A Fast Solution Method for the Inverse Heat Transfer Problem of a Nano Thermal Insulation Composite Material

Abstract

The present invention discloses a fast solution method for the inverse heat transfer problem of a nano-thermal insulation composite material. Based on the deep neural network algorithm, the present invention establishes a mapping relationship between the temperature response of the nano-thermal insulation composite material and its thermal physical properties. Furthermore, the particle swarm algorithm is used to optimize the hyperparameters of the deep neural network, improving the fitting accuracy of the neural network for this mapping relationship, thereby realizing the fast solution of the inverse heat transfer problem of the nano-thermal insulation composite material. The present invention has both theoretical significance and practical application value. When the thermal physical properties of the composite material cannot be measured experimentally due to the limitations of the use conditions, the solution method for the inverse heat transfer problem of the nano-thermal insulation composite material developed based on the present invention can quickly regress accurate results of the material's thermal physical properties through the real-time measurement data of the material temperature, providing theoretical guidance for determining the safety and stability of the thermal insulation structure under the use conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of thermal protection for aircraft, and particularly relates to a rapid solution method for the inverse heat transfer problem of nano-insulating composite materials. Background Technique

[0002] Nano-insulating composite materials have physical properties such as light weight, high porosity, and high specific surface area, and have a low thermal conductivity (0.049 W·m -1 ·K -1 −1, 1000 °C), and are super materials that can meet the insulation requirements of equipment under complex and harsh high-temperature and high-heat flux conditions. To clarify the performance of nano-insulating composite materials and guide the optimal design of material insulation performance, current research has developed methods for obtaining the thermal physical properties of nano-insulating composite materials through various means such as experimental measurement, theoretical modeling, and numerical simulation. In the actual application scenario of the passive thermal protection system on the surface of near-space aircraft, complex internal and external conditions will have various effects on the insulation performance of nano-insulating composite materials, including the influence of microcracks in the composite material under thermal and mechanical loads, and the influence of physical and chemical reactions occurring in the composite material on the heat transfer process, etc. The existing developed models are still difficult to comprehensively consider the influence of multi-scale pore structure changes and physical and chemical reactions in the material on the equivalent thermal conductivity of the material. From the perspective of experimental measurement, due to the limitation of the space conditions of the use scenario of nano-insulating composite materials, it is difficult to arrange complex and expensive thermal physical property measurement devices, etc., so it is difficult to measure the thermal physical properties of this material during its use stage.

[0003] Under the actual use conditions of nano-thermal insulation composite materials, the back temperature experiment is often used to dynamically measure the temperature response inside and on the back of the material, and then qualitatively analyze the performance of the thermal insulation system under the use conditions. However, the back temperature experiment cannot directly measure the thermal physical properties of the material, making it difficult to provide guidance for the optimization of the material preparation process. Therefore, in order to clarify the true thermal physical properties of nano-thermal insulation composite materials in the service environment, scholars have gradually focused on solving the inverse problem of heat and mass transfer inside the material based on the back temperature experiment results of the material, and then calculating and obtaining the thermal physical properties of the material. Reviewing the current traditional methods for solving the inverse problem of heat transfer process in thermal insulation materials, according to their different calculation principles, they can be mainly divided into two categories, including the iterative method and the evolutionary method. However, both methods have the problems of time-consuming calculation and low regression accuracy. In recent years, with the rapid development of machine learning algorithms, the deep neural network algorithm, as an important branch of machine learning algorithms, can quickly and accurately represent the complex functional relationship between high-dimensional arrays. Therefore, the application of deep neural networks in solving the inverse heat transfer problem has great prospects. Czél et al. established a solution model for the inverse heat transfer problem in a cylindrical solid structure using a single-hidden-layer artificial neural network, generated noisy and noiseless material temperature response data, and obtained the regression results of the specific heat capacity at constant volume and thermal conductivity of the material. Liu et al. solved the inverse radiation transfer problem inside a semi-transmissive material using an artificial neural network and predicted the thermal conductivity constant and reflectivity constant of the material under steady-state conditions. It can be found in the above studies that there is still a certain residual between the neural network prediction results and experimental temperature response results, and the problem that the deep neural network is prone to falling into local minima during the optimization process has not been solved. The above problems will all affect the regression accuracy of the inverse problem solver of the deep neural network for the thermal physical properties of nano-thermal insulation composite materials during use. Summary of the Invention

