Numerical simulation method, device, equipment and storage medium for two-phase flow boiling heat transfer based on deep learning

Abstract

This invention provides a method, apparatus, device, and storage medium for numerical simulation of two-phase flow boiling heat transfer based on deep learning, relating to the field of heat transfer technology. The method includes: inputting target input parameters into a target deep neural network model to obtain key parameters of bubble detachment characteristics; constructing a wall heat flux distribution model based on these key parameters; the key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, slip velocity, slip distance, rising diameter, and nucleation density; embedding the bubble wall heat flux distribution model into an Euler-Euler two-fluid framework to obtain an embedded two-fluid framework; the embedded two-fluid framework includes mass conservation equations, momentum conservation equations, and energy conservation equations for the liquid and vapor phases; and performing numerical simulation of two-phase flow boiling heat transfer using the embedded two-fluid framework to obtain the temperature and velocity field distributions of the liquid and vapor phases. This invention has high applicability for numerical simulation of macroscopic boiling phenomena.

Description

Technical Field

[0001] This invention relates to the field of heat transfer technology, and in particular to a method, apparatus, equipment, and storage medium for numerical simulation of two-phase flow boiling heat transfer based on deep learning. Background Technology

[0002] Two-phase flow boiling heat transfer involves both microscopic bubble dynamics (such as bubble nucleation, growth, slip, and detachment) and macroscopic physical field properties (such as flow field and temperature field distribution), making it a typical multi-scale coupled and strongly nonlinear process. Numerical simulation of two-phase flow boiling heat transfer is therefore crucial.

[0003] The traditional numerical simulation method for two-phase flow is the Volume of Fluid (VOF) method. VOF is one of the core methods in Computational Fluid Dynamics (CFD) for simulating the interfacial dynamics of multiphase flows, and it is particularly suitable for tracing the interfaces of immiscible fluids such as gas-liquid and liquid-liquid interfaces. It can be used to study the motion behavior of individual bubbles and is suitable for studying microscale boiling mechanisms, but its applicability to macroscopic boiling phenomena is poor. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a method, apparatus, device, and storage medium for numerical simulation of two-phase flow boiling heat transfer based on deep learning, thereby improving the applicability of numerical simulation for macroscopic boiling phenomena.

[0005] In a first aspect, the present invention provides a numerical simulation method for two-phase flow boiling heat transfer based on deep learning, the method comprising the following steps:

[0006] The collected target input parameters are input into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and a bubble wall heat flux distribution model is constructed based on the key parameters of bubble detachment characteristics. The target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties. The key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding velocity, sliding distance, rising diameter, and nucleation density.

[0007] The heat flux distribution model of the bubble wall is embedded into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework; the embedded two-fluid framework includes the mass conservation equation, momentum conservation equation and energy conservation equation for the liquid phase and the vapor phase.

[0008] The two-phase flow boiling heat transfer was simulated using the embedded two-fluid frame to obtain the temperature and velocity field distributions of the liquid and vapor phases.

[0009] According to the present invention, a numerical simulation method for two-phase flow boiling heat transfer based on deep learning is provided, wherein the target deep neural network model is constructed through the following steps:

[0010] Collect bubble dynamics datasets for multiple working conditions and multiple working media, and determine the input parameters for each sample based on the bubble dynamics datasets;

[0011] The input parameters of each sample are dimensionless to obtain the preprocessed input parameters of each sample; the input parameters of each sample include the fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle and fluid properties corresponding to the sample data;

[0012] The preprocessed sample input parameters are input into the initial deep neural network model to obtain the output parameters corresponding to each sample input parameter.

[0013] The total cost function is determined based on the output parameters corresponding to the input parameters of each sample.

[0014] Based on the total cost function, the weight coefficients of the initial deep neural network model are optimized using the Adam algorithm for adaptive moments estimation until the errors of the training set and the validation set are both lower than a preset threshold, thus obtaining the optimal weight coefficients; the total cost function is the weighted sum of the loss functions of each of the output parameters.

[0015] Based on the optimal weight coefficients, the target deep neural network model is determined.

[0016] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer includes the following steps: collecting bubble dynamics datasets for multiple operating conditions and multiple working fluids, and determining the input parameters for each sample based on the bubble dynamics datasets.

[0017] Determine the data selection criteria; the data selection criteria include the selected working fluid type, coverage pressure range, coverage flow range, tilt angle, and wall superheat; the selected working fluid type includes at least water, liquid nitrogen, and novel cooling working fluids; the coverage pressure range is 0.1 MPa to 20 MPa, the coverage flow range is 0.1 m / s to 10 m / s, and the tilt angle is 0 degrees to 90 degrees;

[0018] Based on the data selection criteria, determine the input data;

[0019] The input data is optimized by feature filtering, and redundant parameters are removed to obtain the input parameters for each sample.

[0020] According to the present invention, a numerical simulation method for two-phase flow boiling heat transfer based on deep learning is provided. The target deep neural network model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is consistent with the dimension of the target input parameters, and the number of nodes in the output layer is consistent with the number of key parameters of the bubble detachment characteristics to be predicted. The step of inputting the collected target input parameters into the target deep neural network model to obtain the key parameters of the bubble detachment characteristics includes:

[0021] The target input parameters are collected and normalized to obtain preprocessed target input parameters.

[0022] The preprocessed target input parameters are input into the input layer to obtain the feature vector matrix;

[0023] The feature vector matrix is ​​input into the hidden layer to obtain the feature vector matrix after weighting;

[0024] The weighted feature vector matrix is ​​input into the output layer to obtain the output vector, and the output vector is mapped to the specific key parameters of the bubble detachment characteristic.

[0025] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer is provided, wherein the bubble wall heat flux distribution model includes heat flux distribution sub-models corresponding to evaporation heat flux, wall quenching heat flux, sliding bubble heat flux, and convection heat flux.

[0026] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer is provided, wherein constructing a bubble wall heat flux distribution model based on the key parameters of the bubble detachment characteristics includes:

[0027] Based on the bubble detachment frequency, the nucleation density, and the rising diameter, a heat flow distribution sub-model corresponding to the evaporation heat flow is constructed;

[0028] Based on the bubble detachment diameter and the wall temperature gradient, a heat flux distribution sub-model corresponding to the wall quenching heat flux is constructed.

[0029] Based on the sweeping area and time of the sliding bubble, calculate the heat flow distribution sub-model corresponding to the heat flow of the sliding bubble;

[0030] The heat flow distribution sub-model corresponding to the convective heat flow is calculated using the remaining heating area.

[0031] According to the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by the present invention, the method further includes embedding the bubble wall heat flux distribution model into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework.

[0032] Determine the target mass exchange and target energy exchange generated by the bubble wall heat flux distribution model;

[0033] The target mass exchange is determined as the mass exchange source term in the Euler-Euler two-fluid model, and the target energy exchange is determined as the energy exchange source term in the Euler-Euler two-fluid model, thus obtaining the embedded two-fluid framework.

