Heat dissipation assembly type selection method and system based on three-dimensional simulation and PINN collaborative optimization

Abstract

The invention discloses a heat dissipation assembly type selection method and system based on three-dimensional simulation and PINN collaborative optimization. The problems that an existing radiator model selection method needs a large amount of calculation resources and time and is low in precision are solved. The method comprises the following steps: constructing a heat dissipation system three-dimensional model comprising a heat dissipation assembly, and simulating the three-dimensional model according to different working conditions of the heat dissipation assembly; acquiring working conditions and simulation output data to establish sample data; constructing a physical information neural network model, and training to obtain a heat dissipation performance prediction model; inputting the working condition data of the to-be-tested heat dissipation assembly into the heat dissipation performance prediction model to obtain heat dissipation performance prediction data; and the optimal heat dissipation assembly is selected by integrating the heat dissipation performance, the cost and the space limiting factors. According to the method, three-dimensional simulation and physical information neural network training are combined for application, and the radiator assembly meeting the heat dissipation requirement can be accurately and efficiently selected.

Description

Technical Field

[0001] This invention relates to the field of heat dissipation equipment selection technology, and in particular to a heat dissipation component selection method and system based on three-dimensional simulation and PINN collaborative optimization. Background Technology

[0002] With the increasing power density of electronic devices, heat dissipation has become a key factor restricting their performance and reliability. Fans and heat sinks, as commonly used heat dissipation components, are crucial for improving heat dissipation efficiency. Traditional selection methods mainly rely on empirical formulas and experimental tests, which suffer from low accuracy, long cycles, and high costs. For example, in some high-power electronic devices, improper selection of heat dissipation components can lead to excessively high temperatures during prolonged operation, resulting in performance degradation and shortened lifespan. In recent years, with the development of computer technology and numerical simulation methods, 3D simulation technology has been widely used in the field of heat dissipation. By establishing a 3D model of the heat dissipation system, the heat dissipation performance under different operating conditions can be simulated and analyzed, providing a certain reference for the selection of heat dissipation components. However, traditional 3D simulation methods often require a large amount of computing resources and time, and the accuracy of simulation results still needs to be improved for complex heat dissipation systems.

[0003] For example, patent number 202311691704.3, entitled "Selection Method for Electric Vehicle Radiators," calculates the coolant flow rate, required air volume, and air speed of the cooling system. Based on the coolant flow rate and air speed, it determines the heat dissipation and compares the heat dissipation of the electric drive with the heat released to select the appropriate radiator. However, this patent requires calculating the coolant flow rate, required air volume, and air speed for each radiator during the selection process, which consumes a large amount of computing resources and has a long cycle time. Summary of the Invention

[0004] This invention primarily addresses the problems of existing heat sink selection methods requiring significant computational resources and time, and exhibiting low accuracy. It provides a heat sink component selection method and system based on 3D simulation and PINN collaborative optimization.

[0005] The above-mentioned technical problems of the present invention are mainly solved by the following technical solution: a heat dissipation component selection method based on three-dimensional simulation and PINN collaborative optimization, comprising the following steps: Construct a 3D model of the heat dissipation system including heat dissipation components, and simulate the 3D model according to different operating conditions of the heat dissipation components; Acquire operating conditions and simulation output data to establish sample data; A physical information neural network model is constructed, and the model is trained based on sample data to obtain a heat dissipation performance prediction model. Input the operating condition data of the heat dissipation component under test into the heat dissipation performance prediction model to obtain heat dissipation performance prediction data; The optimal heat dissipation component was selected by considering factors such as overall heat dissipation performance, cost, and space constraints.

[0006] This invention combines 3D simulation with PINN (Physical Information Neural Network) training. By training the PINN with simulation data, a heat dissipation performance prediction model is obtained. Based on this model, the performance parameters of each heat dissipation unit are acquired, enabling accurate and efficient selection of heat dissipation components that meet the heat dissipation requirements. Selection of heat dissipation components can be achieved simply by using the trained model, eliminating the need for modeling each component. This reduces workload and time, lowers costs, and solves the problems of low accuracy, long cycles, and high costs associated with existing heat dissipation component selection methods.

[0007] As a preferred embodiment, the steps for constructing a three-dimensional model of a heat dissipation system including heat dissipation components include: Based on the structural dimensions of the heat dissipation system, a three-dimensional model is constructed, including the heat source, heat dissipation components, and air domain. The heat dissipation components include the fan and heat sink fins. Set simulation parameters and boundary conditions; Establish the three-dimensional simulation control equations.

[0008] The heat dissipation components in this solution mainly refer to the fan and heat sink fins. Constructing the 3D model of the heat dissipation system involves using 3D modeling software to accurately draw the chip as a heat source model for electronic device heat dissipation systems, as well as drawing the shape and size of the fan blades and the shape and arrangement of the heat sink fins. The constructed 3D model of the heat dissipation system specifically includes: Heat source model: A rectangular structure is used to simulate the heat source of the electronic chip. Its dimensions are length a × width b × height c, where a, b, and c are determined according to the actual chip size, for example, 10mm × 10mm × 2mm. The thermal productivity of the heat source is q, in W / m³. 3 The thermal productivity q of the heat source is calculated by the relationship between the chip's heating power P and the heat source volume V, which is: heating power P / heat source volume V.