[0004] Aiming at the deficiencies in the current numerical regression methods for the thermal physical properties of nano-thermal insulation composite materials, the present invention proposes a fast solution method for the inverse heat transfer problem of nano-thermal insulation composite materials based on the particle swarm optimization-based deep neural network algorithm. Under the limitation of use conditions, when the thermal physical properties of the composite material cannot be measured experimentally, based on the inverse problem solution method developed in the present invention, the accurate results of the thermal physical properties of the material can be quickly regressed through the real-time temperature measurement data of the material, and theoretical guidance can be provided for judging the safety and stability of the thermal insulation system under the use conditions.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A fast solution method for the inverse heat transfer problem of nano-thermal insulation composite materials, comprising the following steps:

[0007] Step 1) Under normal temperature and pressure conditions, test the thermal properties of the nano-insulating composite material to obtain the range of its thermal property changes; in the working conditions where it is impossible to test the thermal properties of the material due to usage conditions, measure the temperature response curves of the heated surface, back surface, and middle surface of the material during service by arranging thermocouple measurement points.

[0008] Step 2) According to the range of changes in the thermal properties of the nano-insulating composite material, randomly generate pseudo-materials with thermal properties similar to those of the nano-insulating composite material; simulate the heat transfer process inside the pseudo-materials, establish the control equation for the heat transfer process inside the pseudo-materials, that is, the numerical model for solving the forward problem, and during the numerical solution of the forward problem of heat transfer inside the pseudo-materials, use the temperature responses of the heated surface and back surface of the nano-insulating composite material during use as boundary conditions, and use the initial temperature distribution of the nano-insulating composite material as the initial condition to calculate and obtain the temperature response curve of the middle surface of the pseudo-materials under the same usage conditions, and establish the forward mapping relationship between the thermal properties of the pseudo-materials and the temperature response.

[0009] Step 3) Repeat Step 2) to obtain multiple forward mapping relationships between the thermal properties of the pseudo-materials and the material temperature response; swap the input and output positions in this forward mapping relationship to form a data set of the corresponding relationship between the temperature response and thermal properties of the pseudo-materials.

[0010] Step 4) Use the data set to train a deep neural network, and based on the trained deep neural network, establish the inverse mapping relationship between the temperature response curve of the middle surface of the nano-insulating composite material and its thermal properties. Through this inverse mapping relationship, form the basic solver for the heat transfer inverse problem of the nano-insulating composite material.

[0011] Step 5) Use the particle swarm algorithm to optimize the hyperparameters of the deep neural network in the basic solver, further find a deep neural network heat transfer inverse problem solver with higher accuracy, and finally combine the experimental results of the temperature response of the middle surface of the nano-insulating composite material to determine the true change process of the thermal properties of the material under usage conditions.

[0012] Compared with the prior art, the beneficial effects of the present invention are:

[0013] 1) In the process of solving the heat transfer inverse problem of the nano-insulating composite material, the deep neural network algorithm is introduced, providing a faster fitting method for establishing the inverse mapping relationship between the material temperature response curve and its thermal properties, and realizing the real-time acquisition of the thermal properties of the insulation system during service.

[0014] 2) The particle swarm optimization algorithm is used to optimize the hyperparameters in the deep neural network, improving the regression accuracy of the deep neural network for the thermal properties of the nano-insulating composite material.

[0015] In summary, the present invention can provide a rapid solution method for the inverse heat transfer problem of a nano-insulating composite material. Based on this method, when the thermal properties of the composite material cannot be measured experimentally due to usage conditions, the dynamic thermal property parameters of the material can be quickly and accurately regressed from the temperature measurement data of the material, providing theoretical guidance for determining the safety and stability of the nano-insulating structure under usage conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the present invention.