[0034] Secondly, the present invention also provides a numerical simulation device for two-phase flow boiling heat transfer based on deep learning, the device comprising the following modules:

[0035] A construction module is used to input the collected target input parameters into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and to construct a bubble wall heat flux distribution model based on the key parameters of bubble detachment characteristics; the target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties; the key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding velocity, sliding distance, rising diameter, and nucleation density;

[0036] The heat flux distribution model of the bubble wall is embedded into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework; the embedded two-fluid framework includes the mass conservation equation, momentum conservation equation and energy conservation equation for the liquid phase and the vapor phase.

[0037] The simulation module is used to simulate the two-phase flow boiling heat transfer numerically through the embedded two-fluid frame, and obtain the temperature field and velocity field distribution of the liquid phase and the vapor phase.

[0038] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the numerical simulation method for two-phase flow boiling heat transfer based on deep learning as described above.

[0039] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the deep learning-based two-phase flow boiling heat transfer numerical simulation method as described above.

[0040] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the deep learning-based two-phase flow boiling heat transfer numerical simulation method as described above.

[0041] The present invention provides a method, apparatus, device, and storage medium for numerical simulation of two-phase flow boiling heat transfer based on deep learning. First, the collected target input parameters are input into a target deep neural network model to obtain key parameters of bubble detachment characteristics. Then, a bubble wall heat flux distribution model is constructed based on these key parameters. The target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties. The key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, slip velocity, slip distance, rising diameter, and nucleation density. Next, the bubble wall heat flux distribution model is embedded into an Euler-Euler two-fluid framework to obtain the embedded two-fluid framework. The embedded two-fluid framework includes mass conservation equations, momentum conservation equations, and energy conservation equations for the liquid and vapor phases. Finally, a numerical simulation of two-phase flow boiling heat transfer is performed using the embedded two-fluid framework to obtain the temperature and velocity field distributions of the liquid and vapor phases.

[0042] In this invention, bubble characteristic parameters (key parameters of bubble detachment characteristics) are obtained based on deep neural network prediction. The two-phase wall boiling phenomenon is modeled based on the bubble characteristic parameters and coupled with the Euler-Euler two-phase flow model to obtain a two-phase boiling heat transfer calculation model with cross-scale and wide applicability. That is, the macroscopic two-fluid Euler-Euler model is used as a framework, and the deep neural network model is used to replace the microscopic bubble detachment model. This model has high applicability for numerical simulation of macroscopic boiling phenomena. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0044] Figure 1 This is one of the flowcharts of the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by the present invention.

[0045] Figure 2 This is the second flowchart of the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by the present invention.

[0046] Figure 3 This is the third flowchart of the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by this invention.

[0047] Figure 4 This is the fourth flowchart of the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by this invention.

[0048] Figure 5 This is the fifth flowchart of the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by the present invention.

[0049] Figure 6 This is a schematic diagram of the structure of the two-phase flow boiling heat transfer numerical simulation device based on deep learning provided by the present invention.

[0050] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0052] To more clearly understand the various embodiments provided by the present invention, the technical content involved in the present invention will first be described as follows:

[0053] Traditional numerical simulation methods for two-phase flow also include the two-fluid (Euler-Euler) method. The two-fluid (Euler-Euler) method is based on the statistical properties of bubbles and often combines the wall heat flux distribution model to describe the boiling phenomenon. It has a wide range of applications. However, its bubble detachment characteristic sub-model relies on empirical relations. Due to the limitations of the empirical relations themselves, the application range of the model is also limited. For example, the sub-model that is applicable to water is no longer applicable to liquid nitrogen. For new working fluids, there is a lack of corresponding experimental relations, which limits the application of the model.

[0054] To address the aforementioned shortcomings in existing technologies, this invention proposes a method that uses a macroscopic Euler-Euler two-fluid model as a framework and employs a deep neural network model to replace the microscopic bubble detachment model in predicting key parameters of bubble detachment characteristics. Furthermore, by constructing a multi-scale coupled framework, a numerical simulation method for two-phase flow boiling heat transfer is developed, which has a wide range of applications, requires fewer computational resources, and has high computational accuracy.

[0055] The following is combined with Figures 1-7 This invention describes a deep learning-based numerical simulation method, apparatus, device, and storage medium for two-phase flow boiling heat transfer.

[0056] Figure 1 This is one of the flowcharts illustrating the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by this invention, as shown below. Figure 1 As shown, the method includes the following:

[0057] Step 101: Input the collected target input parameters into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and construct a bubble wall heat flux distribution model based on the key parameters of bubble detachment characteristics; the target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties; the key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding velocity, sliding distance, rising diameter, and nucleation density;

[0058] First, it should be noted that the execution subject of this invention is an electronic device used to realize numerical simulation of two-phase flow boiling heat transfer based on deep learning, so as to improve the applicability of numerical simulation of phase flow boiling heat transfer.

[0059] The target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties. In fluid mechanics and heat transfer, parameters such as fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties have a significant impact on flow and heat transfer characteristics; that is, the target input parameters are variables affecting bubble escape characteristics. Fluid flow rate is the volume (volume flow rate, cubic meters per second m³ / s) or mass (mass flow rate, kilograms per second kg / s) of fluid passing through the flow channel per unit time. Wall superheat is the wall temperature T. w With fluid saturation temperature T sat The difference This is especially crucial in boiling heat transfer. Fluid pressure is the hydrostatic pressure of the fluid or the operating pressure of the system (Pa or MPa). Contact angle is the angle between the liquid-solid interface and the gas-liquid interface when a droplet contacts a solid surface, characterizing wettability. Wall roughness is the average height (in μm) of the microscopic unevenness of a surface, commonly expressed as Ra or Rz. Inclination angle is the angle of inclination of the flow channel or wall relative to the horizontal direction.

[0060] In this embodiment, the key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding speed, sliding distance, rising diameter, and nucleation density.

[0061] The bubble departure diameter refers to the equivalent diameter of a bubble as it detaches from the wall (usually the diameter assumed by a spherical shape). Factors affecting the bubble departure diameter include wall superheat, surface tension, and contact angle. Bubble departure frequency is the number of bubbles generated by a single nucleation point per unit time. Slip velocity is the upward velocity of a bubble relative to the liquid, determined by the balance of buoyancy and drag. Sliding distance is the horizontal distance a bubble slides along the wall after detachment. The bubble rise diameter is the dynamic diameter of the bubble in free flow after detachment, affected by merging or bursting. Nucleation density is the number of active vaporization nuclei per unit area.

[0062] In traditional numerical simulation methods such as the Euler-Euler method, the bubble escape characteristic sub-model relies on empirical relationships. Due to the inherent limitations of these empirical relationships, the model's application is restricted. For example, a sub-model applicable to water is no longer suitable for liquid nitrogen, and there is a lack of corresponding experimental relationships for novel working fluids, further limiting the model's applicability. Therefore, this invention employs a deep neural network model to predict key parameters of bubble escape characteristics. The collected target input parameters are input into the target deep neural network model to obtain the key parameters of bubble escape characteristics.

