row in the array S i obtained in step three, and use COMSOL to construct a thermal barrier coating microstructure geometric model and divide the thermal barrier coating microstructure calculation grid according to the microstructure parameters of the selected row; construct the force-thermal coupling constitutive equation of the thermal barrier coating and input it into COMSOL as the solution control equation; input the temperature and material parameters of the selected row into COMSOL as the solution material parameters; and input the strain of the selected row into COMSOL as the boundary condition of the solution, complete the microstructure finite element calculation to obtain the Mises stress; finally, solve the failure life of the thermal barrier coating based on the Paris life model;

[0013] Step five, based on the failure life of the thermal barrier coating, solve the life of the thermal barrier coating under different temperature, strain, thermal, mechanical and microstructure parameters to form a life data set; further construct a deep neural network, train the Paris life model based on the life data set, and obtain a deep learning proxy model;

[0014] Step six, based on the deep learning proxy model, predict the array S iThe cumulative distribution function of the lifetime of the thermal barrier coating corresponding to all rows of data is obtained by fitting a normal distribution or a Weibull distribution to the lifetime of the i-th coordinate point. Then, the cumulative distribution function of the lifetime of all coordinate points is obtained by using a loop statement in Python. The cumulative distribution function of the lifetime of the coordinate points is visualized to obtain the failure probability cloud map of the thermal barrier coating of the turbine blade at different service times, thus realizing the reliability evaluation of the thermal barrier coating of the turbine blade.

[0015] Step 7: Use a deep learning surrogate model and variance method to perform sensitivity analysis on M random parameters to obtain the key influencing factors affecting the reliability of the thermal barrier coating on turbine blades.

[0016] Furthermore, the thermal parameters include: the thermal conductivity and coefficient of thermal expansion of each layer of the thermal barrier coating;

[0017] Mechanical parameters include: elastic modulus, Poisson's ratio, and fracture toughness;

[0018] Microstructure parameters include: the thickness, porosity, and interface roughness of each layer of the thermal barrier coating;

[0019] Random parametric statistical distribution functions include the normal distribution, Weibull distribution, log-normal distribution, exponential distribution, gamma distribution, and beta distribution.

[0020] Furthermore, the numerical simulation in step two includes the following steps: establishing a geometric model of the flow field and the macroscopically complex turbine blade thermal barrier coating, dividing the computational mesh, setting boundary conditions and material parameters, and solving for the temperature field and strain field contour maps;

[0021] The geometric model of the complex turbine blade thermal barrier coating includes internal cooling channels and film cooling structure, and the computational mesh includes mesh refinement and boundary layer mesh.

[0022] Boundary conditions include the temperature, pressure, and turbulence of the combustion gases and the temperature, pressure, and turbulence of the cold air, all derived from the actual engine service environment.

[0023] Furthermore, in step three, the (M+4)×N dimensional array S i Let represent the Monte Carlo reliability space of the i-th coordinate point. The reliability space has N points, each containing thermal parameters, mechanical parameters, microstructural parameters, xyz triaxial strain, and the thickness of each of the ceramic layer, oxide layer, and adhesive layer.

[0024] Furthermore, in step four, the geometric model of the thermal barrier coating microstructure is constructed during the modeling process. The microstructure includes the multi-layer structure of the ceramic layer, oxide layer, and adhesive layer of the thermal barrier coating, as well as the microstructure of interface roughness and pores.

[0025] Furthermore, in step four, the parameters of the mechanothermal coupling constitutive equation of the mechanothermal barrier coating include one or more of the following: high-temperature interface oxidation, creep, sintering, and high-temperature corrosion.

[0026] Furthermore, the lifetime dataset of the thermal barrier coating under different temperatures, strains, thermal parameters, mechanical parameters and microstructure parameters obtained in step five; the strain includes the strain in the x and y directions parallel to the thermal barrier coating / substrate interface and the strain in the z direction perpendicular to the interface, and the local strain at the corresponding position of the macroscopic turbine blade is applied to the microscopic finite element boundary conditions in step four to ensure the consistency between microscopic and macroscopic deformation.

[0027]

[0028] in For microscopic strain, For macroscopic response.

[0029] Furthermore, the lifetime dataset in step five also includes experimental spalling failure lifetime. The turbine blade sample is placed in a constant temperature furnace at a preset temperature for high-temperature thermal cycling and high-temperature corrosion tests to obtain the experimental spalling failure lifetime of the thermal barrier coating.