[0017] Figure 2 are the relevant parameters of the pseudo-material in the data sample set and the relationship with the experimental results, where (a) is the value range of the equivalent thermal conductivity of the pseudo-material, and (b) is the value range of the temperature response of the pseudo-material.

[0018] Figure 3 is the particle swarm optimization process of the prediction error of the thermal properties of the aerogel composite material by the present invention, where (a) is the change process of the regression error of the thermal conductivity, and (b) is the change process of the regression error of the temperature.

[0019] Figure 4 is the solution result of the inverse heat transfer problem of the aerogel composite material by the present invention, where (a) is the regression result of the equivalent thermal conductivity, and (b) is the prediction result of the temperature response curve. DETAILED DESCRIPTION OF THE INVENTION

[0020] The following combines the drawings and embodiments to give a detailed description of the specific implementation manners of the present invention.

[0021] As Figure 1 shown, a rapid solution method for the inverse heat transfer problem of a nano-insulating composite material of the present invention can quickly and accurately solve the inverse heat transfer problem of the nano-insulating composite material, and the specific steps are as follows:

[0022] Step 1: Under normal temperature and pressure conditions, test the thermal properties of the nano-insulating composite material to obtain the approximate value range of the material's thermal properties, that is, the thermal property change range. In the case where the thermal property experiment of the insulation system cannot be carried out due to usage conditions and the thermal properties of the material cannot be tested, by arranging thermocouple measurement points, measure the temperature response curves of the heated surface, back surface, and middle surface of the nano-insulating composite material during service.

[0023] Step 2: According to the experimental results of the thermal property change range of the nano-insulating composite material under normal temperature and pressure conditions, randomly generate a pseudo-material whose thermal properties are similar to those of the nano-insulating composite material. The generation method of its thermal properties is as follows:

[0024]

[0025] In the formula, ψreal represents any thermal property parameter of the nano thermal insulation composite material, ψ i * represents the thermal property parameter of the i-th pseudo material, Δψ i * represents the random deviation of the thermal property of the i-th pseudo material. Furthermore, for the pseudo material with random thermal properties, simulate the internal heat transfer process, and establish the control equation for the internal heat transfer process of the pseudo material, that is, the numerical model for solving the forward problem. The mathematical model for the forward problem of heat transfer inside the pseudo material established in the present invention is as follows:

[0026]

[0027]

[0028] In the formula, T represents temperature, σ B is the Stefan-Boltzmann constant, n and κ R,m respectively represent the equivalent refractive coefficient and Rosseland equivalent spectral absorption coefficient of the nano thermal insulation composite material, ρ(T), c p (T), k c (T) and k eff (T) respectively represent the density, specific heat capacity at constant pressure, thermal conductivity of the material when radiation heat transfer is ignored, and equivalent thermal conductivity when radiation heat transfer is considered of the nano thermal insulation composite material, and the above four are all functions of temperature.

[0029] In the specific application of the present invention, the finite volume method can be used to discretely solve the heat transfer equation inside the pseudo material, and combined with the temperature boundary conditions of the heating surface and the back surface of the nano thermal insulation composite material under the use conditions, the temperature response curve of the middle surface of the pseudo material under the same use conditions can be obtained. Specifically, in the calculation process, the temperature responses of the heating surface and the back surface of the nano thermal insulation composite material during the service process can be used as boundary conditions, and the temperature distribution inside the material before the start of the service process (i.e., the initial temperature distribution) can be used as the initial condition. Based on the numerical simulation results, the temperature response curve of the middle surface of the pseudo material under the same use conditions can be obtained, and the following forward mapping relationship between the thermal properties of the pseudo material and the temperature response can be established:

[0030] T ps,i = Φ(ψ ps,i )

[0031] In the formula, T ps,i represents the temperature response curve of the middle surface of the i-th pseudo material, ψ ps,i represents the thermal property parameter of the i-th pseudo material, and Φ represents the forward mapping relationship between the thermal properties and the temperature response.