[0063] Furthermore, the two-phase wall boiling phenomenon can be modeled based on key parameters of bubble detachment characteristics predicted by deep neural networks, resulting in a bubble wall heat flux distribution model. This wall heat flux distribution model describes wall boiling heat transfer. The bubble wall heat flux distribution model includes several sub-models.

[0064] The enhanced boiling heat transfer during forced convection heat transfer can be attributed to the presence of both slip bubbles and fixed bubbles. This is mainly achieved through two mechanisms: 1) latent heat transfer due to the evaporation of the microliquid layer; and 2) instantaneous heat conduction during the reconstruction process of the disrupted thermal boundary layer in the waiting period (i.e., during the initial formation of the next bubble at the same nucleation point).

[0065] Step 102: Embed the bubble wall heat flow distribution model into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework; the embedded two-fluid framework includes the mass conservation equation, momentum conservation equation and energy conservation equation for the liquid phase and the vapor phase.

[0066] Specifically, after obtaining the bubble wall heat flux distribution model, the bubble wall heat flux distribution model is further embedded into an Euler-Euler two-fluid framework. That is, the bubble wall heat flux distribution model and the Euler-Euler two-fluid model are coupled to obtain the embedded two-fluid framework. The embedded two-fluid framework can be understood as a neural network-based two-phase flow boiling model, used to simulate the numerical heat transfer of two-phase flow boiling, i.e., to simulate flow boiling.

[0067] Among them, the Euler-Euler two-fluid model mainly requires solving the mass conservation equation, momentum conservation equation, and energy conservation equation for the vapor and liquid phases, while the wall boiling model mainly calculates the wall heat flux distribution. The wall heat flux distribution includes sub-models such as quenching heat flux, convective heat flux, evaporative heat flux, and slip heat flux. Key parameters such as bubble detachment diameter, bubble detachment frequency, bubble rising diameter, and bubble nucleation density in these heat flux distribution sub-models are predicted using deep neural networks.

[0068] The momentum conservation equations include the following:

[0069]

[0070] in, Indicates time (seconds). Indicates the q-phase void fraction. The density of phase q (kg / m³) is expressed as q phase density. 3 ), This represents the q-phase velocity (meters per second, m / s). Interphase mass exchange (kg / m³) transferred from phase p to phase q 3 ), This represents the interphase mass exchange (kg / m³) transferred from phase q to phase p. 3 ), ρ represents the p-phase, such as the vapor phase, while q represents the q-phase, such as the liquid phase. Indicates the mass source item (kg / m³) 3 ), This represents the gradient operator.

[0071] The mass conservation equation is as follows:

[0072]

[0073] in, Indicates the q-phase void fraction. Represents the q phase density (kg / m³) 3 ), Indicates pressure (Pa). Represents the gradient operator. Represents the gravitational constant. This represents the q-phase velocity (m / s). This represents the interphase mass exchange (kg / m³) transferred from phase p to phase q. 3 ), This represents the interphase mass exchange (kg / m³) transferred from phase q to phase p. 3 ), The interphase velocity (meters per second, m·s) transferred from phase p to phase q -1 ), The interphase velocity (meters per second, m·s) transferred from phase q to phase p -1 ), Interphase force (Newtons per cubic meter, N·m) -3 ), External volume force (N·m) -3 ), Shear lift (N·m) -3 ); Virtual mass force (N·m) -3 ), Wall lubrication force (N·m) -3 ), Turbulent dissipation force (N·m) -3 ), Represents the gradient operator. Let represent the stress tensor (Pa) of phase q.

[0074] The energy conservation equation is as follows:

[0075]

[0076] Indicates the q-phase void fraction. Represents the q phase density (kg / m³) 3 ), This represents the enthalpy value of phase q (kilojoules per kilogram, kJ / kg). This represents the q-phase velocity (m / s). It represents the thermal conductivity of phase q (watts per meter per Kelvin, W / (m·K)). This represents the q-phase temperature (K). This indicates interphase heat exchange (kilojoules per cubic meter, kJ / m³). 3 ), This represents the interphase mass exchange (kg / m³) transferred from phase p to phase q. 3 ), This represents the saturation enthalpy (kJ / kg) between the p phase phases. This represents the interphase mass exchange (kg / m³) transferred from phase q to phase p. 3 ), This represents the saturation enthalpy (kJ / kg) between phases q. Represents the energy source term (kJ / m³). 3 ).

[0077] Step 103: Simulate the two-phase flow boiling heat transfer numerically using the embedded two-fluid frame to obtain the temperature and velocity field distributions of the liquid and vapor phases.

[0078] Specifically, the numerical simulation of two-phase flow boiling heat transfer is performed through the embedded two-fluid frame, that is, the simulation of flow boiling, to obtain, for example, the temperature field and velocity field distribution of the liquid phase and the vapor phase.

[0079] The method provided in this embodiment first inputs the collected target input parameters into a target deep neural network model to obtain key parameters of bubble detachment characteristics, and then constructs a bubble wall heat flux distribution model based on these key parameters. The target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties. The key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, slip velocity, slip distance, rising diameter, and nucleation density. Then, the bubble wall heat flux distribution model is embedded into an Euler-Euler two-fluid framework to obtain the embedded two-fluid framework, which includes the mass conservation equation, momentum conservation equation, and energy conservation equation for the liquid and vapor phases. Subsequently, the two-phase flow boiling heat transfer is simulated numerically using the embedded two-fluid framework to obtain the temperature and velocity field distributions of the liquid and vapor phases.

[0080] In this invention, bubble characteristic parameters (key parameters of bubble detachment characteristics) are obtained based on deep neural network prediction. The two-phase wall boiling phenomenon is modeled based on the bubble characteristic parameters and coupled with the Euler-Euler two-phase flow model to obtain a two-phase boiling heat transfer calculation model with cross-scale and wide applicability. That is, the macroscopic two-fluid Euler-Euler model is used as a framework, and the deep neural network model is used to replace the microscopic bubble detachment model. This model has high applicability for numerical simulation of macroscopic boiling phenomena.

[0081] According to the present invention, a numerical simulation method for two-phase flow boiling heat transfer based on deep learning is provided. The target deep neural network model is constructed through the following steps:

[0082] Collect bubble dynamics datasets for multiple working conditions and multiple working fluids, and determine the input parameters for each sample based on the bubble dynamics datasets;

[0083] The input parameters of each sample are dimensionless to obtain the preprocessed input parameters of each sample; the input parameters of each sample include the fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle and fluid properties corresponding to the sample data;

[0084] The preprocessed input parameters of each sample are input into the initial deep neural network model to obtain the output parameters corresponding to each sample input parameter.

[0085] The total cost function is determined based on the output parameters corresponding to the input parameters of each sample.

