Fluid and particle coupling loss calculation method and device, equipment and storage medium
By constructing a loss calculation method for fluid and particle coupling in a physical information neural network, the dynamic boundary problem in fluid-solid coupling is solved, the prediction accuracy and stability are improved, and it is suitable for loss calculation of fluid and particle coupling.
Patent Information
- Application Number
- CN202510742427.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-05
- Publication Date
- 2025-09-02
AI Technical Summary
The prior art is difficult to effectively capture the dynamic coupling effect of the fluid-solid interface under dynamic boundary conditions during the fluid-solid coupling process, resulting in the model's prediction accuracy drop or failure in complex fluid-solid coupling problems.
Physical information neural networks (PINNs) are used to combine flow networks and particle networks, and by constructing fluid momentum equations, particle dynamic equations and boundary equations, composite functions are generated and loss functions are constructed, and the network is adjusted to improve prediction accuracy and stability.
The prediction accuracy of the physical quantities related to fluid and particles is improved, and the stability of the model is ensured in complex fluid-solid coupling scenarios.
Smart Images
Figure CN120579480A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for calculating loss of fluid-particle coupling. Background Art
[0002] Fluid-structure interaction (FSI) has crucial applications in numerous industrial fields. For example, in aerospace, aircraft aeroelastic analysis, which involves the interaction between airflow and the aircraft structure, is crucial for ensuring flight safety. In mechanical engineering, flow-induced vibrations in engine cooling systems directly impact engine performance and lifespan. In biomedicine, vascular dynamics simulation is crucial for studying the mechanisms of cardiovascular disease. All of these scenarios require accurate simulation and analysis of FSI phenomena. Traditional FSI solvers face numerous challenges in practical application. Firstly, users require in-depth knowledge of physical models to accurately establish the coupling relationship between fluids and solids, and sophisticated meshing techniques are required to ensure computational accuracy and stability. Secondly, meshing often leads to thorny issues such as mesh distortion and negative volume. These issues not only lead to inaccurate calculation results but also hinder convergence in the iterative process, significantly reducing computational efficiency. These issues severely limit the application and effectiveness of traditional solvers for complex FSI problems.
[0003] With the rapid development of artificial intelligence (AI) in recent years, physics-informed neural networks (PINNs), an emerging deep learning technology, have gradually emerged in the fields of fluid and structural simulation. PINNs incorporate a differentiated learning mechanism, enabling semi-supervised or even unsupervised training. This not only greatly reduces the workload of training dataset preparation but also significantly improves the model's transferability, enabling rapid adaptation and application in diverse physical scenarios.
[0004] However, most current research remains limited to the study of fluid dynamics within a fixed flow domain. For dynamic boundary conditions such as large deformations that occur during fluid-structure interaction or significant motion of solid structures, these methods struggle to accurately capture the dynamic coupling effects at the fluid-structure interface. This results in a significant decrease in model prediction accuracy and even potential model failure when faced with complex fluid-structure interaction problems. Therefore, effectively utilizing PINNs to address dynamic boundary issues in fluid-structure interaction and improve the model's prediction accuracy and stability in complex fluid-structure interaction scenarios remains a pressing technical challenge. Summary of the Invention
[0005] In order to solve or partially solve the problems existing in the relevant technologies, the present application provides a method, device, equipment and storage medium for calculating the loss of fluid-particle coupling, which can fully consider the driving or hindering effect of the fluid on the particles, as well as the flow field changes and reaction to the flow field caused by particle movement in the fluid-solid coupling calculation process, thereby improving the accuracy of the prediction of fluid and particle-related physical quantities and ensuring the stability of the model.
[0006] In a first aspect, the present application provides a method for calculating the loss of fluid-particle coupling, which is applied to a physical information neural network. The physical information neural network includes a flow network and a particle network. The flow network is used to fit the flow physical field, and the particle network is used to fit the motion field of solid particles. The method includes: Respectively obtaining flow field physical quantities output by the flow network and particle physical quantities output by the particle network; the flow field physical quantities include at least flow field velocity, flow field pressure, and flow field stress tensor, and the particle physical quantities include at least particle velocity; constructing a fluid momentum equation and a constitutive equation of the fluid, a particle dynamic equation and a boundary equation of the particles according to the flow field physical quantities and the particle physical quantities; generating a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation, and the particle dynamic equation, and generating a second composite function based on the particle velocity, the first composite function, and the boundary equation; A first loss function of the flow network is constructed according to the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; a second loss function of the particle network is constructed according to the particle dynamic equation; the first loss function is used to adjust the flow network, and the second loss function is used to adjust the particle network.
[0007] In a preferred embodiment, generating a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation, and the particle dynamic equation comprises: embedding the fluid momentum equation and the constitutive equation into the particle dynamic equation; Substituting the flow field velocity into the embedded particle dynamic equation obtains the first composite function.