[0032] Step 3: By continuously repeating Step 2, a large number of positive mapping relationships between the thermophysical properties of the pseudo-materials and their temperature responses can be obtained. To ensure the accuracy of the inverse problem solution, the equivalent thermal conductivity and temperature response curve of the nano-insulating composite material must be within the range of variation of the corresponding dataset, that is, within the range of variation of the equivalent thermal conductivity and temperature response curve of the pseudo-materials in the established dataset. After swapping the positions of the input and output in the mapping relationship, a dataset of the corresponding relationship between the temperature response and thermophysical properties of the pseudo-materials is formed.

[0033] Step 4: For the heat transfer inverse problem within the nano-insulating composite material, a deep neural network solution model is established. The above dataset is used to train the deep neural network, and based on the trained deep neural network, an inverse mapping relationship between the temperature response curve and thermophysical properties of the middle surface of the nano-insulating composite material is established. Through this inverse mapping relationship, a basic solver for the heat transfer inverse problem of the nano-insulating composite material is formed.

[0034] The specific establishment process of the inverse mapping relationship in this step can be described as follows:

[0035] First, a neural network model suitable for describing the inverse mapping relationship between the temperature response curve and thermophysical properties is established. This model has at least an input layer, an output layer, and two hidden layers. The number of neurons in the input layer and output layer is set to the number of data points in the temperature measurement process and the dimension of the thermophysical parameters to be regressed, respectively. The hyperparameters in the deep neural network, that is, the initial values of the neuron weights and biases, are taken from a uniform distribution within the range of [-(6 / n in ) 0.5 , (6 / n in ) 0.5 , where n in is the number of data entering this neuron;

[0036] Second, the thermophysical properties and temperature response data of the pseudo-materials in the dataset are normalized. The processing method is as follows:

[0037]

[0038] In the formula, m represents the dimension of the thermophysical property or temperature response vector, and the subscript k is the traversal variable.

[0039] Third, two indicators for measuring the error of the deep neural network inverse problem solver are established, namely the thermophysical property regression error ζ ψ and the temperature response regression error ζ T . The calculation formulas are as follows:

[0040]

[0041]

[0042] Where N s is the number of samples in the test set, and N T represents the number of interpolation points set when regressing the thermal properties. ψ d and ψ p correspond to the true values of the thermal properties of the pseudo-material in the test set and the regression results of the thermal properties of the pseudo-material obtained by using the inverse problem solver, respectively; N t represents the number of interpolation points in the unsteady temperature response curve of the nano-insulating composite material. T exp and T p correspond to the true values of the temperature response of the pseudo-material in the test set and the predicted results of the temperature response of the pseudo-material obtained by using the inverse problem solver, respectively. i and j are traversal variables.

[0043] Finally, combined with the normalized pseudo-material data set, the gradient descent backpropagation algorithm is used to iterate the weights and biases of the neurons in the deep neural network. As the error index of the deep neural network gradually converges, the following reflection relationship between the intermediate surface temperature response curve and the thermal properties of the nano-insulating composite material is obtained:

[0044]

[0045] Where w and b are the weights and biases of the neurons in the deep neural network at convergence, N represents the number of neurons, and N op , N hd1 and N hd2 are the numbers of neurons in the output layer, the first hidden layer, and the second hidden layer, respectively. j, k, and l are all traversal variables, and the ReLU function is the activation function of the neurons.

[0046] Based on the established deep neural network solution model, the following reflection relationship between the intermediate surface temperature response curve and the thermal properties of the nano-insulating composite material can be obtained:

[0047] ψ pred = Θ(T c )

[0048] Where T c represents the temperature response curve of the nano-insulating composite material, ψ pred represents the predicted results of the thermal properties, and Θ represents the reflection relationship between the temperature response and the thermal properties.

[0049] Step 5: To improve the fitting accuracy of the reflection relationship between the intermediate surface temperature response curve and the thermal properties of the nano-insulating composite material by the deep neural network, the particle swarm optimization algorithm is used to optimize the hyperparameters of the deep neural network in the basic solver to form a more accurate heat transfer inverse problem solution model, that is, in the basic solver. Finally, combined with the experimental results of the intermediate surface temperature response of the nano-insulating composite material, the true change process of the thermal properties of the material under the use conditions is determined.