[0086] Based on the total cost function, the Adam algorithm is used to optimize the weight coefficients of the initial deep neural network model through adaptive moments estimation until the errors of the training set and validation set are both lower than the preset threshold, thus obtaining the optimal weight coefficients; the total cost function is the weighted sum of the loss functions of each output parameter.

[0087] The target deep neural network model is determined based on the optimal weight coefficients.

[0088] Specifically, in some embodiments, the steps for constructing the target deep neural network model are as follows:

[0089] First, bubble dynamics datasets for multiple working conditions and multiple working media are collected, and the input parameters of each sample are determined based on the bubble dynamics datasets. Then, the input parameters of each sample are dimensionless to obtain the preprocessed input parameters of each sample.

[0090] The dataset primarily includes sample input parameters such as fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties, as well as key parameters of bubble detachment characteristics such as bubble detachment diameter, bubble detachment frequency, bubble slip distance, and bubble nucleation density. These key parameters serve as label data to guide the model in learning the intrinsic relationship between key parameters and input parameters. Data sources can include experimental data (such as particle image velocimetry and infrared thermometry), simulation data, and publicly available datasets from literature. Determining the sample input parameters based on the dataset constitutes the preprocessing process, including features selection, optimization of sample input parameters, and partitioning into training and validation sets.

[0091] Further, the preprocessed input parameters of each sample are input into the initial deep neural network model to obtain the output parameters corresponding to each sample input parameter. Based on the output parameters corresponding to each sample input parameter, the total cost function is determined. Then, according to the total cost function, the weight coefficients of the initial deep neural network model are optimized using the Adaptive Moment Estimation (Adam) algorithm until the errors of both the training set and the validation set are lower than a preset threshold, thus obtaining the optimal weight coefficients. Notably, the weight coefficients of the initial deep neural network model are not optimized; prediction is performed using the initial deep neural network model, and the total cost function is calculated based on the prediction results (output parameters corresponding to each sample input parameter) and the label data. The total cost function is a weighted sum of the loss functions of each output parameter, expressed as follows:

[0092]

[0093] in, Let w represent the total cost function, and w be the coefficient matrix consisting of the input and output. This is the bias term, usually set to a constant, where i represents the i-th data set and j represents the j-th component. Let the loss function be the loss function for the j-th component of the i-th data set. This represents the predicted value of the j-th component in the i-th data set. This represents the measured value (label data) of the j-th component of the i-th data set, where m represents the total number of m data sets.

[0094] The input and output of a neural network satisfy the following relationship: In the formula, w is the coefficient matrix consisting of input and output, and b is the bias term. Y is the input item, and Y is the output item.

[0095] in, It is expressed as follows:

[0096]

[0097] in, Let represent the loss function for the j-th component of the i-th data set, where i represents the i-th data set and j represents the j-th component. This represents the predicted value of the j-th component in the i-th data set. This represents the measured value (label data) of the j-th component of the i-th data set.

[0098] Then, in order to minimize the total cost function, the Adam algorithm for adaptive moments estimation is used to find the optimal weight coefficients w and b, that is, to obtain the optimal coefficient matrix w composed of input and output and the optimal bias term b.

[0099] Ultimately, the target deep neural network model can be determined based on the optimal weight coefficients. That is, the optimal weight coefficients are used as the updated model parameters to obtain the updated deep neural network model, which is the target deep neural network model.

[0100] The method provided in this embodiment trains a target deep neural network model using a collected bubble dynamics dataset. Then, the target deep neural network model predicts key parameters of bubble detachment characteristics based on target input parameters. Subsequently, the two-phase wall boiling phenomenon is modeled based on the key parameters and coupled with the Euler-Euler two-phase flow model to obtain a cross-scale, widely applicable two-phase boiling heat transfer calculation model, thereby improving the applicability of the model.

[0101] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer is provided, which collects bubble dynamics datasets for multiple operating conditions and multiple working fluids, and determines the input parameters for each sample based on the bubble dynamics datasets, including:

[0102] Determine the data selection criteria; the data selection criteria include the type of working fluid, the coverage pressure range, the coverage flow rate range, the tilt angle, and the wall superheat; the type of working fluid should include at least water, liquid nitrogen, and novel cooling working fluids; the coverage pressure range should include 0.1 MPa to 20 MPa, the flow rate range should include 0.1 m / s to 10 m / s, and the tilt angle should include 0 degrees to 90 degrees;

[0103] Based on the data selection criteria, determine the input data;

[0104] The input data is optimized by feature filtering, and redundant parameters are removed to obtain the input parameters for each sample.

[0105] Specifically, in some embodiments, the specific implementation process of collecting bubble dynamics datasets for multiple operating conditions and multiple working media and determining the input parameters of each sample based on the bubble dynamics dataset includes the following steps:

[0106] First, the data selection criteria are determined, including the type of working fluid, the coverage pressure range, the coverage flow rate range, the tilt angle, and the wall superheat. Further, based on these criteria, the input data is determined.

[0107] To broaden the model's applicability, the selected data should cover as wide a range of operating conditions as possible. This includes data on different working fluids, data across wide pressure and flow ranges, data at different tilt angles, and experimental data under varying wall superheat. For example, the selected working fluids should include at least water, liquid nitrogen, and novel cooling media; the pressure range should cover 0.1 MPa to 20 MPa; the flow range 0.1 m / s to 10 m / s; and the tilt angle 0 degrees to 90 degrees.

[0108] Furthermore, the input data is optimized through feature filtering, and redundant parameters are eliminated to obtain the input parameters for each sample. For example, important feature parameters are identified through feature filtering, thereby optimizing the input data; then, redundant parameters are eliminated to obtain the sample input parameters.

[0109] The sample input parameters can also be divided into training and validation sets. The training set is mainly used to train the weight coefficients of the neural network, ensuring that the error between the network's predicted values ​​and the true values ​​is within an acceptable range. The validation set data is mainly used to verify the generalization properties of the neural network. When both the training set error and the validation set error meet the requirements, the neural network can be considered to have good performance.

[0110] For example, Figure 2 This is the second flowchart of the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by the present invention, as shown below. Figure 2 As shown, the method includes:

[0111] Obtain the raw data pool: including experimental data, numerical simulation data, and literature data;

[0112] Representative data screening includes coverage checks, outlier removal, feature selection, and balancing.

[0113] Dataset validation: physical plausibility, extrapolation testing, and sensitivity analysis;

[0114] Generate a representative dataset: The representative dataset covers all key operating conditions in a balanced manner, and the distribution of bubble parameters conforms to physical laws.

[0115] The method provided in this embodiment selects a dataset that includes a wide range of working conditions and preprocesses the data in the dataset. Then, it uses the preprocessed input data to train an initial deep neural network model, which enables the model to better learn the correlation between data. The trained target deep neural network model has a better prediction effect.

