Method for predicting sweating cooling performance based on regional physical information neural network
Through the prediction method based on the sub-regional physical information neural network, the problem of many variables, high cost and long time-consuming research on the sweat cooling performance of aircraft in the prior art is solved, and more efficient and accurate sweat cooling performance prediction is achieved.
Patent Information
- Application Number
- CN202411926125.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-04-25
AI Technical Summary
When studying the sweat cooling performance of aircraft, the prior art has problems such as numerous variables, high cost and long time, and it is difficult to effectively predict the impact of different porous media on sweat cooling performance.
The prediction method based on the sub-regional physical information neural network is adopted, and the physical information neural network is divided into two parts: cooling medium channel and porous medium area. The fluid heat transfer control equation and the thermal layer temperature distribution control equation are used as loss functions, and the internal connection loss function of the coupling interface is established at the interface to establish a sweat cooling performance prediction model.
It improves the accuracy of sweat cooling performance prediction and the convergence speed of the loss function, reduces cost and time-consuming, and can more accurately predict the sweat cooling performance at hypersonic speed.
Smart Images
Figure CN119885860A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of hypersonic environment thermal management, and in particular to a method for predicting sweating cooling performance based on a regional physical information neural network. Background Art
[0002] With the development of aerospace technology, advanced spacecraft basically need to perform flight missions in scenarios above the first cosmic speed. During the flight, the high-speed and high-total-temperature flow generated by the engine will generate a huge heat load on the wall of the aircraft, which may seriously damage the aircraft. Therefore, it is very important to provide efficient and reliable thermal protection for key parts of the aircraft. The sweat cooling technology originated in the 1940s and is a thermal protection method for high-temperature walls. It has extremely high cooling efficiency in theory and is considered to be an efficient active thermal protection technology.
[0003] At present, the application and research of transpiration cooling in aircraft mainly uses numerical simulation or wind tunnel experiments to study the influence of different porous media on transpiration cooling performance, which has problems such as many variables, high cost and long time consumption.
[0004] Based on this, a prediction method for sweating cooling performance based on regional physical information neural network needs to be developed urgently. Summary of the invention
[0005] The object of the present invention is to provide a method for predicting the sweating cooling performance based on a regional physical information neural network, so as to at least solve one technical problem in the prior art.
[0006] The technical solution of the present invention is:
[0007] A method for predicting sweating cooling performance based on a regional physical information neural network, comprising:
[0008] The physical information neural network is divided into two parts: the cooling medium channel and the porous medium region;
[0009] The fluid heat transfer control equation and the thermal layer temperature distribution control equation are used as the loss functions of the cooling medium channel and the porous medium region respectively;
[0010] Based on the interface between the cooling medium channel and the porous medium region, an internal connection loss function of the coupling interface is established, and a sweat cooling performance prediction model is established;
[0011] The sweating cooling performance prediction model is used to predict the target to be predicted.
[0012] The physical information neural network is divided into two parts: the cooling medium channel and the porous medium area, including:
[0013] The complex model to be predicted is divided into two parts. Each part uses a physical information neural network to perform deep learning and result prediction in different regions, and sets initial conditions.
[0014] Set the initial values of the flow field and temperature field of the cooling medium channel; set the initial values of the flow field, temperature field and porous medium properties of the porous medium; set the boundary conditions of the cooling medium channel and the porous medium area.
[0015] The initial values of the flow field and temperature field of the cooling medium channel are input into the physical information neural network model of the cooling medium channel part for processing.
[0016] The fluid heat transfer control equation is as follows:
[0017]
[0018]
[0019]
[0020]
[0021]
[0022] If the steady-state condition is:
[0023]
[0024] N is u, or v, or T;
[0025] Where u is the velocity component in the x-axis direction, v is the velocity component in the y-axis direction, t is time, and ρ L is the density of the coolant, p is the pressure of the fluid, μ is the viscosity coefficient of the coolant, T is the temperature of the coolant, κ L is the thermal conductivity of the coolant, c pL is the specific heat of the coolant, q is the heat flux transferred by the heat source, and ΔT is the temperature difference between the porous medium and the coolant.