[0030] Furthermore, in step five, the deep learning network is constructed using TensorFlow, which includes an input layer, an output layer, and multiple hidden layers. The number of hidden layers is adjusted according to the degree of non-linearity. Adjacent layers in the deep neural network are connected by weights and the ReLU activation function is used. During training, the deep learning network is updated using mini-batch gradient descent, and the following selection function is used to estimate the importance samples S2 = {x(i) | i = 1, 2, ..., N}:

[0031]

[0032] in Let be the lower bound of the space of the τth most important samples. Let τ be the upper bound of the space of the most important samples, and let DNN be the upper bound of the space of the most important samples. (τ-1) The deep neural network is trained for the (τ-1)th time. By performing mechanical-thermal coupling finite element lifetime prediction on the updated samples, the updated labeled dataset is obtained. The deep neural network is then trained again on the updated labeled dataset until the deep neural network reaches the target accuracy requirement.

[0033] Furthermore, in step five, the deep learning network training process includes numerical simulation lifetime data and experimental peeling failure lifetime. The loss function L(p) for the experimental peeling failure lifetime uses N times (N>1) weights to train the deep learning network, as shown in the following equation:

[0034]

[0035] Where y d,j y is the deep learning prediction value, and yj is the microstructure finite element simulation prediction value.

[0036] Furthermore, the cumulative distribution function of lifetime in step six is ​​described by the following Weibull distribution:

[0037]

[0038] In the formula, t is the service time, F(t) is the probability of failure at time t, and α and β are obtained by fitting using the least squares method. Finally, the coordinates and failure probabilities of all the Monte Carlo data points are mapped onto the computational grid of the turbine blade thermal barrier coating, and a reliability cloud map is obtained through visualization software.

[0039] Furthermore, the sensitivity analysis in step seven involves using a deep learning surrogate model to obtain the lifetime value of each parameter with different values, and then using the variance method to obtain the sensitivity of each parameter, thus achieving sensitivity analysis.

[0040] (III) Beneficial Effects

[0041] The reliability evaluation method proposed in this invention considers the influence of the microstructure of thermal barrier coatings and the essential mechanism of internal force, heat, and chemical coupling failure. It can be applied to the reliability evaluation of thermal barrier coatings for complex turbine blades with three-dimensional film pores under real working conditions. While improving the accuracy of reliability evaluation, it can also ensure the efficiency of reliability evaluation, thus providing a guarantee for the design and safe application of thermal barrier coatings.

[0042] In summary, this invention provides a method for evaluating the reliability of turbine blade thermal barrier coatings through a combination of mechanical, thermal, and chemical forces. This method addresses the current limitations of computationally expensive and impractical methods for evaluating the reliability of complex turbine blade thermal barrier coatings that consider intricate failure mechanisms. It significantly improves the accuracy and efficiency of reliability evaluation for thermal barrier coatings, resulting in substantial economic benefits. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating the evaluation method of the present invention;

[0044] Figure 2 This is a local microscale finite element geometric model in one embodiment of the present invention;

[0045] Figure 3 This is the loss function for the deep learning network training process in one embodiment of the present invention;

[0046] Figure 4 This represents the failure probability of a localized area of ​​the thermal barrier coating in one embodiment of the present invention.

[0047] Figure 5This is a reliability failure probability cloud map of the thermal barrier coating on a turbine blade after 1000 hours, according to one embodiment of the present invention.

[0048] Figure 6 Sensitivity analysis of thermal barrier coating in one embodiment of the present invention; Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0050] like Figure 1 As shown, the turbine blade thermal barrier coating mechanical-thermal-chemical coupling reliability evaluation method of the present invention includes the following steps:

[0051] (1) Measure the thermal, mechanical, and microstructural parameters of the thermal barrier coating on the turbine blades, and calculate the mean and standard deviation of each parameter. In this embodiment, the ceramic layer thickness is 200 μm, the elastic modulus is 40 GPa, the Poisson's ratio is 0.12, the coefficient of thermal expansion is 0.0000012, and the creep parameter is selected as 3.5 × 10⁻⁶. -12 The adhesive layer has a thickness of 200 μm, an elastic modulus of 110 GPa, a Poisson's ratio of 0.3, and a coefficient of thermal expansion of 0.0000016. To simplify this example, only the random characteristics of the mechanical parameters of the oxide layer and the random characteristics of the microstructure of the interface roughness are considered. Based on the characteristics of the data, the random characteristics of these parameters are described by statistical distribution functions such as normal distribution and Weibull distribution, as shown in Table 1.