[0116] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer is provided. The target deep neural network model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is consistent with the dimension of the target input parameters, and the number of nodes in the output layer is consistent with the number of key parameters of the bubble detachment characteristics to be predicted. The collected target input parameters are input into the target deep neural network model to obtain the key parameters of the bubble detachment characteristics, including:

[0117] Collect the target input parameters, normalize the target input parameters, and obtain the preprocessed target input parameters;

[0118] The preprocessed target input parameters are input into the input layer to obtain the feature vector matrix;

[0119] Input the feature vector matrix into the hidden layer to obtain the feature vector matrix after weighting;

[0120] The weighted feature vector matrix is ​​input into the output layer to obtain the output vector, and the output vector is mapped to specific key parameters of bubble detachment characteristics.

[0121] Specifically, in some embodiments, the target deep neural network model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is consistent with the dimension of the target input parameters, and the number of nodes in the output layer is consistent with the number of key parameters of the bubble detachment characteristics to be predicted. For example, if the input is a time series of pressure, temperature, fluid viscosity, etc., the number of nodes is equal to the number of features.

[0122] Correspondingly, the specific implementation process of inputting the collected target input parameters into the target deep neural network model in step 101 to obtain the key parameters of bubble detachment characteristics includes the following steps:

[0123] First, the target input parameters are collected and normalized to obtain preprocessed target input parameters. Then, the preprocessed target input parameters are input into the input layer to obtain the feature vector matrix. For example, the collected target input parameters form a feature vector X = [x1, x2, ..., x...]. n ].

[0124] Normalizing the target input parameters helps the network converge quickly. The normalization formula is:

[0125]

[0126] in, This represents the normalized target input parameters. Input parameters for the target. The minimum value of the target input parameter. This indicates the maximum value of the input parameter.

[0127] Then, the feature vector matrix is ​​input into the hidden layer to obtain the feature vector matrix after weighting.

[0128] Furthermore, the weighted feature vector matrix is ​​input into the output layer to obtain the output vector, and the output vector is mapped to specific key parameters of bubble detachment characteristics (e.g., Y1 = bubble detachment diameter, Y2 = bubble detachment frequency, Y3 = bubble slip distance, Y4 = bubble nucleation density).

[0129] For example, Figure 3 This is the third flowchart of the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by this invention, as shown below. Figure 3As shown, the input layer takes into account fluid wall parameters (including fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, wall tilt angle, and fluid properties). Then, the feature vector output from the input layer is input into the hidden layer. The hidden layer contains a multi-head attention mechanism, including H1_4, H1_3, H1_2, H1_1, H2_2, and H2_1. After processing the feature vector, the hidden layer obtains a weighted feature vector matrix. Then, the weighted feature vector matrix is ​​input into the output layer to obtain the final predicted bubble characteristic parameters (including bubble detachment diameter, detachment frequency, slip velocity, slip distance, rising diameter, and nucleation density).

[0130] The method provided in this embodiment models the two-phase wall boiling phenomenon based on the bubble characteristic parameters predicted by the target deep neural network model, and couples it with the Euler-Euler two-phase flow model to obtain a cross-scale, widely applicable two-phase boiling heat transfer calculation model with high applicability.

[0131] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer is provided. The bubble wall heat flux distribution model includes heat flux distribution sub-models corresponding to evaporation heat flux, wall quenching heat flux, sliding bubble heat flux, and convection heat flux.

[0132] Specifically, in some embodiments, the bubble wall heat flow distribution model includes heat flow distribution sub-models corresponding to evaporation heat flow, wall quenching heat flow, sliding bubble heat flow, and convection heat flow.

[0133] These heat flow distribution sub-models are then used as original terms in energy equations, etc., to couple the bubble wall heat flow distribution model and the Euler-Euler two-fluid model, resulting in a two-phase flow boiling heat transfer model based on a deep learning network, which is used to simulate boiling heat transfer.

[0134] The method provided in this embodiment includes a heat flow distribution model for the bubble wall, comprising heat flow distribution sub-models corresponding to evaporation heat flow, wall quenching heat flow, sliding bubble heat flow, and convection heat flow. The modeled heat flow distribution sub-model is coupled with the Euler-Euler two-fluid model to achieve numerical simulation of two-phase flow boiling heat transfer based on deep learning, which has high applicability.

[0135] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer is provided, which constructs a heat flux distribution model at the bubble wall based on key parameters of bubble detachment characteristics, including:

[0136] Based on bubble detachment frequency, nucleation density, and rising diameter, a heat flow distribution sub-model corresponding to evaporation heat flow is constructed.

[0137] Based on the bubble escape diameter and the wall temperature gradient, a heat flow distribution sub-model corresponding to the wall quenching heat flow is constructed.

[0138] Based on the sweeping area and time of the sliding bubble, calculate the heat flow distribution sub-model corresponding to the heat flow of the sliding bubble;

[0139] The heat flow distribution sub-model corresponding to the convective heat flow is calculated using the remaining heating area.

[0140] Specifically, in some embodiments, step 101, which involves constructing a heat flux distribution model of the bubble wall based on key parameters of bubble detachment characteristics, is implemented through the following steps:

[0141] First, based on the bubble detachment frequency, nucleation density, and rising diameter, a heat flow distribution sub-model corresponding to the evaporation heat flow is constructed. For example, the evaporation heat flow is calculated using the following formula:

[0142]

[0143] Where, q e Evaporation heat flux density (kW / m³) 2 ), R f Indicates a reduction factor. Represents the bubble nucleation density (1 / m 2 ), that is, nucleation density, The diameter of the rising bubble (m). Vapor density (kg / m³) 3 ), The latent heat of vaporization is (kJ / kg). The bubble detachment frequency (1 / s) Bubble volume (m) 3 ).

[0144] Furthermore, based on the bubble escape diameter and the wall temperature gradient, a heat flux distribution sub-model corresponding to the wall quenching heat flux is constructed. For example, the wall quenching heat flux is calculated using the following formula:

[0145]

[0146] in, The heat flux density for wall quenching (kW / m²) 2 ), The liquid phase thermal conductivity is expressed in watts per meter degree (W / (m·K)). Liquid phase density (kg / m³) 3 ), The liquid phase heat capacity (J / (kg·K)) The temperature (K) of the heater surface. D represents the temperature (K) of the liquid. d R is the diameter (m) of the bubble detachment. f Indicates a reduction factor. Bubble nucleation density (1 / m 2 K represents the bubble influence factor. Let be the bubble waiting time (s). The bubble detachment frequency is 1 / s.

[0147] Among them, R f The reduction factor describes the ratio of the actual number of bubbles detached per unit heater surface area to the number of bubble activation nuclei per unit area, i.e.:

[0148] R f =1 / (l s / s+1)

[0149] Where Rf represents the reduction factor, l s Let represent the sliding distance (meters, m); s represents the spacing (meters, m) between two activated nuclei. It is assumed that the nucleation sites are distributed in a square grid, and the bubble slides only in the direction of fluid flow. Therefore, the spacing between activated nuclei can be approximated as... , Bubble nucleation density (1 / m 2 Note that for the sliding distance l s When the distance between activated nuclei is less than s, .