[0050] Specifically, in the process of optimizing the hyperparameters in the deep neural network using the particle swarm optimization algorithm, the hyperparameters in the deep neural network are spanned into a high-dimensional space, and a certain particle swarm is considered. The coordinates X and the movement speed V of each particle can both correspond to random positions in the high-dimensional space, as shown in the following formula:

[0051]

[0052]

[0053] In the formula, N w and N b respectively represent the numbers of weights and biases in the neural network. Both α and β are random parameters, and the superscript 0 represents the initial value. The fitness of each particle in the particle swarm can be calculated according to the solution error of the inverse problem by the deep neural network formed by the high-dimensional coordinates of the particle swarm as hyperparameters. Combining the group learning and information sharing principles in the particle swarm optimization algorithm, the optimization process of the particle in the high-dimensional space will be jointly determined according to the historical optimal solution of the particle itself and the current optimal solution of the particle swarm. Then, in the iterative process of the particle swarm optimization, the iterative update methods of the particle speed and position are as follows:

[0054]

[0055]

[0056] In the formula, both r1 and r2 are random parameters, and the value range is [0, 1]. c1 and c2 are acceleration learning constants. ω is a learning factor. P and G are respectively the individual optimal value of the particle and the global optimal value of the particle swarm in the current optimization iteration step. As the particle coordinates continue to evolve, the above two optimal values are also continuously updated with the progress of the optimization process, and the method is as follows:

[0057]

[0058]

[0059] In the formula, N p is the number of optimization steps, ζ is the error index of the prediction result of the deep neural network, and k represents the current iteration step. As the optimization process progresses, the global optimal value of the particle swarm will be continuously optimized, and finally the optimized hyperparameter value will be obtained. Substituting the optimization result into the deep neural network structure, that is, for the heat transfer inverse problem of the nano-insulating composite material, a fast and accurate deep neural network solution model is formed. Combining the intermediate surface temperature response curve data of the nano-insulating composite material obtained in step 1, the true change process of the thermal properties of the insulating material under the service conditions can be quickly and accurately regressed through this solution model.

[0060] In a specific embodiment of the present invention, to solve the inverse heat conduction problem in a typical nano-thermal insulation composite material, namely an aerogel composite material, the steps are as follows: Step 1: Under normal temperature and pressure conditions, test the thermal physical properties of the aerogel composite material to obtain a rough range of the material's thermal physical property values; and through experimental methods, the temperature response curve of the middle surface of the material was tested when the peak temperature of the material surface was 1100K and the bottom of the material was an adiabatic boundary.

[0061] Step 2: According to the experimental results of the variation range of the thermal physical properties of the aerogel composite material under normal temperature and pressure conditions, randomly generate a pseudo-material similar to the aerogel composite material. Furthermore, for the pseudo-material with random thermal physical properties, establish a mathematical model for the forward heat transfer problem inside the material.

[0062] The density and specific heat capacity of the aerogel composite material can be obtained according to relevant experiments, and the calculation formula for the specific heat capacity at constant volume (c V =ρ·c p ) is as follows:

[0063] ρ(T)c p (T)=94922.5 + 985.8T - 8.229×10 -1 T 2 +2.41×10 -4 T 3 / MJ·m -3 ·K -1

[0064] After that, use the finite volume method to discretely solve the heat transfer equation inside the pseudo-material and establish the following forward mapping relationship between the equivalent thermal conductivity and the temperature response:

[0065] T ps,i =Φ(k ps,i )[[ID=3?]]

[0066] In the formula, k ps represents the equivalent thermal conductivity of the pseudo-material.

[0067] Step 3: By continuously repeating Step 2, a large number of forward mapping relationships between the equivalent thermal conductivity of the pseudo-material and its temperature response can be obtained. To ensure the accuracy of the inverse problem solution, the equivalent thermal conductivity and the temperature response curve must be included within the variation range of the corresponding data set. In this example, the inclusion relationships between the variation ranges of the equivalent thermal conductivity and the temperature response curve in the pseudo-material data set and the experimental values of the material and the predicted values in the literature are respectively as Figure 2 shown in (a) and (b). After swapping the positions of the input and output in the mapping relationship, a data set of the corresponding relationship between the temperature response of the pseudo-material and the equivalent thermal conductivity is formed.