[0026] The thermal layer temperature control equation is as follows:
[0027]
[0028]
[0029]
[0030]
[0031]
[0032] If the steady-state condition is:
[0033]
[0034] N is (ρc+φρ L c pL )T;
[0035] Where ρ is the density of the porous medium, c is the specific heat of the porous medium, φ is the porosity, and ρ L is the density of the coolant, c pL is the specific heat of the coolant, T is the porous medium temperature, T avg is the average surface temperature of the porous medium, k is the thermal conductivity of the porous medium, κ L is the thermal conductivity of the coolant, T c is the temperature of the coolant as it flows into the porous medium, is the mass flux vector of the coolant;
[0036] The upper surface of the target to be predicted is set as:
[0037]
[0038]
[0039] In the formula, q r is the heating flux of the external flow field, T w is the upper wall temperature of the porous medium, q 0 is a constant determined by the airflow boundary layer, T r is the mainstream temperature.
[0040] The method of establishing an internal connection loss function of a coupling interface based on the interface between the cooling medium channel and the porous medium region and establishing a sweating cooling performance prediction model comprises:
[0041] Define the internal connection loss function based on the coupling surface between the cooling medium channel and the porous medium region, and set the interface boundary conditions;
[0042] Define the total loss function, which includes the internal connection loss function, the data loss function of the porous medium part and the cooling medium channel part, and the partial differential equation loss function, and set the weight of each loss function;
[0043] The loss function of each prediction is returned, and the loss function is continuously optimized and iterated to achieve overall minimization. Model training is performed to obtain a sweating cooling performance prediction model.
[0044] The interface boundary condition is:
[0045] u| + =u| - =ue ;
[0046] v| + =v| - =v e ;
[0047] T| + =T| - =T e ;
[0048] Where + represents the porous medium area, - represents the cooling medium channel; the velocity component of the interface in the x-axis direction is u e , the velocity in the y-axis direction is v e , temperature is T e .
[0049] The step of using the sweating cooling performance prediction model to predict the target to be predicted includes:
[0050] Calculating the cooling efficiency of the sweating cooling model using the prediction data obtained by the sweating cooling performance prediction model;
[0051] By fitting different coolant density, coolant thermal conductivity, coolant specific heat, coolant viscosity coefficient, coolant initial flow rate, porous medium porosity, porous medium density, porous medium thermal conductivity, porous medium specific heat and mainstream temperature data, the cooling efficiency of the corresponding model is predicted, thereby obtaining the influence of different porous media on its sweating cooling performance.
[0052] The cooling efficiency calculation formula of the sweating cooling performance prediction model is as follows:
[0053] Where η is the cooling efficiency; T r is the mainstream temperature; T avg is the average surface temperature of the porous medium; T c is the temperature of the coolant as it flows into the porous medium.
[0054] For the predicted data obtained using the sweating cooling performance prediction model:
[0055] Draw the error and loss function result graphs to judge the loss function of the predicted data:
[0056] If the overall loss function is less than a preset threshold, the predicted data is used to calculate the cooling efficiency of the sweat cooling model.
[0057] The error and loss function result graphs are plotted. If the loss function is greater than the preset threshold, the loss function weights of the cooling medium channel part, the porous medium part, and the internal connection conditions are reset to achieve a prediction model that conforms to the data.
[0058] The beneficial effects of the present invention include at least:
[0059] The prediction method described in the present invention partitions the movement of the fluid in the cooling medium channel and the heat exchange of the fluid in the porous medium to improve the prediction accuracy and the convergence speed of the loss function, and sets a loss function composed of an inner connection loss function, a data loss function and a partial differential equation loss function. The effectiveness and accuracy of the model are verified in combination with numerical simulation results and real physical laws, which brings convenience to deep learning work and more accurately predicts the performance of sweat cooling under hypersonic conditions; at the same time, it reduces costs and shortens time. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 It is the structural diagram of the regional physical information neural network;
[0061] Figure 2 It is the neural network flow chart;
[0062] Figure 3 Schematic diagram of cooling medium channel and porous medium. DETAILED DESCRIPTION
[0063] The present application is further described below in conjunction with the accompanying drawings.
[0064] In order to more efficiently predict the sweating cooling performance, the prediction method using the Physics Informed Neural Network (PINN) is particularly important. PINN combines deep learning and partial differential equations, can learn physical laws from data, has wide applicability and flexibility, and is suitable for solving complex physical problems.