[0052] Table 1

[0053]

[0054] (2) A geometric model file of the complex thermal barrier coating of a turbine blade was created using SolidWorks and saved as a .x_t file. The model was then imported into ICEM to generate the computational mesh, and ANSYS CFX was imported to input the boundary conditions and material parameters. The inlet pressure was set to 0.34 MPa, the inlet temperature to 2000 K, and the outlet pressure to 0.2 MPa. After obtaining the temperature field, the temperature field was imported into Workbench to solve for the stress and strain fields. The coordinates of the points on the interface of the thermal barrier coating blade, as well as the corresponding temperature and strain fields, were exported as a yep.csv file.

[0055] (3) Write code in Python to sequentially input the temperature and strain at different positions of the blade from yep.csv using a loop, and construct a density distribution function. Input the mean and standard deviation of each parameter measured in step one, and use the Monte Carlo method to obtain 1000 samples S={x (i) |i=1,2,…,N}, and define it as an array S;

[0056] (4) Construct a geometric model of the thermal barrier coating microstructure considering local microscale structures in COMSOL software. In this example, the average interface roughness R = 12 μm is taken. For a more comprehensive model, non-periodic interface roughness and microstructures such as internal pores can be considered during modeling. Figure 2 When setting boundary conditions, Figure 2 The bottom of the model is fixed, the displacement in the x-direction of the left boundary is zero, and the boundary displacement is set as u = l0·ε, where l0 is the width of the microscale model, and ε is the local strain extracted from the corresponding position of the macroscopic turbine blade. Its function is to take into account the influence of the complex three-dimensional geometry of the turbine blade on the internal stress and life of the local thermal barrier coating. In this example, only three failure mechanisms are considered: high-temperature oxidation, thermal mismatch, and creep. The mechanical-thermal-chemical coupling constitutive equation is combined with Paris's law to obtain the failure life. The specific equation is as follows:

[0057]

[0058]

[0059] ε=ε e +ε ox +ε th +ε cr (3)

[0060]

[0061]

[0062]

[0063]

[0064]

[0065]

[0066] In the formula, σ represents the symmetric Cauchy stress tensor, ε is the total strain, u is the displacement, and ε e It is elastic strain, ε ox It is growth strain, ε th It is thermal strain, ε crHere, f is the thermal strain, c represents the oxygen concentration, n represents the volume fraction of oxides, G is the energy release rate, D is the thermal barrier coating damage (when D = 1, the thermal barrier coating fails), and the others are material parameters. a and b are fitting parameters; in this example, a = 10. -7 b = 1.345. In the above formula, equation (5) considers creep, equations (6) and (7) consider interfacial oxidation, and equation (4) couples thermal mismatch, high-temperature oxidation, and creep. Optionally, failure mechanisms such as corrosion and sintering can also be considered.

[0067] (5) Based on the above localized thermo-mechanical coupling finite element simulation, the lifetime of the thermal barrier coating under different temperatures, strains, and different oxide layer elastic moduli, Poisson's ratios, and fracture toughnesses is solved to form a lifetime dataset. A deep neural network is then constructed, in which a fully connected layer is used. The input layer has 7 neurons, the hidden layer contains two layers with 100 and 30 neurons respectively, and the output layer has 1 neuron. The ReLU activation function is used, and the mean squared error loss function is used as follows. The lifetime is used as the output variable, and temperature, strain, thermal parameters, and mechanical parameters are used as input variables. The deep learning network is trained using mini-batch gradient descent, and the loss function changes during training as follows: Figure 3 It can be observed that the loss function has fallen below 0.2% after 100 training iterations, indicating that the accuracy training is complete.

[0068]

[0069] Where y d,j y is a deep learning prediction. j These are the predicted values ​​from the finite element simulation of the microstructure.

[0070] (6) Based on the deep learning proxy model trained in the previous step, the thermal barrier coating lifetime at all points in the sample space S established in step three of the reliability analysis can be predicted. Then, α and β are obtained by fitting the Weibull distribution and other functions using the least squares method. Finally, the cumulative distribution function of the lifetime at a certain coordinate point is obtained through formula (11), such as Figure 4 As can be seen, the failure probability of the thermal barrier coating in this localized area has reached as high as 80% after approximately 900 hours of service, which is extremely dangerous. By calculating all coordinate points using a loop in Python, the failure probability of the turbine blade's thermal barrier coating can be obtained. By selecting a specific moment, a cloud map of the failure probability of each region of the blade's thermal barrier coating at that moment can be obtained, as shown below. Figure 5 It can be seen that at 1000 hours, the failure probability of the thermal barrier coating on the upper right and lower left of the turbine blade is close to 1, indicating that the thermal barrier coating in this area will basically peel off at this time.