[0009] Fan Model: An axial flow fan model is adopted, consisting of a hub and blades. The hub diameter is d1, the blade chord convergence is n, and the blade chord length varies radially according to a certain rule, such as l1 linearly increasing from the hub to l2 at the blade tip. The blade installation angle is taken within the range of 15°-30°. The fan's rotation region is processed using sliding mesh technology to simulate the fan's rotational effect.

[0010] Heat dissipation fin model: A rectangular fin array is used, with fins made of aluminum (thermal conductivity k = 170 W / (m·K)). The dimensions of a single fin are length L × width W × thickness t, the fin spacing is s, the fin height is h, the number of rows is m, and the number of columns is p. A close contact model is used between the fins and the heat source, and the contact thermal resistance is negligible. Air domain model: A cuboid region surrounds the heat source, fan, and heat sink fins, with a size 3-5 times the overall size of the heat source, fan, and heat sink fins to ensure sufficient flow field development. The air domain's inlet and outlet are set in appropriate locations to simulate the actual airflow environment. The model is meshed using appropriate mesh types and sizes to minimize computational load while maintaining accuracy. For example, finer meshes are used on heat sources and heat sink fin surfaces to accurately capture temperature and flow field changes, while coarser meshes are used in air regions far from the heat sink components.

[0011] Set simulation parameters, including the physical parameters of air, such as density ρ = 1.225 kg / m³. 3 Specific heat capacity c p =10006 J / (kg·K), thermal conductivity k=0.026 W / (m·K), heat source power, fan speed, airflow, air pressure, and other parameters. These parameters should be set appropriately according to the actual application scenario. For example, for electronic devices operating in a normal temperature environment, the physical properties of air at the corresponding temperature should be used, and the heat source power should be determined based on the chip's rated power.

[0012] Define boundary conditions, such as setting constant heat flux density boundary conditions on the heat source surface, velocity inlet and pressure outlet boundary conditions at the fan inlet and outlet, and no-slip boundary conditions on the model walls. Ensure that the boundary conditions are set in accordance with actual physical conditions to improve the accuracy of simulation results.

[0013] Establish three-dimensional simulation control equations, including continuity equations, momentum equations, and energy equations.

[0014] As a preferred approach, the three-dimensional simulation control equations include: The continuity equation states that the partial derivative of air density with respect to time plus the divergence of the product of air density and velocity vector equals 0. The momentum equation states that the product of air density and the first element is equal to the sum of the negative pressure gradient and the divergence of the second element, as well as the product of air density and the vector product of gravitational acceleration. The first unit is the partial derivative of the velocity vector with respect to time plus the product of the velocity vector and the gradient of the velocity vector. The second unit is the sum of the velocity vector gradient and the transpose of the velocity vector gradient multiplied by the dynamic viscosity; The energy equation, the product of air density and specific heat capacity in the third unit, equals the sum of divergence and heat source term in the fourth unit; The third unit is the partial derivative of temperature with respect to time plus the dot product of the velocity vector and the temperature gradient. The fourth unit is the product of thermal conductivity and temperature gradient.

[0015] Specifically, the continuity equation is expressed as follows: The partial derivative of air density ρ with respect to time t is +∇・(product of air density ρ and velocity vector u) = 0. For steady-state flow, the partial derivative of air density ρ with respect to time t is 0, and the continuity equation simplifies to ∇・(product of air density ρ and velocity vector u) = 0.

[0016] The momentum equation is specifically expressed as: the product of air density ρ and the first element = -∇pressure p + divergence of the second element + the product of air density ρ and gravitational acceleration vector g; The first unit is the partial derivative of velocity vector u with respect to time t, plus the product of velocity vector u and ∇ velocity vector u; The second unit is (the sum of the ∇ velocity vector u and the transpose of the ∇ velocity vector u) * dynamic viscosity μ.

[0017] Where ∇ velocity vector u is the gradient of velocity vector u, and for dynamic viscosity μ, when air is at 25℃, dynamic viscosity μ = 1.81 * 10 -5 Pa・s.

[0018] The energy equation is specifically expressed as: air density ρ and specific heat capacity c p The product of the third unit equals the divergence and heat source term S of the fourth unit. t sum; Among them, for the heat source region S t =q, other regions S t =0.

[0019] The third unit is the partial derivative of temperature T with respect to time t, plus the dot product of velocity vector u and ∇temperature T; ∇temperature T is the temperature gradient; The fourth unit is the product of thermal conductivity k and ∇temperature T.

[0020] As a preferred embodiment, the simulation of the 3D model according to different operating conditions of the heat dissipation component includes: The finite volume method was used to discretize the three-dimensional simulation control equations. A first-order implicit scheme was used for the time term, a second-order upwind scheme for the convection term, and a central difference scheme for the diffusion term. The pressure-velocity coupling was performed using the SIMPLE algorithm. The iterative convergence criterion was that all residuals were less than 1*102. -6 .

[0021] This scheme uses computational fluid dynamics software to simulate the established three-dimensional model, solve the three-dimensional simulation control equations, and obtain the temperature field and flow field distribution in the heat dissipation system under different operating conditions. Through simulation, the heat transfer path in the heat dissipation system and the air flow can be seen intuitively, providing data support for subsequent analysis.

[0022] The simulation calculations used the finite volume method to discretize the governing equations. The time term employed a first-order implicit scheme, the convection term a second-order upwind scheme, and the diffusion term a central difference scheme. The pressure-velocity coupling was performed using the SIMPLE algorithm, and the iterative convergence criterion was that all residuals were less than 1*10. -6 Furthermore, the flow field distribution no longer changes with the number of iterations in temperature-dependent applications.