[0065] At present, in the field of transpiration cooling, there is no formula or prediction method that directly links the characteristics of porous media to the performance of transpiration cooling. Various parameters of porous media, such as porosity, thermal conductivity and structure, will have a significant impact on the performance of transpiration cooling, and the cost of experimenting on each parameter is extremely high. In traditional PINN deep learning, a single fully connected neural network is difficult to handle the complex model of transpiration cooling. Therefore, this embodiment proposes a prediction method for transpiration cooling performance based on a regional physical information neural network. This method not only greatly reduces time and labor costs, but also can accurately predict the cooling efficiency of different porous media and different cooling media in the transpiration cooling system, providing a precise and effective new way for quantitative evaluation of transpiration cooling performance under hypersonic conditions. Specific embodiment 1:
[0067] The technical solution adopted in this embodiment is as follows:
[0068] The method for predicting the sweating cooling performance based on the regional physical information neural network described in this embodiment includes: dividing the physical information neural network into two parts, namely, the cooling medium channel and the porous medium region; using the fluid heat transfer control equation and the thermal layer temperature distribution control equation as the loss functions of the cooling medium channel and the porous medium region respectively; establishing an internal connection loss function of the coupling interface based on the interface between the cooling medium channel and the porous medium region, and establishing a sweating cooling performance prediction model; and using the sweating cooling performance prediction model to predict the target to be predicted.
[0069] The specific steps are as follows: Figure 2 :
[0070] S1: Divide the complex model to be predicted into two parts, each of which uses a Figure 1 The physical information neural network is used to conduct deep learning and result prediction in different regions, and the initial conditions are set;
[0071] S2: Set the initial values of the flow field and temperature field of the cooling medium channel, such as Figure 3 ; Set the initial values of porous medium flow field, temperature field and porous medium properties; set the boundary conditions of cooling medium channel and porous medium area.
[0072] S3: Input the initial values of the flow field and temperature field of the cooling medium channel into the physical information neural network model of the cooling medium channel part for processing.
[0073] The flow heat transfer control equation is as follows:
[0074]
[0075]
[0076]
[0077]
[0078]
[0079] If the steady-state condition is:
[0080]
[0081] N can be u, v, or T;
[0082] Where u is the velocity component in the x-axis direction, v is the velocity component in the y-axis direction, t is time, and ρ L is the density of the coolant, p is the pressure of the fluid, μ is the viscosity coefficient of the coolant, T is the temperature of the coolant, κ L is the thermal conductivity of the coolant, c pLis the specific heat of the coolant, q is the heat flux transferred by the heat source, and ΔT is the temperature difference between the porous medium and the coolant.
[0083] S4: Define the boundary conditions of the flow heat transfer control equations and collect N according to the initial and boundary conditions. 01 、N b1 data points, according to the domain, N f1 Configuration points.
[0084] S5: A fully connected neural network is set to predict the flow field and temperature field of the cooling medium channel, 10 hidden layers are set, and 130 units are set in each hidden layer.
[0085] S6: Set the activation function to Swish.
[0086] S7: Define loss functions, including data loss function and partial differential equation loss function, and set the weights of each loss function.
[0087] S8: Set the initial value of the learning rate to 0.001.
[0088] S9: Set the optimizer to Adam.
[0089] S10: Input the porous medium flow field, temperature field and initial values of porous medium properties into the physical information neural network of the porous medium part for processing.
[0090] S11: The temperature of each point in the porous medium is obtained based on the control equation of the thermal layer temperature.
[0091] The thermal layer temperature control equation is as follows:
[0092]
[0093]
[0094]
[0095]
[0096]
[0097] If the steady-state condition is:
[0098]
[0099] N is (ρc+φρ L c pL )T;
[0100] Where ρ is the density of the porous medium, c is the specific heat of the porous medium, φ is the porosity, and ρ Lis the density of the coolant, c pL is the specific heat of the coolant, T is the porous medium temperature, T avg is the average surface temperature of the porous medium, κ is the thermal conductivity of the porous medium, L is the thermal conductivity of the coolant, T c is the temperature of the coolant as it flows into the porous medium, is the mass flux vector of the coolant.
[0101] The upper surface is set as:
[0102]
[0103]
[0104] In the formula, q r is the heating flux of the external flow field, T w is the upper wall temperature of the porous medium, q 0 is a constant determined by the airflow boundary layer, T r is the mainstream temperature.
[0105] S12: Define the boundary conditions of the thermal layer temperature control equation and collect N according to the initial and boundary conditions. 02 、N b2 data points, according to the domain, N f2 Configuration points.
[0106] S13: A physical information neural network for predicting sweating cooling performance is set, 10 hidden layers are set, and 130 units are set in each hidden layer.
[0107] S14: Set the activation function to Swish.