[0071]

[0072] (7) Finally, a deep learning surrogate model is used to obtain the lifetime value for each parameter with different values. Furthermore, the sensitivity of each parameter is obtained based on the variance method, thus identifying the key influencing factors affecting the reliability of the thermal barrier coating, such as... Figure 6 It can be found that fracture toughness is the parameter with the greatest impact on reliability, followed by strain, interface roughness, coefficient of thermal expansion, temperature, elastic modulus, and Poisson's ratio.

[0073] This embodiment realizes the reliability evaluation of the mechanical-thermal-chemical coupling of thermal barrier coatings on turbine blades. It can be seen that the method comprehensively considers the influence of microstructure factors of thermal barrier coatings and mechanical-thermal-chemical coupling failure mechanisms. The reliability evaluation of thermal barrier coatings on complex blades faces the challenges of many parameters, large nonlinearity, and high computational cost. The application of this method improves the accuracy and efficiency of reliability evaluation and provides an evaluation means for the safe application and optimized design of thermal barrier coatings.

Claims

1. A method for evaluating the reliability of a thermomechanical coupling of a thermal barrier coating of a turbine blade, characterized in that, The method comprises the following steps: Step one, measure the thermal parameters, mechanical parameters and microstructure parameters of the turbine blade thermal barrier coating, and calculate the mean value and standard deviation of each parameter; Step two, establish a macroscopic complex turbine blade thermal barrier coating geometry model and its calculation grid, including: using numerical simulation to obtain the temperature field and strain field of the turbine blade thermal barrier coating, and exporting the coordinates, temperature and three-direction strain of the data points of the calculation grid to an excel file; Step five, based on the failure life of the thermal barrier coating, the life of the thermal barrier coating under different temperatures, strains, thermal, mechanical and microstructure parameters is solved to form a life data set; further, a deep neural network is constructed, and the Paris life model is trained based on the life data set to obtain a deep learning proxy model; the strain includes x, y direction strain parallel to the thermal barrier coating and z direction strain perpendicular to the interface, or x, y direction strain parallel to the substrate interface and z direction strain perpendicular to the interface; and the local strain of the corresponding position of the macroscopic complex turbine blade is applied to the micro finite element boundary condition in step four to ensure the consistency of micro and macro deformation; Step three, read the temperature and strain of three directions of each coordinate point in the excel file by using the loop statement in Python to form a 4×N array; then, based on the density distribution function of M parameters measured in step one, N values are taken out respectively by using the Monte Carlo method, thereby forming a (M+4)×N dimensional array S i , i represents the i-th coordinate point; Step four, the array S obtained in step three i A row is selected, and based on the microstructure parameters of the selected row, a geometric model of the thermal barrier coating microstructure is constructed using COMSOL, and a computational mesh for the thermal barrier coating microstructure is generated. The mechanical-thermal coupling constitutive equation of the thermal barrier coating is constructed and input into COMSOL as the governing equation for solving. The temperature and material parameters of the selected row are input into COMSOL as material parameters for solving. The strain of the selected row is input into COMSOL as the boundary condition for solving, and the Mises stress is obtained through finite element calculation of the microstructure. Finally, the failure lifetime of the thermal barrier coating is obtained based on the Paris lifetime model. The microstructure of the thermal barrier coating microstructure is constructed during the modeling process. The microstructure includes the multilayer structure of the ceramic layer, oxide layer, and adhesive layer of the thermal barrier coating, as well as the interface roughness and pore microstructure. The parameters of the mechanical-thermal coupling constitutive equation of the thermal barrier coating include one or more of the following: high-temperature interface oxidation, creep, sintering, and high-temperature corrosion. Step seven, the sensitivity of M random parameters is analyzed by using the deep learning proxy model and variance method to obtain the key influencing factors affecting the reliability of the turbine blade thermal barrier coating. ； wherein is the micro-strain, is the macro-strain; Step six, predicting array S based on deep learning agent model i The life of the thermal barrier coating corresponding to all the row data in the middle is fitted with a normal distribution or a Weibull distribution to obtain the cumulative distribution function of the life of the i-th coordinate point. Further, a loop statement in Python is used to obtain the cumulative distribution function of the life of all coordinate points. The obtained cumulative distribution function of the life of the coordinate points is visualized to obtain a cloud map of the failure probability of the turbine blade thermal barrier coating serving different times, that is, the reliability evaluation of the turbine blade thermal barrier coating is realized. The thermal parameters include the thermal conductivity and thermal expansion coefficient of each layer of the thermal barrier coating; 2. The method of claim 1, wherein the method further comprises: The mechanical parameters include the elastic modulus, Poisson's ratio and fracture toughness; The microstructure parameters include the thickness, porosity and interface roughness of each layer of the thermal barrier coating; The random parameter statistical distribution function includes normal distribution, Weibull distribution, lognormal distribution, exponential distribution, gamma distribution and beta distribution. The numerical simulation in step two comprises the following steps: establishing a flow field and a macroscopic complex turbine blade thermal barrier coating geometry model, dividing the calculation grid, setting the boundary conditions and material parameters, and solving the temperature field and strain field cloud map; 3. The method of claim 1, wherein the method further comprises: The macroscopic complex turbine blade thermal barrier coating geometry model includes an internal cold gas channel and a film cooling structure, and the calculation grid includes grid refinement and boundary layer grid; The boundary conditions include the temperature, pressure and turbulence of the gas from the real engine service environment, and the temperature, pressure and turbulence of the cold gas. The life data set in step five also includes experimental spalling failure life, the turbine blade sample is placed in a preset temperature constant temperature furnace for high temperature thermal cycle and high temperature corrosion test to obtain the experimental spalling failure life of the thermal barrier coating.