[0008] In a preferred embodiment, generating a second composite function based on the particle velocity, the first composite function and the boundary equation comprises: embedding the first composite function into the boundary equation; Substituting the particle velocity into the embedded boundary equation yields a second composite function.
[0009] In a preferred embodiment, the first loss function of the flow network constructed according to the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function includes: Setting a first initial condition equation of the fluid physical field; Setting respective first coefficients for the first initial condition equation, the fluid momentum equation, the constitutive equation, the boundary equation, and the second composite function; The first loss function is constructed based on the first coefficient, the first initial condition equation, the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function.
[0010] In a preferred embodiment, constructing the second loss function of the particle network according to the particle dynamic equation includes: Setting a second initial condition equation for the solid particle motion field; Setting respective second coefficients for the second initial condition equation and the particle dynamic equation; The second loss function is constructed based on the second coefficient, the second initial condition equation and the particle dynamic equation.
[0011] In a preferred embodiment, the fluid momentum equation is constructed using the following method: calculating partial derivatives of the flow field physical quantities with respect to input data of the flow network; constructing the fluid momentum equation based on the flow field velocity and the partial derivative; The constitutive equation is constructed using the following method: constructing the constitutive equation based on the partial derivative, the flow field pressure, and the flow field stress tensor; The particle dynamic equation of the particles is constructed as follows: calculating derivatives of the particle velocity with respect to input data of the particle network; The particle dynamic equation is constructed based on the particle velocity and the derivative.
[0012] In a preferred embodiment, the flow network and the granular network are two independent multi-layer perceptron networks.
[0013] In a second aspect, the present application provides a device for calculating loss of fluid-particle coupling, which is applied to a physical information neural network. The physical information neural network includes a flow network and a particle network. The flow network is used to fit the flow physical field, and the particle network is used to fit the motion field of solid particles. The device includes: an acquisition module, configured to respectively acquire flow field physical quantities output by the flow network and particle physical quantities output by the particle network; the flow field physical quantities at least include flow field velocity, flow field pressure, and flow field stress tensor, and the particle physical quantities at least include particle velocity; A construction module, configured to construct a fluid momentum equation and a constitutive equation of the fluid, a particle dynamic equation and a boundary equation of the particles according to the flow field physical quantities and the particle physical quantities; A coupling module is configured to generate a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation, and the particle dynamic equation; and generate a second composite function based on the particle velocity, the first composite function, and the boundary equation; A loss function module is used to construct a first loss function of the flow network based on the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; and to construct a second loss function of the particle network based on the particle dynamic equation; the first loss function is used to adjust the flow network, and the second loss function is used to adjust the particle network.
[0014] A third aspect of the present application provides an electronic device, including: processor; and The memory stores executable codes thereon, and when the executable codes are executed by the processor, the processor is caused to execute the method described above.
[0015] A fourth aspect of the present application provides a computer-readable storage medium having executable code stored thereon. When the executable code is executed by a processor of an electronic device, the processor is caused to execute the method described above.
[0016] The technical solution provided by this application may have the following beneficial effects: An embodiment of the present application provides a method for calculating the loss of fluid and particle coupling, which is applied to a physical information neural network, wherein the physical information neural network includes a flow network and a particle network, wherein the flow network is used to fit the flow physical field, and the particle network is used to fit the motion field of solid particles; the method includes: respectively obtaining the flow field physical quantities output by the flow network and the particle physical quantities output by the particle network; the flow field physical quantities include at least the flow field velocity, the flow field pressure and the flow field stress tensor, and the particle physical quantities include at least the particle velocity; constructing the fluid momentum equation and the constitutive equation of the fluid, the particle dynamic equation and the boundary equation of the particles based on the flow field physical quantities and the particle physical quantities; generating a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation and the particle dynamic equation, and generating a second composite function based on the particle velocity, the first composite function and the boundary equation; constructing a first loss function of the flow network based on the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; constructing a second loss function of the particle network based on the particle dynamic equation; the first loss function is used to adjust the flow network, and the second loss function is used to adjust the particle network. Through the above method, the driving or hindering effect of the fluid on the particles, as well as the changes in the flow field caused by the movement of particles and the reaction to the flow field are fully considered, which improves the accuracy of the prediction of fluid and particle-related physical quantities and ensures the stability of the model.
[0017] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] The above and other objects, features and advantages of the present application will become more apparent by describing in more detail exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.