[0023] As a preferred approach, the operating conditions of the heat dissipation components are obtained, including fan speed, number of blades, fin height, fin spacing, and fin thickness. Based on the simulation results, the performance parameters corresponding to the working conditions of the heat dissipation component are obtained. The performance parameters include the highest surface temperature of the heat dissipation fins, the lowest surface temperature of the heat dissipation fins, the average surface temperature of the heat dissipation fins, the air outlet temperature, and the heat dissipation efficiency. The operating conditions and performance parameters of the heat dissipation components are preprocessed to generate sample data.

[0024] This solution acquires simulation data on the operating conditions and results of the heat dissipation components in the simulation model. The heat dissipation components include a fan and heat sink fins. The operating conditions include fan speed, number of blades, fin height, fin spacing, and fin thickness. The simulation results are the performance parameters of the heat dissipation components, including the highest surface temperature of the heat sink fins, the lowest surface temperature of the heat sink fins, the average surface temperature of the heat sink fins, the air outlet temperature, and the heat dissipation efficiency. The obtained operating conditions and performance parameters are preprocessed using normalization, mapping them to the [0,1] interval. The normalization method is min-max normalization. After preprocessing, sample data is obtained, each containing the operating conditions of the heat dissipation components and the corresponding simulation performance parameters.

[0025] As a preferred option, the physical information neural network model includes: The input layer contains 8 neurons, which correspond to the operating conditions of the heat sink components and air parameters, respectively. The hidden layers consist of three layers, each containing 32, 64, and 32 neurons respectively, all using the ReLU activation function, and all layers are fully connected. The output layer contains 5 neurons, and the outputs correspond to the performance parameters.

[0026] Set the model's overall loss function.

[0027] This scheme uses a Physical Information Neural Network (PINN) model, including the number of neurons in the input layer, hidden layer, and output layer. The input layer mainly takes in parameters related to the heat dissipation system, such as fan speed, heat sink fin size, and heat source power. The hidden layer uses a suitable activation function (such as the ReLU function) for nonlinear transformation. The output layer outputs the predicted heat dissipation performance parameters, such as heat dissipation efficiency and temperature distribution.

[0028] The specific structure of the physical information neural network model is as follows: The input layer contains 8 neurons, with the following input parameters: fan speed n (rpm), number of blades N, fin height h (mm), fin spacing s (mm), fin thickness t (mm), heat source power P (W), and air density ρ (kg / m³). 3 ), air thermal conductivity k (W / (m·K)).

[0029] The hidden layers consist of three layers: the first layer contains 32 neurons using the ReLU activation function, the second layer contains 64 neurons using the ReLU activation function, and the third layer contains 32 neurons using the ReLU activation function. All layers are fully connected, and the weights and biases are determined during training.

[0030] The output layer contains 5 neurons, each outputting the highest surface temperature T of the heat sink fins. max (°C), Minimum surface temperature of heat sink fins T min (°C), Average surface temperature of heat sink fins (T) ave (°C), Air outlet temperature T out (°C), heat dissipation efficiency η (%), where heat dissipation efficiency η is the surface temperature of the heat source T. source and air outlet temperature T out The difference is related to the surface temperature T of the heat source. source and ambient air temperature T air The ratio of the differences.

[0031] Determine the comprehensive loss function, which combines the residuals of the physical model and the data fitting error as the loss function.

[0032] As a preferred approach, the comprehensive loss function includes both data loss and physical loss, and is a weighted sum of the data loss and physical loss. The data loss is calculated using mean squared error, and the physical loss is obtained by inputting the prediction results of the physical information neural network model into the control equation to calculate the residual.

[0033] In this scheme, the comprehensive loss function L is composed of the data loss L. d and physical loss L p It consists of two parts, specifically data loss L. d and physical loss Lp The weighted sum, data loss L d and physical loss L p The weights are summed to 1, where the data loss L d The weight is set to 0.7. Data loss function L d Mean square error is used. Physical loss L p The residuals are obtained by substituting the model's predicted structure into the continuity equation, momentum equation, and energy equation, and are expressed as: 1 / M∑(continuity equation residuals) 2 +Momentum equation residuals 2 + Energy equation residuals 2 ), where M represents the number of physical constraint points.

[0034] As a preferred embodiment, the step of training the model based on evolutionary data to obtain a heat dissipation performance prediction model includes: The sample data is divided into training and testing sets. The physical information neural network model is trained and tested. A comprehensive loss function is used, and backpropagation is performed through the Adam optimizer to iteratively optimize the model parameters, minimize the comprehensive loss function, and obtain the heat dissipation performance prediction model after training and passing the test.

[0035] In this scheme, the sample data is divided proportionally, typically in a 7:3 ratio, into training and test sets. The training set is input into the physical information neural network model for training, using the Adam optimizer for parameter updates. The initial learning rate is set to 0.001, gradually decreasing with each training iteration (decreasing to 0.9 times the original rate every 1000 iterations). The partial derivatives of the loss function with respect to each weight and bias are calculated, and the weights and biases are updated via backpropagation using the chain rule. A maximum of 10,000 iterations is set, and each iteration uses a batch of samples (batch size = 32) for training. When the loss function value is less than 1 * 10^- ... -4 Training may stop when the maximum number of iterations is reached. After training, the performance of the physical information neural network model is evaluated using test set data, and the RMSE and coefficient of determination R are calculated. 2 (Coefficient of determination R) 2 The closer the value is to 1, the better the model performance. If the performance is not up to standard, adjust the number of neurons in the hidden layer of the physical information neural network model, the learning rate, or increase the training data, and retrain the model. After the performance meets the standard, obtain the heat dissipation performance prediction model based on the trained physical information neural network model.