[0150] Furthermore, based on the sweeping area and time of the sliding bubble, a heat flow distribution sub-model corresponding to the heat flow of the sliding bubble is calculated.

[0151] For a sliding bubble, the slip heat flux density is:

[0152]

[0153] in, This represents the slip heat flux density (kW / m²). 2 ), The sliding distance (meters, m) and average bubble diameter D are represented. avg =(D d +D l ) / 2, D d It is the diameter of the bubble detachment in meters (D). l R is the diameter of the rising bubble (m). f Indicates a reduction factor. The liquid phase thermal conductivity is (W / (m K)). Liquid phase density (kg / m³) 3 ), The liquid phase heat capacity (J / (kg·K)) The temperature of the heater surface (K). Where K is the temperature of the liquid (K), and K is the bubble influence factor. The bubble detachment frequency (1 / s) Indicates the slip time (seconds). Bubble nucleation density (1 / m 2 ).

[0154] Furthermore, the heat flow distribution sub-model corresponding to the convective heat flow is calculated using the remaining heating area. For example, the convective heat flow is calculated using the following formula:

[0155] In this process, forced convection always exists in the heating surface region unaffected by stationary and slipping bubbles, and the forced convection heat transfer area ratio is:

[0156]

[0157] in, To force convection heat transfer area ratio, Indicates a reduction factor. Bubble nucleation density (1 / m 2 ), Let D be the sliding displacement (m). avg D is the average bubble diameter (m). d It is the diameter (m) of the bubble detachment, t w Let the bubble waiting time be (s). Let K be the bubble detachment frequency (1 / s) and K be the bubble influence factor. Indicates the slip time (seconds).

[0158] Forced convection heat flux (i.e., wall convection heat flux density) q c It can be expressed as:

[0159]

[0160] This indicates the temperature (K) of the heater surface. Indicates the temperature (K) of the liquid. The convective heat transfer coefficient (watts per square kelvin, W / m²) 2 ·K), The area ratio for forced convection heat transfer.

[0161] Therefore, the total wall heat flux (i.e., the total wall heat flux density) can be determined based on the above heat flux components:

[0162]

[0163] in, The total heat flux density of the wall is The wall evaporation heat flux density, For wall quenching heat flux density, The wall slip heat flux density, The wall convective heat flux density is given.

[0164] The method provided in this embodiment uses a deep neural network to predict key parameters such as bubble detachment diameter, bubble detachment frequency, bubble rising diameter, and bubble nucleation density. Then, a wall heat flow distribution model is constructed based on these key parameters. Subsequently, the wall heat flow distribution model is coupled with the Euler-Euler two-fluid model to realize numerical simulation of two-phase flow boiling heat transfer, which has strong applicability.

[0165] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer is provided, which embeds a bubble wall heat flux distribution model into an Euler-Euler two-fluid framework to obtain an embedded two-fluid framework, comprising:

[0166] Determine the target mass exchange and target energy exchange generated by the wall boiling model;

[0167] The target mass exchange is defined as the mass exchange source term in the Euler-Euler two-fluid model, and the target energy exchange is defined as the energy exchange source term in the Euler-Euler two-fluid model, thus obtaining the embedded two-fluid framework.

[0168] Specifically, in some embodiments, step 102 can be implemented by the following steps:

[0169] The steps for coupling the bubble wall heat flux distribution model and the Euler-Euler two-fluid model are as follows:

[0170] First, calculate the mass exchange generated by the wall boiling model. The calculation formula is as follows:

[0171]

[0172] in, This represents the mass exchange that occurs during boiling (kilograms per cubic meter, kg / m³). 3 ), The wall evaporation heat flux density (kW / m²) 2 ), It is the latent heat of vaporization (kilojoules per kilogram, kJ / kg). Liquid phase heat capacity (joules per kilogram per Kelvin, J / (kg·K)) It represents the temperature (Kelvin, K).

[0173] The heat source generated by boiling is:

[0174]

[0175] The energy source term for boiling (kilojoules per cubic meter, kJ / m³) 3 ), Mass exchange that occurs during boiling (kg / m³) 3 ), The latent heat of vaporization is (kJ / kg).

[0176] Furthermore, the mass obtained from the above calculations is exchanged. Energy exchange S H By incorporating the mass exchange and energy source terms from the three major conservation equations in the Euler-Euler two-fluid model, the coupling between the Euler-Euler two-fluid model and the wall boiling model is achieved.

[0177] The method provided in this embodiment embeds the bubble wall heat flow distribution model into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework.

[0178] Figure 4 This is the fourth flowchart of the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by this invention, as shown below. Figure 4 As shown, the method includes:

[0179] First, the deep neural network predicts the key parameters of the bubble detachment characteristics: the deep neural network includes an input layer, a hidden layer, and an output layer.

[0180] Then, based on the key parameters of bubble detachment characteristics, a heat flux distribution model for the bubble wall is constructed. The total heat flux distribution in the bubble wall heat flux distribution model includes evaporation heat flux, quenching heat flux, convection heat flux, and slip heat flux.

[0181] Furthermore, the wall heat flux contribution output by the bubble wall heat flux distribution model is used as the energy exchange source term and mass exchange source term, and coupled into the Euler-Euler two-fluid model. The Euler-Euler two-fluid model includes the mass conservation equation, momentum conservation equation, and energy conservation equation.

[0182] Figure 5 This is the fifth flowchart illustrating the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by this invention. Figure 5 As shown, the method includes:

[0183] Acquire experimental data, numerical simulation data, and literature data, as well as fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties;

[0184] The input parameters in the acquired dataset are processed by dimensionless transformation and feature filtering.

[0185] The processed dataset is input into the initial deep neural network model to train the target deep neural network model. The target deep neural network model is then used to predict the bubble characteristic output parameters. These parameters include bubble detachment diameter, detachment frequency, slip velocity, slip distance, rising diameter, and nucleation density.

[0186] Subsequently, CFD coupling was performed based on the output parameters of bubble characteristics to predict the boiling heat transfer values ​​in real time.

[0187] The method provided in this embodiment updates the two-phase flow field parameters by dynamically coupling the bubble slip velocity and slip distance through an interphase force model.

[0188] The following describes the deep learning-based two-phase flow boiling heat transfer numerical simulation device provided by the present invention. The deep learning-based two-phase flow boiling heat transfer numerical simulation device described below can be referred to in correspondence with the deep learning-based two-phase flow boiling heat transfer numerical simulation method described above.

[0189] Figure 6 This is a schematic diagram of the structure of the deep learning-based two-phase flow boiling heat transfer numerical simulation device provided by the present invention, as shown below. Figure 6 As shown, the deep learning-based two-phase flow boiling heat transfer numerical simulation device 600 includes the following modules:

[0190] The construction module 610 is used to input the collected target input parameters into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and to construct a bubble wall heat flux distribution model based on the key parameters of bubble detachment characteristics; the target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties; the key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding velocity, sliding distance, rising diameter, and nucleation density;

[0191] The heat flux distribution model of the bubble wall is embedded into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework; the embedded two-fluid framework includes the mass conservation equation, momentum conservation equation and energy conservation equation for the liquid phase and the vapor phase.