[0019] Figure 1 1 is a flow chart of a method for calculating the loss of fluid-particle coupling according to an embodiment of the present application; Figure 2 1 is another flow chart of a method for calculating the loss of fluid-particle coupling shown in an embodiment of the present application; Figure 3 Schematic diagram of the flow field velocity field prediction results shown in the embodiment of the present application; Figure 4 Schematic diagram of particle motion trajectory prediction results shown in an embodiment of the present application; Figure 5 is a schematic diagram of the physical information neural network structure shown in an embodiment of the present application; Figure 6Schematic diagram of the working condition shown in the embodiment of the present application; Figure 7 Schematic diagram of the structure of the loss calculation device for fluid-particle coupling shown in an embodiment of the present application; Figure 8 It is a structural diagram of an electronic device shown in an embodiment of the present application. DETAILED DESCRIPTION
[0020] The following describes embodiments of the present application in more detail with reference to the accompanying drawings. Although the accompanying drawings illustrate embodiments of the present application, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.
[0021] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.
[0022] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0023] Currently, most research is still limited to the field of fluid dynamics within a fixed flow domain. For dynamic boundary conditions such as large deformations that occur during fluid-structure interaction or significant motion of solid structures, relevant methods struggle to accurately capture the dynamic coupling effects at the fluid-structure interface. This results in a significant decrease in model prediction accuracy and even potential model failure when faced with complex fluid-structure interaction problems. Therefore, effectively utilizing PINNs to address dynamic boundary issues in fluid-structure interaction and improve the model's prediction accuracy and stability in complex fluid-structure interaction scenarios remains a pressing technical challenge.
[0024] To address the above-mentioned issues, an embodiment of the present application provides a method for calculating the loss of fluid-particle coupling, which can fully consider the driving or hindering effect of the fluid on the particles, as well as the changes in the flow domain and the reaction to the flow field caused by the movement of particles, thereby improving the accuracy of the prediction of physical quantities related to fluids and particles and ensuring the stability of the model.
[0025] The technical solutions of the embodiments of the present application are described in detail below with reference to the accompanying drawings.
[0026] Figure 1 It is a flow chart of a method for calculating the loss of fluid-particle coupling shown in an embodiment of the present application.
[0027] See also Figure 1 The method is applied to a physical information neural network, which includes a flow network and a particle network. The flow network is used to fit the flow physical field, and the particle network is used to fit the solid particle motion field. The method includes: Step 110, respectively obtain the flow field physical quantities output by the flow network and the particle physical quantities output by the particle network; the flow field physical quantities include at least flow field velocity, flow field pressure and flow field stress tensor, and the particle physical quantities include at least particle velocity.
[0028] The physical information neural network incorporates the physical laws described by differential equations into its loss function to guide the learning process to obtain solutions that are more consistent with the basic physical laws. The physical information neural network of the embodiment of the present application includes a flow network (flow_net) and a particle network (particle_net). In one example, the flow network includes a 10-layer neural network: the input layer has 3 neurons; the output layer has 5 neurons; the hidden layer has 8 layers, and each layer has 10 neurons. The particle network also includes a 10-layer neural network: the input layer has 1 neuron; the output layer has 2 neurons; the hidden layer also has 8 layers, and each layer has 10 neurons. Both the flow network and the particle network use the SiLU (Sigmoid-Weighted Linear Unit) function as the activation function.
[0029] Mobile network channel width , inlet velocity As a reference physical quantity, the sampling point input data set is dimensionless and obtained Then, the sampling points of the input layer are forward propagated through the multi-layer neural network to obtain the dimensionless output physical quantity , used to fit the flow physics The particle network uses the same reference physical quantity and , non-dimensionalize the input data of the sampling points and obtain The input layer is forward propagated through the particle network to obtain the dimensionless particle velocity , used to fit the particle motion field ,in is the input data of the flow network, is the input data of the particle network, is the flow field physical quantity output by the flow network, is the particle physical quantity output by the particle network, where represents the dimensionless spatial coordinates of the sampling points, represents the dimensionless time of the sampling point, denote the dimensionless flow field velocity, represents the dimensionless flow field pressure, They represent the dimensionless flow field stress tensors in different directions, represent the dimensionless particle velocity, respectively.
[0030] Step 120 : constructing the fluid momentum equation and constitutive equation of the fluid, the particle dynamic equation and boundary equation of the particles according to the flow field physical quantities and the particle physical quantities.
[0031] Under incompressible steady fluid conditions, the reference fluid density can be given and Reynolds number The fluid momentum equation is constructed based on the flow field physical quantities. The fluid constitutive equation is constructed based on the flow field physical quantities and the Reynolds number. The particle dynamic equation is constructed based on the particle physical quantities and the Reynolds number. The boundary equation is constructed based on the flow field physical quantities and the particle physical quantities.
[0032] Step 130: Generate a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation, and the particle dynamic equation; and generate a second composite function based on the particle velocity, the first composite function, and the boundary equation.