[0036] As a preferred approach, different operating conditions of the heat dissipation components are input into the heat dissipation performance prediction model to obtain the corresponding heat dissipation performance prediction data of the heat dissipation components. Based on performance prediction data, heat dissipation components with lower cost and smaller footprint are selected, provided that heat dissipation requirements are met.

[0037] A heat dissipation performance prediction model is deployed, and the parameters of different fan and heat sink fin combinations, i.e., the operating data of the heat dissipation components, are input into the model to obtain corresponding prediction results such as heat dissipation efficiency and temperature distribution. Based on the prediction results, considering factors such as heat dissipation performance, cost, and space constraints, the optimal fan and heat sink fin combination is selected. Preferably, under the premise of meeting the heat dissipation requirements, a heat dissipation component with lower cost and smaller space occupation is selected as the final heat dissipation component.

[0038] A heat dissipation component selection system based on 3D simulation and PINN collaborative optimization includes: The simulation unit constructs a three-dimensional model of the heat dissipation system, including heat dissipation components, and simulates the three-dimensional model according to different operating conditions of the heat dissipation components. The data acquisition and processing unit acquires operating condition and simulation output data to establish sample data. The neural network unit constructs a physical information neural network model, trains the model based on sample data, and obtains a heat dissipation performance prediction model after training. The selection unit deploys a heat dissipation performance prediction model, inputs the operating condition data of the heat dissipation component under test to obtain heat dissipation performance prediction data, and selects the optimal heat dissipation component by comprehensively considering heat dissipation performance, cost and space constraints.

[0039] Therefore, the advantages of the present invention are: 1. By combining 3D simulation with physical information neural network training, it is possible to accurately and efficiently select heat sink components that meet heat dissipation requirements.

[0040] 2. The performance parameters of heat dissipation components can be obtained and selected based on the heat dissipation performance prediction model after training, without the need to model each heat dissipation component, which reduces workload and time and lowers costs. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of a framework structure of the system of the present invention.

[0042] Figure 2 This is a flowchart illustrating one method of the present invention.

[0043] 1-Simulation Unit 2-Data Acquisition and Processing Unit 3-Neural Network Unit 4-Selection Unit. Detailed Implementation

[0044] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0045] Example: This embodiment presents a method for selecting heat dissipation components based on 3D simulation and PINN collaborative optimization, such as... Figure 2 As shown, it includes the following steps: S1. Construct a 3D model of the heat dissipation system including heat dissipation components, and simulate the 3D model according to different operating conditions of the heat dissipation components.

[0046] As a preferred embodiment, the specific process of constructing the three-dimensional model of the heat dissipation system includes: S11. Based on the structural dimensions of the heat dissipation system, construct a three-dimensional model including the heat source, heat dissipation components, and air domain. The heat dissipation components include the fan and heat dissipation fins. S12. Set simulation parameters and boundary conditions; S13. Establish the three-dimensional simulation control equations.

[0047] In this embodiment, the heat dissipation components mainly refer to the fan and heat sink fins. Constructing a three-dimensional model of the heat dissipation system includes using three-dimensional modeling software to accurately draw the chip as a heat source model for the heat dissipation system of electronic devices, as well as drawing the shape and size of the fan blades and the shape and arrangement of the heat sink fins.

[0048] Specifically, the constructed 3D model of the heat dissipation system includes: Heat source model: A rectangular structure is used to simulate the heat source of the electronic chip. Its dimensions are length a × width b × height c, where a, b, and c are determined according to the actual chip size, for example, 10mm × 10mm × 2mm. The thermal productivity of the heat source is q, in W / m³. 3 The thermal productivity q of the heat source is calculated by the relationship between the chip's heating power P and the heat source volume V, which is: heating power P / heat source volume V.

[0049] Fan Model: An axial flow fan model is adopted, consisting of a hub and blades. The hub diameter is d1, the blade chord convergence is n, and the blade chord length varies radially according to a certain rule, such as l1 linearly increasing from the hub to l2 at the blade tip. The blade installation angle is taken within the range of 15°-30°. The fan's rotation region is processed using sliding mesh technology to simulate the fan's rotational effect.

[0050] Heat dissipation fin model: A rectangular fin array is used, with fins made of aluminum (thermal conductivity k = 170 W / (m·K)). The dimensions of a single fin are length L × width W × thickness t, the fin spacing is s, the fin height is h, the number of rows is m, and the number of columns is p. A close contact model is used between the fins and the heat source, and the contact thermal resistance is negligible. Air domain model: A cuboid region surrounds the heat source, fan, and heat sink fins, with a size 3-5 times the overall size of the heat source, fan, and heat sink fins to ensure sufficient flow field development. The air domain's inlet and outlet are set in appropriate locations to simulate the actual airflow environment. The model is meshed using appropriate mesh types and sizes to minimize computational load while maintaining accuracy. For example, finer meshes are used on heat sources and heat sink fin surfaces to accurately capture temperature and flow field changes, while coarser meshes are used in air regions far from the heat sink components.

[0051] Specifically, set the simulation parameters, including the physical parameters of air, such as density ρ = 1.225 kg / m³. 3 Specific heat capacity c p =1006 J / (kg·K), thermal conductivity k=0.026 W / (m·K), heat source power, fan speed, airflow, air pressure, and other parameters. These parameters should be set appropriately according to the actual application scenario. For example, for electronic devices operating in a normal temperature environment, the physical properties of air at the corresponding temperature should be used, and the heat source power should be determined based on the chip's rated power.