4. The method of claim 1, wherein the method is used to evaluate the reliability of the TBCs of the turbine blade under the combined action of thermal and mechanical stresses. The (M+4) x N dimensional array S described in step three i is a Monte Carlo reliability space representing the i-th coordinate point, the reliability space having N points, each point containing thermal parameters, mechanical parameters and microstructure parameters, xyz triaxial strain, and the thickness of each layer of the ceramic layer, the oxide layer, and the bonding layer.

5. The method of claim 1, wherein the method is used to evaluate the reliability of the TBCs of the turbine blade under the combined action of thermal and mechanical stresses. ​ 6. The method of claim 1, wherein the method is used to evaluate the reliability of the TBCs of the turbine blade under the combined action of thermal and mechanical stresses. The step five constructs the deep learning network by using TensorFlow, which comprises an input layer, an output layer and multiple hidden layers, the number of hidden layers being adjusted according to the degree of nonlinearity; the adjacent layers in the deep neural network are connected through weights, and a ReLU activation function is used; during the training of the deep learning network, a small batch gradient descent method is used for updating, and the following selection function is used to estimate important samples S2={x(i)|i=1, 2, …, N}: ； in For the first The lower bound of the secondary important sample space For the first The upper limit of the secondary important sample space For the first Deep neural network after training; The updated labeled data set is obtained by coupling the updated samples with the thermomechanical coupling finite element life prediction, the deep neural network is trained again with the updated labeled data set, and the deep neural network is trained until the target precision requirement is reached.

7. The method of claim 6, wherein the method further comprises: The deep learning network training process in the fifth step includes numerical simulation life data and experimental spalling failure life, and the loss function L of the experimental spalling failure life is p The deep learning network is trained with N times of weight, where N>1, as follows: ； wherein, is the deep learning predicted value, is the microstructure finite element simulation predicted value.

8. The method of claim 1, wherein the method is used to evaluate the reliability of the TBCs of the turbine blade under the combined action of thermal and mechanical stresses. The cumulative distribution function of the life in the step six is a Weibull distribution described as follows: ； wherein t is the service time, F ( t ) is t the probability of failure at time t, The coordinates and failure probability of all the Monte Carlo data points are mapped onto the computational grid of the turbine blade thermal barrier coating, and the reliability nephogram is obtained by visualization software.

9. The method of claim 1, wherein the method is used to evaluate the reliability of the TBCs of the turbine blade under the combined action of thermal and mechanical stresses. The sensitivity analysis in the step seven is to obtain the life value of each parameter at different values by using the deep learning proxy model, and further obtain the sensitivity of each parameter based on the variance method, so that the sensitivity analysis is realized.

Images