[0033] During fluid-solid coupling, the coupling effect of the flow network on the particle network occurs in the particle dynamic equation. The input data of the fluid sampling points near the particles are obtained and input into the flow network. The output data is calculated through the fluid momentum equation and the constitutive equation. The output data includes the flow field velocity. The flow field velocity is substituted into the particle dynamic equation to generate the first composite function.
[0034] The coupling effect of the particle network on the flow network occurs in the boundary equation. The input data of the particle sampling points on the boundary are obtained and input into the particle network. The particle velocity is calculated by the first composite function and substituted into the boundary equation to generate the second composite function.
[0035] Step 140, constructing a first loss function of the flow network based on the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; constructing a second loss function of the particle network based on the particle dynamic equation; the first loss function is used to adjust the flow network, and the second loss function is used to adjust the particle network.
[0036] In the loss calculation, a first loss function for the flow network is constructed based on the fluid momentum equation, constitutive equation, boundary equation, and a second composite function. A second loss function for the particle network is constructed based on the particle dynamic equation. Backpropagation is performed on the flow network and the particle network, respectively, based on the first and second loss functions. The weights and biases for the next forward propagation are then determined. The flow network and the particle network are then adjusted based on their respective weights and biases until the values of the first and second loss functions fall below a threshold, or the number of iterations reaches a maximum.
[0037] An embodiment of the present application provides a method for calculating the loss of fluid and particle coupling, which is applied to a physical information neural network. The physical information neural network includes a flow network and a particle network. The flow network is used to fit the flow physical field, and the particle network is used to fit the motion field of solid particles. The method includes: respectively obtaining the flow field physical quantities output by the flow network and the particle physical quantities output by the particle network; the flow field physical quantities include at least the flow field velocity, flow field pressure and flow field stress tensor, and the particle physical quantities include at least the particle velocity; constructing the fluid momentum equation and constitutive equation of the fluid, the particle dynamic equation and the boundary equation of the particles based on the flow field physical quantities and the particle physical quantities; generating a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation and the particle dynamic equation, and generating a second composite function based on the particle velocity, the first composite function and the boundary equation; constructing a first loss function of the flow network based on the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; constructing a second loss function of the particle network based on the particle dynamic equation; the first loss function is used to adjust the flow network, and the second loss function is used to adjust the particle network. Through the above method, the driving or hindering effect of the fluid on the particles, as well as the changes in the flow field caused by the movement of particles and the reaction to the flow field are fully considered, which improves the accuracy of the prediction of fluid and particle-related physical quantities and ensures the stability of the model.
[0038] Figure 2 This is another flow chart of the method for calculating the loss of fluid-particle coupling shown in an embodiment of the present application.
[0039] See also Figure 2 The method is applied to a physical information neural network, which includes a flow network and a particle network. The flow network is used to fit the flow physical field, and the particle network is used to fit the solid particle motion field. The method includes: Step 210, respectively obtain the flow field physical quantities output by the flow network and the particle physical quantities output by the particle network; the flow field physical quantities include at least flow field velocity, flow field pressure and flow field stress tensor, and the particle physical quantities include at least particle velocity.
[0040] The physical information neural network incorporates the physical laws described by differential equations into its loss function to guide the learning process to obtain solutions that are more consistent with the basic physical laws. The physical information neural network of the embodiment of the present application includes a flow network (flow_net) and a particle network (particle_net). In one example, the flow network includes a 10-layer neural network: the input layer has 3 neurons; the output layer has 5 neurons; the hidden layer has 8 layers, and each layer has 10 neurons. The particle network also includes a 10-layer neural network: the input layer has 1 neuron; the output layer has 2 neurons; the hidden layer also has 8 layers, and each layer has 10 neurons. Both the flow network and the particle network use the SiLU (Sigmoid-Weighted Linear Unit) function as the activation function.
[0041] Mobile network channel width , inlet velocity As a reference physical quantity, the sampling point input data set is dimensionless and obtained Then, the sampling points of the input layer are forward propagated through the multi-layer neural network to obtain the dimensionless output physical quantity , used to fit the flow physics The particle network uses the same reference physical quantity and , non-dimensionalize the input data of the sampling points and obtain The input layer is forward propagated through the particle network to obtain the dimensionless particle velocity , used to fit the particle motion field ,in is the input data of the flow network, is the input data of the particle network, is the flow field physical quantity output by the flow network, is the particle physical quantity output by the particle network, where represents the dimensionless spatial coordinates of the sampling points, represents the dimensionless time of the sampling point, denote the dimensionless flow field velocity, represents the dimensionless flow field pressure, They represent the dimensionless flow field stress tensors in different directions, represent the dimensionless particle velocity, respectively.
[0042] In an optional embodiment of the present application, the flow network and the granular network are two independent multi-layer perceptron networks.