[0068] Step 4: Establish a deep neural network solution model for the heat transfer inverse problem within the aerogel composite material.

[0069] Step 5: Use the particle swarm optimization algorithm to optimize the hyperparameters within the deep neural network, forming a more accurate solution model for the heat transfer inverse problem.

[0070] The variation process of the solution accuracy of the neural network solver for the heat transfer inverse problem of the aerogel composite material is as Figure 3 shown. As Figure 3 shown in (a) and (b) therein, as the particle swarm optimization process progresses, the regression errors of the solver for the equivalent thermal conductivity of the material and the temperature response curve continuously shrink. In the face of random initial hyperparameter values, through the particle swarm optimization process, the regression accuracy of the deep neural network inverse problem solver will converge to a certain global minimum. The above results indicate that the convergence result of the particle swarm optimization is independent of the initial values of the deep neural network hyperparameters, proving that the particle swarm optimization algorithm adopted in the present invention can complete the iterative optimization of the deep neural network hyperparameters and can improve the solution accuracy of the deep neural network for the heat transfer inverse problem.

[0071] Based on the particle swarm optimization deep neural network algorithm, the solution results for the heat transfer inverse problem of the aerogel composite material are as Figure 4 shown. Figure 4 (a) therein shows the regression results of the present method for the equivalent thermal conductivity of the material. It can be found that the regression accuracy of the deep neural network algorithm for the thermal conductivity (<4%) is already much lower than the traditional GA regression error. On this premise, the regression accuracy of the particle swarm optimization deep neural network algorithm for the equivalent thermal conductivity of the material can be lower than 1.5%. After particle swarm optimization, the accuracy of the deep neural network inverse problem solution model is significantly improved. The inverse problem solver established by combining the two is an effective tool for establishing the relationship between the temperature response of the material and its thermal properties in the operating environment. The comparison between the predicted results of the present method for the material temperature response curve and the experimental values is as Figure 4 shown in (b) therein. As can be seen from Figure 4 (b) therein, the absolute prediction error of the present method for the temperature response curve is 2.6K, and the relative error is less than 0.4%, which once again proves the accuracy of the solution results of the present method for the heat transfer inverse problem. From the perspective of calculation time consumption, the training time of the deep neural network is about 5 minutes, and using the trained neural network to solve the inverse problem is often instantaneous. The above results indicate that the present method has obvious advantages in terms of convenience and accuracy compared with the traditional iterative method or evolutionary method for solving the heat transfer inverse problem.

[0072] In summary, the present invention proposes a fast solution method for the inverse heat conduction problem of a nano-thermal insulation composite material. Under the condition that the thermal properties of the composite material cannot be measured experimentally due to usage conditions, the inverse heat conduction problem solution method developed based on the present invention can quickly regress accurate results of the thermal properties of the material through real-time temperature measurement data of the material, providing theoretical guidance for determining the safety and stability of the nano-thermal insulation structure under usage conditions.