[0052] Define boundary conditions, such as setting constant heat flux density boundary conditions on the heat source surface, velocity inlet and pressure outlet boundary conditions at the fan inlet and outlet, and no-slip boundary conditions on the model walls. Ensure that the boundary conditions are set in accordance with actual physical conditions to improve the accuracy of simulation results.

[0053] Establish three-dimensional simulation control equations, including continuity equations, momentum equations, and energy equations.

[0054] Specifically, in the continuity equation, the partial derivative of air density with respect to time plus the divergence of the product of air density and velocity vector equals 0. The momentum equation states that the product of air density and the first element is equal to the sum of the negative pressure gradient and the divergence of the second element, as well as the product of air density and the vector product of gravitational acceleration. The first unit is the partial derivative of the velocity vector with respect to time plus the product of the velocity vector and the gradient of the velocity vector. The second unit is the sum of the velocity vector gradient and the transpose of the velocity vector gradient multiplied by the dynamic viscosity; The energy equation, the product of air density and specific heat capacity in the third unit, equals the sum of divergence and heat source term in the fourth unit; The third unit is the partial derivative of temperature with respect to time plus the dot product of the velocity vector and the temperature gradient. The fourth unit is the product of thermal conductivity and temperature gradient.

[0055] Based on the above description, the further continuity equation can be expressed as: partial derivative of air density ρ with respect to time t + ∇・(product of air density ρ and velocity vector u) = 0. For steady-state flow, the partial derivative of air density ρ with respect to time t is 0, and the continuity equation simplifies to ∇・(product of air density ρ and velocity vector u) = 0.

[0056] The momentum equation is expressed as: the product of air density ρ and the first element = -∇pressure p + divergence of the second element + the product of air density ρ and gravitational acceleration vector g; The first unit is: the partial derivative of velocity vector u with respect to time t + the product of velocity vector u and ∇ velocity vector u; The second unit is: (the sum of the ∇ velocity vector u and the transpose of the ∇ velocity vector u) * dynamic viscosity μ.

[0057] Where ∇ velocity vector u is the gradient of velocity vector u, and for dynamic viscosity μ, when air is at 25℃, dynamic viscosity μ = 1.81 * 10 -5 Pa・s.

[0058] The energy equation is expressed as: air density ρ and specific heat capacity c p The product of the third unit equals the divergence and heat source term S of the fourth unit. t sum; Among them, for the heat source region S t =q, other regions S t =0.

[0059] The third unit is the partial derivative of temperature T with respect to time t, plus the dot product of velocity vector u and ∇temperature T; ∇temperature T is the temperature gradient; The fourth unit is the product of thermal conductivity k and ∇temperature T.

[0060] As a preferred embodiment, the three-dimensional model is simulated according to different operating conditions of the heat dissipation component, including: S14. The three-dimensional simulation control equations are discretized using the finite volume method. The time term uses a first-order implicit scheme, the convection term uses a second-order upwind scheme, and the diffusion term uses a central difference scheme. The pressure-velocity coupling is performed using the SIMPLE algorithm. The iterative convergence criterion is that all residuals are less than 1*10. -6 .

[0061] This scheme uses computational fluid dynamics software to simulate the established three-dimensional model, solve the three-dimensional simulation control equations, and obtain the temperature field and flow field distribution in the heat dissipation system under different operating conditions. Through simulation, the heat transfer path in the heat dissipation system and the air flow can be seen intuitively, providing data support for subsequent analysis.

[0062] The simulation calculations used the finite volume method to discretize the governing equations. The time term employed a first-order implicit scheme, the convection term a second-order upwind scheme, and the diffusion term a central difference scheme. The pressure-velocity coupling was performed using the SIMPLE algorithm, and the iterative convergence criterion was that all residuals were less than 1*10. -6 Furthermore, the flow field distribution no longer changes with the number of iterations in temperature-dependent applications.

[0063] S2. Obtain operating conditions and simulation output data to establish sample data.

[0064] As a preferred embodiment of this solution, it specifically includes: S21. Obtain the operating conditions of the heat dissipation components, including fan speed, number of blades, fin height, fin spacing, and fin thickness; S22. Based on the simulation results, obtain the performance parameters corresponding to the working conditions of the heat dissipation component. The performance parameters include the highest surface temperature of the heat dissipation fins, the lowest surface temperature of the heat dissipation fins, the average surface temperature of the heat dissipation fins, the air outlet temperature, and the heat dissipation efficiency. S23. Preprocess the operating conditions and performance parameters of the heat dissipation components to generate sample data.

[0065] This solution acquires simulation data on the operating conditions and results of the heat dissipation components in the simulation model. The heat dissipation components include a fan and heat sink fins. The operating conditions include fan speed, number of blades, fin height, fin spacing, and fin thickness. The simulation results are the performance parameters of the heat dissipation components, including the highest surface temperature of the heat sink fins, the lowest surface temperature of the heat sink fins, the average surface temperature of the heat sink fins, the air outlet temperature, and the heat dissipation efficiency. The obtained operating conditions and performance parameters are preprocessed using normalization, mapping them to the [0,1] interval. The normalization method is min-max normalization. After preprocessing, sample data is obtained, each containing the operating conditions of the heat dissipation components and the corresponding simulation performance parameters.