[0043] The flow network and the granular network are two independent multilayer perceptron (MLP) networks. An MLP is a feedforward artificial neural network composed of multiple neurons (neuronal nodes) arranged in a hierarchical structure, including an input layer, hidden layers, and an output layer. Neurons between layers are connected by weights, and information is propagated from the input layer to the output layer. The flow network and the granular network each have their own Adam (Adaptive Moment Estimation) optimizer, and their iteration process, loss calculation process, and learning rate are independently configured.
[0044] Step 220 : constructing the fluid momentum equation and constitutive equation of the fluid, the particle dynamic equation and boundary equation of the particles according to the flow field physical quantities and the particle physical quantities.
[0045] Under incompressible steady fluid conditions, the reference fluid density can be given and Reynolds number The fluid momentum equation is constructed based on the flow field physical quantities. The fluid constitutive equation is constructed based on the flow field physical quantities and the Reynolds number. The particle dynamic equation is constructed based on the particle physical quantities and the Reynolds number. The boundary equation is constructed based on the flow field physical quantities and the particle physical quantities.
[0046] In an optional embodiment of the present application, the fluid momentum equation is constructed using the following method: Calculate partial derivatives of flow field quantities with respect to input data of the flow network; The fluid momentum equation is constructed based on the flow field velocity and partial derivatives.
[0047] Mobile network channel width , inlet velocity As a reference physical quantity, the sampling point input data set is dimensionless and obtained Then, the sampling points of the input layer are forward propagated through the multi-layer neural network to obtain the dimensionless output physical quantity , using the self-differentiation mechanism of MLP, the partial derivatives of the flow field physical quantities with respect to the input data (time and space) are obtained respectively Under incompressible steady flow conditions, the reference fluid density and Reynolds number , the following formulas (1) and (2) give the dimensionless fluid momentum equations: (1) (2) The constitutive equation is constructed in the following way: The constitutive equation is constructed based on partial derivatives, flow field pressure and flow field stress tensors.
[0048] The constitutive equation is constructed based on the partial derivatives, flow field pressure and flow field stress tensors. The dimensionless constitutive equation can be constructed by referring to the following formulas (3)-(6):
[0049] The particle dynamic equation of the particle is constructed as follows: Calculate the derivative of particle velocity with respect to input data of the particle network; The particle dynamic equations are constructed based on particle velocity and derivatives.
[0050] Particle networks use the same reference physical quantity and , non-dimensionalize the input data of the sampling points and obtain The input layer is forward propagated through the particle network to obtain the dimensionless particle velocity , the trajectory of the particle can be calculated using the trapezoidal method , which can be calculated by formula (7) and (8): (7) (8) Then, according to the position of the particle at each time point, the coordinates of the surface and boundary layer sampling points are calculated respectively, which can be calculated by formulas (9) and (10): (9) (10) Where R represents the boundary layer radius. When , it indicates the sampling point on the particle surface; and when the sampling radius When , it indicates the sampling point near the particle. Indicates the center angle range , using uniform sampling. Using the self-differentiation mechanism of the particle network, the derivative of the particle velocity with respect to time is calculated , In the Stokes region, the drag coefficient , the particle dynamic equation can be constructed based on the particle velocity and derivative. The particle dynamic equation can be expressed by formula (11) and (12): (11) (12) in, represents the dimensionless particle density , represents the dimensionless particle diameter .
[0051] In addition, the boundary condition equations of fluid and particles can be set, which can be expressed by formulas (13)-(16):
[0052] (15) (16) Among them, inlet is the condition related to the entrance, outlet is the condition related to the exit, and wall is the condition related to the wall.
[0053] Step 230 , embedding the fluid momentum equation and the constitutive equation into the particle dynamic equation; substituting the flow field velocity into the embedded particle dynamic equation to obtain a first composite function.
[0054] In fluid-solid coupling, the coupling effect of the flow network on the particle network occurs in the particle dynamic equation and , obtain the input data of the fluid sampling points near the particle , input it into the flow network, and calculate the output data through the fluid momentum equation and constitutive equation. The output data includes the flow field velocity at the sampling point near the particle ( , ), the flow field velocity , Substituting into the particle dynamic equation, the first composite function is generated, which can be expressed by formula (17): (17) Step 240: embed the first composite function into the boundary equation; substitute the particle velocity into the embedded boundary equation to obtain a second composite function.
[0055] The coupling effect of the particle network on the flow network occurs in the boundary equation , get the input data of the particle sampling points on the boundary , input it into the particle network, and calculate the particle velocity through the first composite function ( , ), particle velocity ( , ) is substituted into the boundary equation to generate the second composite function, which can be expressed by formula (18): (18) Step 250, constructing a first loss function of the flow network based on the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; constructing a second loss function of the particle network based on the particle dynamic equation; the first loss function is used to adjust the flow network, and the second loss function is used to adjust the particle network.