Claims

1. A rapid solution method for the inverse heat transfer problem of a nano thermal insulation composite material, characterized in that, It includes the following steps: Step 1) Under normal temperature and pressure conditions, test the thermal properties of the nano thermal insulation composite material to obtain the range of its thermal property changes; in the working conditions where it is impossible to test the thermal properties of the material due to usage conditions, by arranging thermocouple measuring points, measure the temperature response curves of the heated surface, back surface, and middle surface of the material during service; Step 2) According to the range of changes in the thermal properties of the nano thermal insulation composite material, randomly generate pseudo materials with thermal properties similar to those of the nano thermal insulation composite material; simulate the internal heat transfer process of the pseudo materials, establish the control equation for the internal heat transfer process of the pseudo materials, that is, the numerical model for solving the forward problem, and during the numerical solution process of the forward problem of internal heat transfer in the pseudo materials, use the temperature responses of the heated surface and back surface of the nano thermal insulation composite material during use as boundary conditions, and use the initial temperature distribution of the nano thermal insulation composite material as the initial condition to calculate and obtain the temperature response curve of the middle surface of the pseudo materials under the same usage conditions, and establish the forward mapping relationship between the thermal properties of the pseudo materials and the temperature response; Step 3) Repeat Step 2) to obtain multiple forward mapping relationships between the thermal properties of the pseudo materials and the material temperature response; Swap the input and output positions in this forward mapping relationship to form a data set of the corresponding relationship between the temperature response and thermal properties of the pseudo materials; Step 4) Use the data set to train a deep neural network, and based on the trained deep neural network, establish the inverse mapping relationship between the temperature response curve of the middle surface of the nano thermal insulation composite material and the thermal properties. Through this inverse mapping relationship, form the basic solver for the heat transfer inverse problem of the nano thermal insulation composite material; Step 5) Use the particle swarm algorithm to optimize the hyperparameters of the deep neural network in the basic solver, further search for a deep neural network heat transfer inverse problem solver with higher accuracy, and finally combine the experimental results of the temperature response of the middle surface of the nano thermal insulation composite material to determine the true change process of the thermal properties of the material under usage conditions.

2. The rapid solution method for the inverse heat transfer problem of the nano-thermal insulation composite material according to claim 1, characterized in that, In Step 2), the generation method of the thermal properties of the pseudo materials is as follows: ψ i * = ψ real + Δψ i * In the formula, ψ real represents any thermal property parameter of the nano thermal insulation composite material, ψ i * represents the thermal property parameter of the i-th pseudo material, Δψ i * represents the random deviation of the thermal property of the i-th pseudo material.

3. The rapid solution method for the inverse heat transfer problem of the nano thermal insulation composite material according to claim 1, characterized in that, In Step 2), the mathematical model for the forward problem of internal heat transfer in the pseudo materials is as follows: where T represents temperature, σ B is the Stefan-Boltzmann constant, n and κ R,m respectively represent the equivalent refractive index and the Rosseland equivalent spectral absorption coefficient of the nano thermal insulation composite material, ρ(T), c p (T), k c (T) and k eff (T) respectively represent the density, specific heat capacity at constant pressure, thermal conductivity of the material ignoring radiative heat transfer, and equivalent thermal conductivity considering radiative heat transfer of the nano thermal insulation composite material, and all four are functions of temperature.

4. The rapid solution method for the inverse heat transfer problem of the nano-thermal insulation composite material according to claim 3, characterized in that, Use the finite volume method to discretely solve the mathematical equation for the forward problem of internal heat transfer in the pseudo materials, and combine the temperature boundary conditions of the heated surface and back surface of the nano thermal insulation composite material under usage conditions to obtain the temperature response curve of the middle surface of the pseudo materials under the same usage conditions.

5. The rapid solution method for the inverse heat transfer problem of the nano thermal insulation composite material according to any one of claims 1 to 4, characterized in that In Step 2), the forward mapping relationship between the thermal properties of the pseudo materials and the temperature response is established as follows: T ps,i = Φ(ψ ps,i ) where, T ps,i represents the temperature response curve of the middle surface of the i-th pseudo material, and ψ ps,i represents the thermophysical property parameter of the i-th pseudo material, and Φ represents the positive mapping relationship between the thermophysical property and the temperature response.

6. The rapid solution method for the inverse heat transfer problem of the nano thermal insulation composite material according to claim 1, characterized in that, When establishing the data set of the corresponding relationship between the temperature response and thermal properties of the pseudo materials in Step 3), to ensure the accuracy of the inverse problem solution, the equivalent thermal conductivity and temperature response curve of the nano thermal insulation composite material are included within the range of changes in the equivalent thermal conductivity and temperature response curve of the pseudo materials in the established data set.

7. The rapid solution method for the inverse heat transfer problem of the nano-thermal insulation composite material according to claim 1, wherein The inverse mapping relationship is expressed as: ψ pred = Θ(T c ) where, T c represents the temperature response curve of the middle surface of the nano thermal insulation composite material, ψ pred represents the prediction result of the thermal physical properties, and Θ represents the reflection relationship between the temperature response and the thermal physical properties.