[0192] The simulation module 620 is used to simulate the two-phase flow boiling heat transfer numerically through the embedded two-fluid frame, and obtain the temperature field and velocity field distribution of the liquid phase and the vapor phase.

[0193] The device provided in this embodiment includes a construction module 610 and a simulation module 620. The construction module 610 is used to input the collected target input parameters into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and to construct a bubble wall heat flux distribution model based on the key parameters of bubble detachment characteristics. The target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties. The key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, slip velocity, slip distance, rising diameter, and nucleation density. Then, the bubble wall heat flux distribution model is embedded into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework, which includes the mass conservation equation, momentum conservation equation, and energy conservation equation for the liquid and vapor phases. Subsequently, the simulation module 620 is used to simulate the two-phase flow boiling heat transfer numerically through the embedded two-fluid framework to obtain the temperature field and velocity field distribution of the liquid and vapor phases.

[0194] In this invention, bubble characteristic parameters (key parameters of bubble detachment characteristics) are obtained based on deep neural network prediction. The two-phase wall boiling phenomenon is modeled based on the bubble characteristic parameters and coupled with the Euler-Euler two-phase flow model to obtain a two-phase boiling heat transfer calculation model with cross-scale and wide applicability. That is, the macroscopic two-fluid Euler-Euler model is used as a framework, and the deep neural network model is used to replace the microscopic bubble detachment model. This model has high applicability for numerical simulation of macroscopic boiling phenomena.

[0195] According to the present invention, a numerical simulation method for two-phase flow boiling heat transfer based on deep learning is provided, wherein the target deep neural network model is constructed through the following steps:

[0196] Collect bubble dynamics datasets for multiple working conditions and multiple working media, and determine the input parameters for each sample based on the bubble dynamics datasets;

[0197] The input parameters of each sample are dimensionless to obtain the preprocessed input parameters of each sample; the input parameters of each sample include the fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle and fluid properties corresponding to the sample data;

[0198] The preprocessed sample input parameters are input into the initial deep neural network model to obtain the output parameters corresponding to each sample input parameter.

[0199] The total cost function is determined based on the output parameters corresponding to the input parameters of each sample.

[0200] Based on the total cost function, the weight coefficients of the initial deep neural network model are optimized using the Adam algorithm for adaptive moments estimation until the errors of the training set and the validation set are both lower than a preset threshold, thus obtaining the optimal weight coefficients; the total cost function is the weighted sum of the loss functions of each of the output parameters.

[0201] Based on the optimal weight coefficients, the target deep neural network model is determined.

[0202] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer includes the following steps: collecting bubble dynamics datasets for multiple operating conditions and multiple working fluids, and determining the input parameters for each sample based on the bubble dynamics datasets.

[0203] Determine the data selection criteria; the data selection criteria include the selected working fluid type, coverage pressure range, coverage flow range, tilt angle, and wall superheat; the selected working fluid type includes at least water, liquid nitrogen, and novel cooling working fluids; the coverage pressure range is 0.1 MPa to 20 MPa, the coverage flow range is 0.1 m / s to 10 m / s, and the tilt angle is 0 degrees to 90 degrees;

[0204] Based on the data selection criteria, the input data is determined;

[0205] The input data is optimized by feature filtering, and redundant parameters are removed to obtain the input parameters for each sample.

[0206] According to the present invention, a numerical simulation method for two-phase flow boiling heat transfer based on deep learning is provided. The target deep neural network model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is consistent with the dimension of the target input parameters, and the number of nodes in the output layer is consistent with the number of key parameters of the bubble detachment characteristics to be predicted.

[0207] The construction module 610 is specifically used for:

[0208] The target input parameters are collected and normalized to obtain preprocessed target input parameters.

[0209] The preprocessed target input parameters are input into the input layer to obtain the feature vector matrix;

[0210] The feature vector matrix is ​​input into the hidden layer to obtain the feature vector matrix after weighting;

[0211] The weighted feature vector matrix is ​​input into the output layer to obtain the output vector, and the output vector is mapped to the specific key parameters of the bubble detachment characteristic.

[0212] According to the present invention, a deep learning-based numerical simulation method for two-phase flow boiling heat transfer is provided, wherein the bubble wall heat flux distribution model includes heat flux distribution sub-models corresponding to evaporation heat flux, wall quenching heat flux, sliding bubble heat flux, and convection heat flux.

[0213] According to the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by the present invention, the construction module 610 is further used for:

[0214] Based on the bubble detachment frequency, the nucleation density, and the rising diameter, a heat flow distribution sub-model corresponding to the evaporation heat flow is constructed;

[0215] Based on the bubble detachment diameter and the wall temperature gradient, a heat flux distribution sub-model corresponding to the wall quenching heat flux is constructed.

[0216] Based on the sweeping area and time of the sliding bubble, calculate the heat flow distribution sub-model corresponding to the heat flow of the sliding bubble;

[0217] The heat flow distribution sub-model corresponding to the convective heat flow is calculated using the remaining heating area.

[0218] According to the numerical simulation method for two-phase flow boiling heat transfer based on deep learning provided by the present invention, the construction module 610 is specifically used for:

[0219] Determine the target mass exchange and target energy exchange generated by the bubble wall heat flux distribution model;

[0220] The target mass exchange is determined as the mass exchange source term in the Euler-Euler two-fluid model, and the target energy exchange is determined as the energy exchange source term in the Euler-Euler two-fluid model, thus obtaining the embedded two-fluid framework.

[0221] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call logic instructions in the memory 730 to execute a deep learning-based numerical simulation method for two-phase flow boiling heat transfer, which includes:

[0222] The collected target input parameters are input into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and a bubble wall heat flux distribution model is constructed based on the key parameters of bubble detachment characteristics. The target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties. The key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding velocity, sliding distance, rising diameter, and nucleation density.

[0223] The heat flux distribution model of the bubble wall is embedded into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework; the embedded two-fluid framework includes the mass conservation equation, momentum conservation equation and energy conservation equation for the liquid phase and the vapor phase.

[0224] The two-phase flow boiling heat transfer was simulated using the embedded two-fluid frame to obtain the temperature and velocity field distributions of the liquid and vapor phases.

[0225] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0226] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the deep learning-based two-phase flow boiling heat transfer numerical simulation method provided by the above methods, the method comprising:

[0227] The collected target input parameters are input into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and a bubble wall heat flux distribution model is constructed based on the key parameters of bubble detachment characteristics. The target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties. The key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding velocity, sliding distance, rising diameter, and nucleation density.

[0228] The heat flux distribution model of the bubble wall is embedded into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework; the embedded two-fluid framework includes the mass conservation equation, momentum conservation equation and energy conservation equation for the liquid phase and the vapor phase.

[0229] The two-phase flow boiling heat transfer was simulated using the embedded two-fluid frame to obtain the temperature and velocity field distributions of the liquid and vapor phases.