[0056] In the loss calculation, a first loss function for the flow network is constructed based on the fluid momentum equation, constitutive equation, boundary equation, and a second composite function. A second loss function for the particle network is constructed based on the particle dynamic equation. Backpropagation is performed on the flow network and the particle network, respectively, based on the first and second loss functions. The weights and biases for the next forward propagation are then determined. The flow network and the particle network are then adjusted based on their respective weights and biases until the values of the first and second loss functions fall below a threshold, or the number of iterations reaches a maximum.
[0057] In an optional embodiment of the present application, step 250 includes: Set the first initial condition equation of the fluid physics field; Setting the first coefficients of the first initial condition equation, the fluid momentum equation, the constitutive equation, the boundary equation, and the second composite function respectively; A first loss function is constructed based on the first coefficient, the first initial condition equation, the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function.
[0058] The first initial condition equation of the fluid physics field can be expressed by formula (19): (19) The first initial condition equations are , the fluid momentum equation 、 , constitutive equation - , boundary equation - Setting the first coefficient, in one example, the fluid momentum equation can be 、 and constitutive equation - The first coefficient is set to 1, and the first initial condition equation and boundary equations - The first coefficient is set to 2, and the second composite function The first coefficient is set to 5.
[0059] Calculate the loss function of each equation, multiply the loss function of each equation by the corresponding coefficient and then add them together to obtain the first loss function of the flow network. The first loss function can be expressed by formula (20): (20) In an optional embodiment of the present application, step 250 further includes: Set the second initial condition equation of the solid particle motion field; Setting the second coefficients for the second initial condition equation and the particle dynamic equation respectively; A second loss function is constructed based on the second coefficient, the second initial condition equation and the particle dynamic equation.
[0060] The second initial condition equation of the solid particle motion field can be expressed by formula (21): (twenty one) The second initial condition equations are , particle dynamic equation 、 , set the second coefficient. In one example, the particle dynamic equation can be 、 The second coefficient is set to 1, and the second initial condition equation The second coefficient is set to 5.
[0061] Calculate the loss function of each equation, multiply the loss function of each equation by the corresponding coefficient and then add them together to obtain the second loss function of the particle network. The second loss function can be expressed by formula (22): 5 (twenty two) Among them, the loss function of each equation is obtained by the mean square error of the equation, and the loss function can be expressed by formula (23): (twenty three) in, represents the fluid momentum equation, constitutive equation, particle dynamic equation, initial condition equation or boundary condition equation, Indicates the corresponding number of sampling points.
[0062] refer to Figure 3 The velocity field prediction result obtained after adopting the solution of the embodiment of the present application is shown in the figure, where the horizontal axis represents the spatial coordinate. , the vertical axis represents the spatial coordinate . refer to Figure 4 The particle motion trajectory prediction result obtained after adopting the solution of the embodiment of the present application, where the horizontal axis represents time , the vertical axis represents the horizontal axis of the motion trajectory ( ). refer to Figure 5This is a schematic diagram of the physical information neural network structure, where "I" means that the data on the left is directly transmitted to the right without any processing. The output of the flow network is as follows: Figure 3 As shown, the output of the particle network is as follows Figure 4 shown.
[0063] In order to better understand the technical solutions of the embodiments of this application, refer to Figure 6 , is a schematic diagram of the working condition shown in the embodiment of the present application: First, determine the spatiotemporal control domain of the flow and particle motion time control domain The embodiment of this application is mainly aimed at two-dimensional transient processes, so It should contain two orthogonal spatial dimensions and one time dimension. Should contain a velocity entry ( ), a pressure outlet ( ) and multiple wall boundaries. According to the inlet width and inlet flow rate, the reference physical quantity is determined respectively and Then, according to the density and viscosity of the fluid, the dimensionless number is determined. Determine the dimensionless particle density based on the particle density and diameter and dimensionless particle diameter .
[0064] Set the number of sampling points and the distribution of sampling points respectively. It mainly includes the fluid time and space control domain The sampling points in the velocity inlet boundary are also included. , pressure outlet , wall The sampling points on the For the particle motion time control domain, set the time sampling point and the initial conditions. The sampling points. Then, the sampling point coordinates and time are dimensionlessly normalized according to the reference physical quantity and substituted into the physical information neural network of the embodiment of the present application as the input set. The model finally obtained after the training process is , including two neural network architectures, flow network flow-net and particle network particle-net, as well as their weights and bias data, for interpolated physical quantity prediction.
[0065] The prediction process can be carried out in the same flow spatiotemporal control domain and particle motion time control domain Within, any other sampling points are selected and substituted into the model , and get the flow velocity at the corresponding point , and the particle speed .