[0108] S15: Define the internal connection loss function based on the coupling surface between the cooling medium channel and the porous medium region, and set the interface boundary conditions.
[0109] Assume that the velocity component of the interface in the x-axis direction is u e , the velocity in the y-axis direction is v e , temperature is T e , the interface boundary conditions are:
[0110] u| + =u| - =u e ;
[0111] v| + =v| - =v e ;
[0112] T| + =T| -=T e ;
[0113] Among them, + represents the porous medium area, and - represents the cooling medium channel.
[0114] S16: define the total loss function, which includes the internal connection loss function, the data loss function of the S11 porous medium part and the S7 cooling medium channel part, and the partial differential equation loss function, and set the weight of each loss function.
[0115] S17: Set the initial value of the learning rate to 0.001.
[0116] S18: Set the optimizer to Adam.
[0117] S19: Call the model set above for training.
[0118] S20: Return the loss function of each prediction, and continuously optimize and iterate to minimize the loss function as a whole.
[0119] S21: The training steps for both parts are set to 200,000. Model training is terminated after reaching the number of steps.
[0120] S22: Run the above program to perform model training.
[0121] S23: Draw the error and loss function result graph. If the overall loss function is greater than 1×10 -7 , then return to S16 to reset the loss function weights of the cooling medium channel part, the porous medium part and the internal connection condition. If the loss function is less than 1×10 -7 , it means that the model prediction is consistent with the physical laws and given data conditions, and a set of predicted data is obtained.
[0122] S24: Calculate the cooling efficiency of the sweat cooling model using the predicted data obtained by the above model.
[0123] The cooling efficiency calculation formula of the sweat cooling model is as follows:
[0124]
[0125] Where η is the cooling efficiency.
[0126] S25: By fitting different coolant density, coolant thermal conductivity, coolant specific heat, coolant viscosity coefficient, coolant initial flow rate, porous medium porosity, porous medium density, porous medium thermal conductivity, porous medium specific heat, and mainstream temperature data, the cooling efficiency of the corresponding model is predicted, thereby obtaining the influence of different porous media on their sweating cooling performance.
[0127] The present invention sets a function with x, y, t, u, v, p, and T, divides the complex model into two parts: the cooling medium channel and the porous medium, establishes physical information neural networks for each part, uses the deep learning function of the PyTorch framework and the existing simulation results, performs regional physical information neural networks and predicts the η value, thereby accurately completing the prediction of the sweat cooling performance under hypersonic speed, and providing a certain theoretical basis for the thermal management research of sweat cooling under hypersonic speed.
[0128] The present invention combines the NS equations describing fluid momentum with the thermal layer temperature control equations describing the relationship between sweating and control temperature and writes them into the program. The movement of the fluid in the cooling medium channel and the heat exchange of the fluid in the porous medium are predicted by partition, which improves the prediction accuracy and the convergence speed of the loss function, and sets a loss function composed of an inner connection loss function, a data loss function and a partial differential equation loss function. The effectiveness and accuracy of the model are verified by combining numerical simulation results and real physical laws, which brings convenience to deep learning work and more accurately predicts the performance of sweating cooling under hypersonic speed.
[0129] The above disclosures are only several specific implementation scenarios of the present invention, but the present invention is not limited thereto, and any changes that can be thought of by a person skilled in the art should fall within the protection scope of the present invention. The above serial numbers of the present invention are only for description and do not represent the advantages and disadvantages of the implementation scenarios.
Claims
1. A method for predicting sweating cooling performance based on a regional physical information neural network, characterized in that: include: The physical information neural network is divided into two parts: the cooling medium channel and the porous medium region; The fluid heat transfer control equation and the thermal layer temperature distribution control equation are used as the loss functions of the cooling medium channel and the porous medium region respectively; Based on the interface between the cooling medium channel and the porous medium region, an internal connection loss function of the coupling interface is established, and a sweat cooling performance prediction model is established; The sweating cooling performance prediction model is used to predict the target to be predicted.
2. The method for predicting sweating cooling performance based on regional physical information neural network according to claim 1, characterized in that: The physical information neural network is divided into two parts: the cooling medium channel and the porous medium area, including: The complex model to be predicted is divided into two parts. Each part uses a physical information neural network to perform deep learning and result prediction in different regions, and sets initial conditions. Set the initial values of the flow field and temperature field of the cooling medium channel; set the initial values of the flow field, temperature field and porous medium properties of the porous medium; set the boundary conditions of the cooling medium channel and the porous medium area. The initial values of the flow field and temperature field of the cooling medium channel are input into the physical information neural network model of the cooling medium channel part for processing.