[0230] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the deep learning-based numerical simulation method for two-phase flow boiling heat transfer provided by the methods described above, the method comprising:

[0231] The collected target input parameters are input into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and a bubble wall heat flux distribution model is constructed based on the key parameters of bubble detachment characteristics. The target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties. The key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding velocity, sliding distance, rising diameter, and nucleation density.

[0232] The heat flux distribution model of the bubble wall is embedded into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework; the embedded two-fluid framework includes the mass conservation equation, momentum conservation equation and energy conservation equation for the liquid phase and the vapor phase.

[0233] The two-phase flow boiling heat transfer was simulated using the embedded two-fluid frame to obtain the temperature and velocity field distributions of the liquid and vapor phases.

[0234] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0235] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0236] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A numerical simulation method for two-phase flow boiling heat transfer based on deep learning, characterized in that, include: The collected target input parameters are input into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and a bubble wall heat flux distribution model is constructed based on the key parameters of bubble detachment characteristics. The target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties. The key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding velocity, sliding distance, rising diameter, and nucleation density. The bubble wall heat flux distribution model is embedded into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework; the embedded two-fluid framework includes the mass conservation equation, momentum conservation equation, and energy conservation equation for the liquid and vapor phases; the process of embedding the bubble wall heat flux distribution model into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework includes: determining the target mass exchange and target energy exchange generated by the bubble wall heat flux distribution model; The target mass exchange is determined as the mass exchange source term in the Euler-Euler two-fluid model, and the target energy exchange is determined as the energy exchange source term in the Euler-Euler two-fluid model, thus obtaining the embedded two-fluid framework; The two-phase flow boiling heat transfer was simulated using the embedded two-fluid frame to obtain the temperature and velocity field distributions of the liquid and vapor phases.

2. The numerical simulation method for two-phase flow boiling heat transfer based on deep learning according to claim 1, characterized in that, The target deep neural network model is constructed through the following steps: Collect bubble dynamics datasets for multiple working conditions and multiple working media, and determine the input parameters for each sample based on the bubble dynamics datasets; The input parameters of each sample are dimensionless to obtain the preprocessed input parameters of each sample; the input parameters of each sample include the fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle and fluid properties corresponding to the sample data; The preprocessed sample input parameters are input into the initial deep neural network model to obtain the output parameters corresponding to each sample input parameter. The total cost function is determined based on the output parameters corresponding to the input parameters of each sample. Based on the total cost function, the weight coefficients of the initial deep neural network model are optimized using the Adam algorithm for adaptive moments estimation until the errors of the training set and the validation set are both lower than a preset threshold, thus obtaining the optimal weight coefficients; the total cost function is the weighted sum of the loss functions of each of the output parameters. Based on the optimal weight coefficients, the target deep neural network model is determined.

3. The numerical simulation method for two-phase flow boiling heat transfer based on deep learning according to claim 2, characterized in that, The process involves collecting bubble dynamics datasets for multiple operating conditions and multiple working media, and determining the input parameters for each sample based on the bubble dynamics datasets, including: Determine the data selection criteria; the data selection criteria include the selected working fluid type, coverage pressure range, coverage flow range, tilt angle, and wall superheat; the selected working fluid type includes at least water and liquid nitrogen; the coverage pressure range is 0.1 MPa to 20 MPa, the coverage flow range is 0.1 m / s to 10 m / s, and the tilt angle is 0 degrees to 90 degrees; Based on the data selection criteria, the input data is determined; The input data is optimized by feature filtering, and redundant parameters are removed to obtain the input parameters for each sample.

4. The numerical simulation method for two-phase flow boiling heat transfer based on deep learning according to claim 1, characterized in that, The target deep neural network model includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is consistent with the dimension of the target input parameters, and the number of nodes in the output layer is consistent with the number of key parameters of the bubble detachment characteristics to be predicted. The process of inputting the collected target input parameters into the target deep neural network model to obtain key parameters of bubble detachment characteristics includes: The target input parameters are collected and normalized to obtain preprocessed target input parameters. The preprocessed target input parameters are input into the input layer to obtain the feature vector matrix; The feature vector matrix is ​​input into the hidden layer to obtain the feature vector matrix after weighting; The weighted feature vector matrix is ​​input into the output layer to obtain the output vector, and the output vector is mapped to the specific key parameters of the bubble detachment characteristic.

5. The numerical simulation method for two-phase flow boiling heat transfer based on deep learning according to claim 1, characterized in that, The bubble wall heat flow distribution model includes heat flow distribution sub-models corresponding to evaporative heat flow, wall quenching heat flow, sliding bubble heat flow, and convective heat flow.

6. The numerical simulation method for two-phase flow boiling heat transfer based on deep learning according to claim 5, characterized in that, The step of constructing a heat flux distribution model on the bubble wall based on the key parameters of the bubble detachment characteristics includes: Based on the bubble detachment frequency, the nucleation density, and the rising diameter, a heat flow distribution sub-model corresponding to the evaporation heat flow is constructed; Based on the bubble detachment diameter and the wall temperature gradient, a heat flux distribution sub-model corresponding to the wall quenching heat flux is constructed. Based on the sweeping area and time of the sliding bubble, calculate the heat flow distribution sub-model corresponding to the heat flow of the sliding bubble; The heat flow distribution sub-model corresponding to the convective heat flow is calculated using the remaining heating area.

7. A numerical simulation device for two-phase flow boiling heat transfer based on deep learning, characterized in that, include: A construction module is used to input the collected target input parameters into the target deep neural network model to obtain key parameters of bubble detachment characteristics, and to construct a bubble wall heat flux distribution model based on the key parameters of bubble detachment characteristics; the target input parameters include fluid flow rate, wall superheat, fluid pressure, contact angle, wall roughness, tilt angle, and fluid properties; the key parameters of bubble detachment characteristics include bubble detachment diameter, detachment frequency, sliding velocity, sliding distance, rising diameter, and nucleation density; The bubble wall heat flux distribution model is embedded into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework. The embedded two-fluid framework includes the mass conservation equation, momentum conservation equation, and energy conservation equation for the liquid and vapor phases. The process of embedding the bubble wall heat flux distribution model into the Euler-Euler two-fluid framework to obtain the embedded two-fluid framework includes: determining the target mass exchange and target energy exchange generated by the bubble wall heat flux distribution model; determining the target mass exchange as the mass exchange source term in the Euler-Euler two-fluid model, and determining the target energy exchange as the energy exchange source term in the Euler-Euler two-fluid model, thus obtaining the embedded two-fluid framework. The simulation module is used to simulate the two-phase flow boiling heat transfer numerically through the embedded two-fluid frame, and obtain the temperature field and velocity field distribution of the liquid phase and the vapor phase.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the deep learning-based two-phase flow boiling heat transfer numerical simulation method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the deep learning-based two-phase flow boiling heat transfer numerical simulation method as described in any one of claims 1 to 6.

Images