[0066] The embodiments of the present application provide a method for calculating the loss of fluid-particle coupling, which fully considers the driving or hindering effect of the fluid on the particles, as well as the changes in the flow domain and the reaction to the flow field caused by the movement of the particles, thereby improving the accuracy of the prediction of physical quantities related to the fluid and particles and ensuring the stability of the model; under the conditions of known fluid density, dynamic viscosity, spherical particle diameter and density, the model can predict the motion trajectory of spherical solid particles and the flow field distribution in the flow domain.
[0067] Corresponding to the aforementioned embodiment of the method for realizing the application function, the present application also provides a fluid and particle coupling loss calculation device, an electronic device and corresponding embodiments.
[0068] Figure 7 Schematic diagram of the structure of the fluid and particle coupling loss calculation device shown in an embodiment of the present application.
[0069] See also Figure 7 The device is applied to a physical information neural network, which includes a flow network and a particle network. The flow network is used to fit the flow physical field, and the particle network is used to fit the solid particle motion field. The device includes: An acquisition module 710 is configured to respectively acquire flow field physical quantities output by the flow network and particle physical quantities output by the particle network; the flow field physical quantities include at least flow field velocity, flow field pressure, and flow field stress tensor, and the particle physical quantities include at least particle velocity; A construction module 720 is used to construct the fluid momentum equation and constitutive equation of the fluid, the particle dynamic equation and boundary equation of the particle based on the flow field physical quantities and the particle physical quantities; A coupling module 730 is configured to generate a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation, and the particle dynamic equation; and to generate a second composite function based on the particle velocity, the first composite function, and the boundary equation. The loss function module 740 is used to construct a first loss function of the flow network based on the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; and to construct a second loss function of the particle network based on the particle dynamic equation; the first loss function is used to adjust the flow network, and the second loss function is used to adjust the particle network.
[0070] In an optional embodiment of the present application, the coupling module 730 includes: The first embedding submodule is used to embed the fluid momentum equation and the constitutive equation into the particle dynamic equation; The first composite function submodule is used to substitute the flow field velocity into the embedded particle dynamic equation to obtain the first composite function.
[0071] In an optional embodiment of the present application, the coupling module 730 further includes: a second embedding submodule, for embedding the first composite function into the boundary equation; The second composite function submodule is used to substitute the particle velocity into the embedded boundary equation to obtain a second composite function.
[0072] In an optional embodiment of the present application, the loss function module 740 includes: A first initial submodule, used for setting a first initial condition equation of a fluid physical field; A first coefficient submodule is used to set respective first coefficients for the first initial condition equation, the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; The first loss function submodule is used to construct a first loss function based on the first coefficient, the first initial condition equation, the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function.
[0073] In an optional embodiment of the present application, the loss function module 740 further includes: The second initial condition submodule is used to set the second initial condition equation of the solid particle motion field; A second coefficient submodule, for setting respective second coefficients for the second initial condition equation and the particle dynamic equation; The second loss function submodule is used to construct a second loss function based on the second coefficient, the second initial condition equation and the particle dynamic equation.
[0074] An embodiment of the present application provides a loss calculation device for fluid-particle coupling, which fully considers the driving or hindering effect of the fluid on the particles, as well as the changes in the flow domain and the reaction to the flow field caused by the movement of the particles, thereby improving the accuracy of the prediction of physical quantities related to the fluid and particles and ensuring the stability of the model; under the conditions of known fluid density, dynamic viscosity, spherical particle diameter and density, the model can predict the motion trajectory of spherical solid particles and the flow field distribution in the flow domain.
[0075] Regarding the apparatus in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated again here.
[0076] Figure 8 It is a structural diagram of an electronic device shown in an embodiment of the present application.
[0077] See also Figure 8 , the electronic device 800 includes a memory 810 and a processor 820.
[0078] The processor 820 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. Memory 810 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage. ROM may store static data or instructions required by processor 820 or other computer modules. Permanent storage may be a readable and writable storage device. Permanent storage may be a non-volatile storage device that maintains stored instructions and data even when the computer is powered off. In some embodiments, the permanent storage device utilizes a mass storage device (e.g., a magnetic or optical disk, flash memory). In other embodiments, the permanent storage device may be a removable storage device (e.g., a floppy disk, optical drive). System memory may be a readable and writable storage device or a volatile readable and writable storage device, such as dynamic random access memory (DRAM). System memory may store some or all instructions and data required by the processor during operation. Furthermore, memory 810 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (e.g., DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), as well as magnetic disks and / or optical disks. In some embodiments, the memory 810 may include a readable and / or writable removable storage device, such as a compact disc (CD), a read-only digital versatile disc (e.g., DVD-ROM, dual-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (e.g., SD card, mini SD card, Micro-SD card, etc.), a magnetic floppy disk, etc. Computer-readable storage media do not include carrier waves and transient electronic signals transmitted wirelessly or wired.