3. The method for predicting the sweating cooling performance based on the regional physical information neural network according to claim 1 is characterized in that: The fluid heat transfer control equation is as follows: If the steady-state condition is: N is u, or v, or T; Where u is the velocity component in the x-axis direction, v is the velocity component in the y-axis direction, t is time, and ρ L is the density of the coolant, p is the pressure of the fluid, μ is the viscosity coefficient of the coolant, T is the temperature of the coolant, κ L is the thermal conductivity of the coolant, c pL is the specific heat of the coolant, q is the heat flux transferred by the heat source, and ΔT is the temperature difference between the porous medium and the coolant.
4. The method for predicting the sweating cooling performance based on the regional physical information neural network according to claim 1 is characterized in that: The thermal layer temperature control equation is as follows: If the steady-state condition is: N is (ρc + φρ L c pL )T; Where ρ is the density of the porous medium, c is the specific heat of the porous medium, φ is the porosity, and ρ L is the density of the coolant, c pL is the specific heat of the coolant, T is the porous medium temperature, T avg is the average surface temperature of the porous medium, κ is the thermal conductivity of the porous medium, L is the thermal conductivity of the coolant, T c is the temperature of the coolant as it flows into the porous medium, is the mass flux vector of the coolant; The upper surface of the target to be predicted is set as: In the formula, q r is the heating flux of the external flow field, T w is the wall temperature of the porous medium, q0 is a constant determined by the air flow boundary layer, T r is the mainstream temperature.
5. The method for predicting the sweating cooling performance based on the regional physical information neural network according to claim 1 is characterized in that: The method of establishing an internal connection loss function of a coupling interface based on the interface between the cooling medium channel and the porous medium region and establishing a sweating cooling performance prediction model comprises: Define the internal connection loss function based on the coupling surface between the cooling medium channel and the porous medium region, and set the interface boundary conditions; Define the total loss function, which includes the internal connection loss function, the data loss function of the porous medium part and the cooling medium channel part, and the partial differential equation loss function, and set the weight of each loss function; The loss function of each prediction is returned, and the loss function is continuously optimized and iterated to achieve overall minimization. Model training is performed to obtain a sweating cooling performance prediction model.
6. The method for predicting the sweating cooling performance based on the regional physical information neural network according to claim 1 is characterized in that: The interface boundary condition is: in| + =in| - =in e ; v| + =v| - =v e ; T| + =T| - =T e ; Where + represents the porous medium area, - represents the cooling medium channel; the velocity component of the interface in the x-axis direction is u e , the velocity in the y-axis direction is v e , temperature is T e .
7. The method for predicting the sweating cooling performance based on the regional physical information neural network according to claim 1, characterized in that: The step of using the sweating cooling performance prediction model to predict the target to be predicted includes: Calculating the cooling efficiency of the sweating cooling model using the prediction data obtained by the sweating cooling performance prediction model; By fitting different coolant density, coolant thermal conductivity, coolant specific heat, coolant viscosity coefficient, coolant initial flow rate, porous medium porosity, porous medium density, porous medium thermal conductivity, porous medium specific heat and mainstream temperature data, the cooling efficiency of the corresponding model is predicted, thereby obtaining the influence of different porous media on its sweating cooling performance.
8. The method for predicting the sweating cooling performance based on the regional physical information neural network according to claim 7 is characterized in that: The cooling efficiency calculation formula of the sweating cooling performance prediction model is as follows: Where η is the cooling efficiency; T r is the mainstream temperature; T avg is the average surface temperature of the porous medium; T c is the temperature of the coolant as it flows into the porous medium.
9. The method for predicting the sweating cooling performance based on the regional physical information neural network according to claim 7 is characterized in that: For the predicted data obtained using the sweating cooling performance prediction model: Draw the error and loss function result graphs to judge the loss function of the predicted data: If the overall loss function is less than a preset threshold, the predicted data is used to calculate the cooling efficiency of the sweat cooling model.
10. The method for predicting the sweating cooling performance based on the regional physical information neural network according to claim 7, characterized in that: The error and loss function result graphs are plotted. If the loss function is greater than the preset threshold, the loss function weights of the cooling medium channel part, the porous medium part, and the internal connection conditions are reset to achieve a prediction model that conforms to the data.