[0066] S3. Construct a physical information neural network model, train the model based on sample data, and obtain a heat dissipation performance prediction model after training.

[0067] As a preferred embodiment, the constructed physical information neural network model structure includes: The input layer contains 8 neurons, which correspond to the operating conditions of the heat sink components and air parameters, respectively. The hidden layers consist of three layers, each containing 32, 64, and 32 neurons respectively, all using the ReLU activation function, and all layers are fully connected. The output layer contains 5 neurons, and the outputs correspond to the performance parameters.

[0068] Set the model's overall loss function.

[0069] Specifically, this step sets up a Physical Information Neural Network (PINN) model, including the number of neurons in the input layer, hidden layer, and output layer. The input layer mainly takes in parameters related to the heat dissipation system, such as fan speed, heat sink fin size, and heat source power; the hidden layer uses a suitable activation function (such as the ReLU function) for nonlinear transformation; and the output layer outputs the predicted heat dissipation performance parameters, such as heat dissipation efficiency and temperature distribution.

[0070] The specific structure of the physical information neural network model is as follows: The input layer contains 8 neurons, with the following input parameters: fan speed n (rpm), number of blades N, fin height h (mm), fin spacing s (mm), fin thickness t (mm), heat source power P (W), and air density ρ (kg / m³). 3 ), air thermal conductivity k (W / (m·K)).

[0071] The hidden layers consist of three layers: the first layer contains 32 neurons using the ReLU activation function, the second layer contains 64 neurons using the ReLU activation function, and the third layer contains 32 neurons using the ReLU activation function. All layers are fully connected, and the weights and biases are determined during training.

[0072] The output layer contains 5 neurons, each outputting the highest surface temperature T of the heat sink fins. max (°C), Minimum surface temperature of heat sink fins T min (°C), Average surface temperature of heat sink fins (T) ave (°C), Air outlet temperature T out (°C), heat dissipation efficiency η (%), where heat dissipation efficiency η is the surface temperature of the heat source T. source and air outlet temperature T out The difference is related to the surface temperature T of the heat source. source and ambient air temperature T air The ratio of the differences.

[0073] Determine the comprehensive loss function, which combines the residuals of the physical model and the data fitting error as the loss function.

[0074] As a preferred embodiment, the comprehensive loss function includes data loss and physical loss, and is a weighted sum of data loss and physical loss. The data loss is calculated using mean squared error, and the physical loss is obtained by inputting the prediction results of the physical information neural network model into the control equation to calculate the residual.

[0075] Specifically, in this scheme, the comprehensive loss function L is composed of the data loss L... d and physical loss L p It consists of two parts, specifically data loss L.d and physical loss L p The weighted sum, data loss L d and physical loss L p The weights are summed to 1, where the data loss L d The weight is set to 0.7. Data loss function L d Mean square error is used. Physical loss L p The residuals are obtained by substituting the model's predicted structure into the continuity equation, momentum equation, and energy equation, and are expressed as: 1 / M∑(continuity equation residuals) 2 +Momentum equation residuals 2 + Energy equation residuals 2 ), where M represents the number of physical constraint points.

[0076] As a preferred embodiment, after obtaining the dataset, the sample data is divided into a training set and a test set. The physical information neural network model is trained and tested. A comprehensive loss function is used, and backpropagation is performed through the Adam optimizer to iteratively optimize the model parameters, minimize the comprehensive loss function, and obtain the heat dissipation performance prediction model after training and passing the test.

[0077] In this scheme, the sample data is divided proportionally, typically in a 7:3 ratio, into training and test sets. The training set is input into the physical information neural network model for training, using the Adam optimizer for parameter updates. The initial learning rate is set to 0.001, gradually decreasing with each training iteration (decreasing to 0.9 times the original rate every 1000 iterations). The partial derivatives of the loss function with respect to each weight and bias are calculated, and the weights and biases are updated via backpropagation using the chain rule. A maximum of 10,000 iterations is set, and each iteration uses a batch of samples (batch size = 32) for training. When the loss function value is less than 1 * 10^- ... -4 Training may stop when the maximum number of iterations is reached. After training, the performance of the physical information neural network model is evaluated using test set data, and the RMSE and coefficient of determination R are calculated. 2 (Coefficient of determination R) 2 The closer the value is to 1, the better the model performance. If the performance is not up to standard, adjust the number of neurons in the hidden layer of the physical information neural network model, the learning rate, or increase the training data, and retrain the model. After the performance meets the standard, obtain the heat dissipation performance prediction model based on the trained physical information neural network model.

[0078] S4. Input the operating condition data of the heat dissipation component under test into the heat dissipation performance prediction model to obtain heat dissipation performance prediction data; select the optimal heat dissipation component by comprehensively considering heat dissipation performance, cost, and space constraints. Specifically, this includes: S41. Input the heat dissipation components under different operating conditions into the heat dissipation performance prediction model to obtain the corresponding heat dissipation performance prediction data of the heat dissipation components; S42. Based on the performance prediction data, select heat dissipation components that are low in cost and occupy little space, provided that the heat dissipation requirements are met.

[0079] Specifically, a heat dissipation performance prediction model is deployed. Parameters of different fan and heat sink fin combinations, i.e., the operating conditions of the heat dissipation components, are input into the model to obtain prediction results such as heat dissipation efficiency and temperature distribution. Based on the prediction results, considering factors such as heat dissipation performance, cost, and space constraints, the optimal fan and heat sink fin combination is selected. Preferably, under the premise of meeting heat dissipation requirements, a heat dissipation component with lower cost and smaller space occupation is selected as the final heat dissipation component.