[0079] The memory 810 stores executable codes. When the executable codes are processed by the processor 820 , the processor 820 may execute part or all of the above-mentioned methods.
[0080] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.
[0081] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium), which stores executable code (or computer program or computer instruction code) and, when executed by a processor of an electronic device (or server, etc.), enables the processor to perform part or all of the steps of the above-mentioned method according to the present application.
[0082] The present application also provides a computer program product, which includes computer instructions, and when the computer instructions are executed by a processor, the method described above is implemented.
[0083] The embodiments of the present application have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to the technology in the market, or to enable other persons skilled in the art to understand the embodiments disclosed herein.
Claims
1. A method for calculating the loss of fluid-particle coupling, characterized in that: Applied to a physical information neural network, the physical information neural network includes a flow network and a particle network, the flow network is used to fit the flow physical field, and the particle network is used to fit the solid particle motion field; the method includes: Respectively obtaining flow field physical quantities output by the flow network and particle physical quantities output by the particle network; the flow field physical quantities include at least flow field velocity, flow field pressure, and flow field stress tensor, and the particle physical quantities include at least particle velocity; constructing a fluid momentum equation and a constitutive equation of the fluid, a particle dynamic equation and a boundary equation of the particles according to the flow field physical quantities and the particle physical quantities; generating a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation, and the particle dynamic equation, and generating a second composite function based on the particle velocity, the first composite function, and the boundary equation; A first loss function of the flow network is constructed according to the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; a second loss function of the particle network is constructed according to the particle dynamic equation; the first loss function is used to adjust the flow network, and the second loss function is used to adjust the particle network.
2. The method according to claim 1, characterized in that Generating a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation, and the particle dynamic equation includes: embedding the fluid momentum equation and the constitutive equation into the particle dynamic equation; Substituting the flow field velocity into the embedded particle dynamic equation obtains the first composite function.
3. The method according to claim 2, characterized in that Generating a second composite function based on the particle velocity, the first composite function, and the boundary equation comprises: embedding the first composite function into the boundary equation; Substituting the particle velocity into the embedded boundary equation yields a second composite function.
4. The method according to claim 1, wherein The first loss function of the flow network constructed according to the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function includes: Setting a first initial condition equation of the fluid physical field; Setting respective first coefficients for the first initial condition equation, the fluid momentum equation, the constitutive equation, the boundary equation, and the second composite function; The first loss function is constructed based on the first coefficient, the first initial condition equation, the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function.
5. The method according to claim 1, characterized in that The second loss function of the particle network constructed according to the particle dynamic equation includes: Setting a second initial condition equation for the solid particle motion field; Setting respective second coefficients for the second initial condition equation and the particle dynamic equation; The second loss function is constructed based on the second coefficient, the second initial condition equation and the particle dynamic equation.
6. The method according to claim 1, characterized in that The fluid momentum equation is constructed as follows: calculating partial derivatives of the flow field physical quantities with respect to input data of the flow network; constructing the fluid momentum equation based on the flow field velocity and the partial derivative; The constitutive equation is constructed using the following method: constructing the constitutive equation based on the partial derivative, the flow field pressure, and the flow field stress tensor; The particle dynamic equation of the particles is constructed as follows: calculating derivatives of the particle velocity with respect to input data of the particle network; The particle dynamic equation is constructed based on the particle velocity and the derivative.
7. The method according to claim 1, characterized in that The flow network and the particle network are two independent multi-layer perceptron networks.
8. A device for calculating loss of fluid-particle coupling, characterized in that: Applied to a physical information neural network, the physical information neural network includes a flow network and a particle network, the flow network is used to fit the flow physical field, and the particle network is used to fit the solid particle motion field; the device includes: an acquisition module, configured to respectively acquire flow field physical quantities output by the flow network and particle physical quantities output by the particle network; the flow field physical quantities at least include flow field velocity, flow field pressure, and flow field stress tensor, and the particle physical quantities at least include particle velocity; A construction module, configured to construct a fluid momentum equation and a constitutive equation of the fluid, a particle dynamic equation and a boundary equation of the particles according to the flow field physical quantities and the particle physical quantities; A coupling module is configured to generate a first composite function based on the flow field velocity, the fluid momentum equation, the constitutive equation, and the particle dynamic equation; and generate a second composite function based on the particle velocity, the first composite function, and the boundary equation; A loss function module is used to construct a first loss function of the flow network based on the fluid momentum equation, the constitutive equation, the boundary equation and the second composite function; and to construct a second loss function of the particle network based on the particle dynamic equation; the first loss function is used to adjust the flow network, and the second loss function is used to adjust the particle network.
9. An electronic device, characterized in that: include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having executable codes stored thereon, wherein when the executable codes are executed by a processor of an electronic device, the processor is caused to execute the method according to any one of claims 1 to 7.