8. The rapid solution method for the inverse heat transfer problem of the nano thermal insulation composite material according to claim 1 or 7, characterized in that The establishment process of the inverse mapping relationship is as follows: First, a neural network model suitable for describing the reflection relationship between the temperature response curve and the thermal properties is established. This model has at least an input layer, an output layer, and two hidden layers. The number of neurons in the input layer and the output layer are respectively set to the number of data points in the temperature measurement process and the dimension of the thermal property parameters to be regressed; the hyperparameters in the deep neural network, that is, the initial values of the neuron weights and biases, are within [-(6 / n in ) 0.5 ,(6 / n in ) 0.5 and are taken from a uniform distribution, where n in is the number of data input to this neuron; Secondly, normalize the thermal property and temperature response data of the pseudo materials in the data set, and the processing method is as follows: In the formula, m represents the dimension of the thermal property or temperature response vector, and the subscript k is the traversal variable; Next, two metrics for measuring the error of the deep neural network inverse problem solver are established, namely the thermophysical property regression error ζ ψ and the temperature response regression error ζ T , and the calculation formulas are as follows: Where N s is the number of samples in the test set, N T represents the number of interpolation points set when regressing the thermal properties, ψ d and ψ p correspond to the true value of the thermal properties of the pseudo-material in the test set and the regression result of the thermal properties of the pseudo-material obtained by using the inverse problem solver, respectively; N t represents the number of interpolation points in the unsteady temperature response curve of the nano-insulating composite material, T exp and T p correspond to the true value of the temperature response of the pseudo-material in the test set and the predicted result of the temperature response of the pseudo-material obtained by using the inverse problem solver, respectively, and i and j are traversal variables; Finally, combined with the normalized pseudo-material data set, the gradient descent backpropagation algorithm is used to iterate the weights and biases of the neurons in the deep neural network. As the error index of the deep neural network gradually converges, the following reflection relationship between the intermediate surface temperature response curve of the nano-thermal insulation composite material and its thermal properties is obtained: Wherein, w and b are respectively the weights and biases of neurons in the deep neural network at convergence, N represents the number of neurons, N op , N hd1 and N hd2 are respectively the numbers of neurons in the output layer, the first hidden layer, and the second hidden layer, j, k, and l are all traversal variables, and the ReLU function is the activation function of neurons.

9. The rapid solution method for the inverse heat transfer problem of the nano thermal insulation composite material according to claim 8, characterized in that The particle swarm optimization algorithm is used to optimize the hyperparameters in the deep neural network, and the process is as follows: The hyperparameters in the deep neural network are spanned into a high-dimensional space, and a certain particle swarm is considered. The coordinates X and the movement speed V of each particle correspond to random positions in the high-dimensional space, as shown in the following formula: Where N w and N b respectively represent the numbers of weights and biases in the neural network. Both α and β are random parameters, and the superscript 0 represents the initial value. The optimization process of the particle in the high-dimensional space is jointly determined according to the historical optimal solution of the particle itself and the existing optimal solution of the particle swarm. In the iterative process of particle swarm optimization, the iterative update methods of particle velocity and position are as follows: In the formula, r1 and r2 are both random parameters, and the value ranges are [0, 1]. c1 and c2 are acceleration learning constants, ω is a learning factor, P and G are the individual optimal value of the particle and the global optimal value of the particle swarm in this optimization iteration step, respectively. As the particle coordinates continue to evolve, the above two optimal values are also continuously updated with the progress of the optimization process, and the method is as follows: where N p is the number of optimization steps, ζ is the error index of the prediction result of the deep neural network, k represents the current iteration step. As the optimization process progresses, the global optimal value of the particle swarm will be continuously optimized, and finally the optimized hyperparameter values will be obtained. Substitute the optimization results into the deep neural network structure, that is, for the heat transfer inverse problem of nano-insulating composite materials, a fast and accurate deep neural network solution model is formed.

Images