[0080] The technical solution of the method of the present invention will be further described in detail below with reference to specific examples.

[0081] Taking a heat dissipation system of an electronic device as an example, the operation of the method of the present invention includes the following steps: S1. Construct a three-dimensional model of the heat dissipation system.

[0082] First, a 3D model of the electronic device's heat dissipation system was created using SolidWorks software. This included modeling the chip as a heat source with dimensions of 10mm × 10mm × 2mm, using a 50mm diameter axial fan with 7 blades, and aluminum heat sink fins with a height of 20mm, a spacing of 3mm, and a thickness of 1mm, arranged in an array above the heat source. The model was then meshed, using 0.5mm tetrahedral meshes on the heat source and heat sink fin surfaces, and 2mm meshes on the airflow. Then, the simulation parameters were set in Fluent software, with the air properties set according to the ambient temperature condition of 25℃ (density ρ = 1.225 kg / m³). 3 Specific heat capacity c p =1006 J / (kg·K), thermal conductivity k=0.026 W / (m·K)), the heat source's heating power is set to 50 W, and the fan speed is set to 3000 rpm. The boundary conditions are set as follows: the heat source surface has a constant heat flux density boundary condition, with a heat flux density of 50 W / cm³. 2 The fan inlet is a velocity inlet boundary condition, with the velocity calculated based on the fan speed and airflow. The fan outlet is a pressure outlet boundary condition, set to standard atmospheric pressure. The model wall is a no-slip boundary condition.

[0083] Establish three-dimensional simulation control equations, including continuity equations, momentum equations, and energy equations.

[0084] Next, three-dimensional model simulation calculations were performed. The finite volume method was used to discretize the three-dimensional simulation control equations. The time term used a first-order implicit scheme, the convection term used a second-order upwind scheme, and the diffusion term used a central difference scheme. The pressure-velocity coupling was performed using the SIMPLE algorithm. The iterative convergence criterion was that all residuals were less than 1*10. -6 After a period of iterative calculations, the temperature and flow field distribution within the heat dissipation system under this operating condition was obtained. The simulation results showed that the highest surface temperature of the heat sink fins was 65℃, and the fan outlet airflow was 0.05 m³ / s. 3 Data such as / s and wind pressure of 50Pa are used to select the corresponding data from the heat dissipation performance prediction model output.

[0085] S3. Construct a physical information neural network model.

[0086] The model structure is as follows: the input layer has 8 neurons, which are used to input fan speed, number of blades, fin height, fin spacing, fin thickness, heat source power, air density, and air thermal conductivity, respectively; the hidden layer has 3 layers, with 32, 64, and 32 neurons in each layer, respectively; the output layer has 5 neurons, which are used to output the predicted maximum surface temperature of the heat dissipation fins, minimum surface temperature of the heat dissipation fins, average surface temperature of the heat dissipation fins, air outlet temperature, and heat dissipation efficiency, respectively.

[0087] The comprehensive loss function was determined to be the weighted sum of the physical model residuals and the data fitting error (weight coefficient α=0.7). The Adam optimizer was used to adjust the model parameters, and the initial learning rate was set to 0.001. The training process is as follows: One hundred sets of simulation data under different operating conditions (obtained by changing parameters such as fan speed and heat sink fin size) were collected to train the PINN model, and the data were processed using min-max normalization. The maximum number of iterations was set to 10,000, the batch size to 32, and the learning rate was reduced to 0.9 times the original value every 1,000 iterations. After 5,000 iterations, the model's root mean square error (RMSE) decreased to below 0.05, and the coefficient of determination R0 was [not specified]. 2 The accuracy reaches 0.98, meeting the precision requirements.

[0088] S4. Input the operating condition data of the heat dissipation component to be tested into the heat dissipation performance prediction model to obtain heat dissipation performance prediction data; select the optimal heat dissipation component by comprehensively considering heat dissipation performance, cost and space constraints.

[0089] Suppose there are three different types of fans and four different types of heat sink fins to choose from. Input their parameters into a heat dissipation performance prediction model to obtain the predicted heat dissipation performance under different combinations. For example, fan A has 9 blades and a speed range of 2500-3500 rpm. When combined with heat sink fin C (fin height 25 mm, fin spacing 2.5 mm), the predicted heat dissipation efficiency is 85%, and the highest surface temperature of the heat sink fins is 60℃. Fan B has 6 blades and a speed range of 3000-4000 rpm. When combined with heat sink fin D (fin height 18 mm, fin spacing 3.5 mm), the predicted heat dissipation efficiency is 80%, and the highest surface temperature of the heat sink fins is 68℃, etc.

[0090] Taking into account factors such as heat dissipation performance, cost, and internal space constraints of the electronic device, the combination of fan A and heat sink C was ultimately selected as the optimal solution for the heat dissipation system of the electronic device. This embodiment also includes a heat dissipation component selection system based on 3D simulation and PINN collaborative optimization, used to implement the above method, such as... Figure 1 As shown, the system includes: Simulation Unit 1 constructs a three-dimensional model of a heat dissipation system including heat dissipation components, and simulates the three-dimensional model according to different operating conditions of the heat dissipation components; Data acquisition and processing unit 2 acquires operating condition and simulation output data to establish sample data; Neural network unit 3 constructs a physical information neural network model, trains the model based on sample data, and obtains a heat dissipation performance prediction model after training. Selection Unit 4 deploys a heat dissipation performance prediction model, inputs the operating condition data of the heat dissipation component under test to obtain heat dissipation performance prediction data, and selects the optimal heat dissipation component by comprehensively considering heat dissipation performance, cost and space constraints.

[0091] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

[0092] Although this paper uses terms such as simulation unit, data acquisition and processing unit, neural network unit, and selection unit frequently, the possibility of using other terms is not excluded. These terms are used merely for the convenience of describing and explaining the essence of this invention; interpreting them as any additional limitation would contradict the spirit of this invention.

Claims

1. A method for selecting heat dissipation components based on 3D simulation and PINN collaborative optimization, characterized in that, Includes the following steps: Construct a 3D model of the heat dissipation system including heat dissipation components, and simulate the 3D model according to different operating conditions of the heat dissipation components; Acquire operating conditions and simulation output data to establish sample data; A physical information neural network model is constructed, and the model is trained based on sample data to obtain a heat dissipation performance prediction model. Input the operating condition data of the heat dissipation component under test into the heat dissipation performance prediction model to obtain heat dissipation performance prediction data; The optimal heat dissipation component was selected by considering factors such as overall heat dissipation performance, cost, and space constraints.

2. The heat dissipation component selection method based on three-dimensional simulation and PINN collaborative optimization according to claim 1, characterized in that, The steps for constructing a three-dimensional model of a heat dissipation system including heat dissipation components include: Based on the structural dimensions of the heat dissipation system, a three-dimensional model is constructed, including the heat source, heat dissipation components, and air domain. The heat dissipation components include the fan and heat sink fins. Set simulation parameters and boundary conditions; Establish the three-dimensional simulation control equations.

3. The heat dissipation component selection method based on three-dimensional simulation and PINN collaborative optimization according to claim 2, characterized in that, The three-dimensional simulation control equations include: The continuity equation states that the partial derivative of air density with respect to time plus the divergence of the product of air density and velocity vector equals 0. The momentum equation states that the product of air density and the first element is equal to the sum of the negative pressure gradient and the divergence of the second element, as well as the product of air density and the vector product of gravitational acceleration. The first unit is the partial derivative of the velocity vector with respect to time plus the product of the velocity vector and the gradient of the velocity vector. The second unit is the sum of the velocity vector gradient and the transpose of the velocity vector gradient multiplied by the dynamic viscosity; The energy equation, the product of air density and specific heat capacity in the third unit, equals the sum of divergence and heat source term in the fourth unit; The third unit is the partial derivative of temperature with respect to time plus the dot product of the velocity vector and the temperature gradient. The fourth unit is the product of thermal conductivity and temperature gradient.

4. The heat dissipation component selection method based on three-dimensional simulation and PINN collaborative optimization according to claim 3, characterized in that, The simulation of the 3D model based on different operating conditions of the heat dissipation components includes: The finite volume method was used to discretize the three-dimensional simulation control equations. A first-order implicit scheme was used for the time term, a second-order upwind scheme for the convection term, and a central difference scheme for the diffusion term. The pressure-velocity coupling was performed using the SIMPLE algorithm. The iterative convergence criterion was that all residuals were less than 1*102. -6 .

5. The heat dissipation component selection method based on three-dimensional simulation and PINN co-optimization as described in claim 2, 3, or 4, characterized in that: Obtain the operating conditions of the heat dissipation components, including fan speed, number of blades, fin height, fin spacing, and fin thickness; Based on the simulation results, the performance parameters corresponding to the working conditions of the heat dissipation component are obtained. The performance parameters include the highest surface temperature of the heat dissipation fins, the lowest surface temperature of the heat dissipation fins, the average surface temperature of the heat dissipation fins, the air outlet temperature, and the heat dissipation efficiency. The operating conditions and performance parameters of the heat dissipation components are preprocessed to generate sample data.

6. The heat dissipation component selection method based on three-dimensional simulation and PINN collaborative optimization according to claim 5, characterized in that, Physical information neural network models include: The input layer contains 8 neurons, which correspond to the operating conditions of the heat sink components and air parameters, respectively. The hidden layers consist of three layers, each containing 32, 64, and 32 neurons respectively, all using the ReLU activation function, and all layers are fully connected. The output layer contains 5 neurons, and the outputs correspond to the performance parameters. Set the model's overall loss function.

7. The heat dissipation component selection method based on three-dimensional simulation and PINN collaborative optimization according to claim 6, characterized in that: The overall loss function includes data loss and physical loss, and is a weighted sum of data loss and physical loss; The data loss is calculated using mean squared error, and the physical loss is obtained by inputting the prediction results of the physical information neural network model into the control equation to calculate the residual.

8. The heat dissipation component selection method based on three-dimensional simulation and PINN collaborative optimization according to claim 7, characterized in that, The process of training the model based on evolutionary data to obtain a heat dissipation performance prediction model includes: The sample data is divided into training and testing sets. The physical information neural network model is trained and tested. A comprehensive loss function is used, and backpropagation is performed through the Adam optimizer to iteratively optimize the model parameters, minimize the comprehensive loss function, and obtain the heat dissipation performance prediction model after training and passing the test.

9. The heat dissipation component selection method based on three-dimensional simulation and PINN collaborative optimization according to claim 8, characterized in that: By inputting different operating conditions of the heat dissipation components into the heat dissipation performance prediction model, the corresponding heat dissipation performance prediction data of the heat dissipation components can be obtained. Based on performance prediction data, heat dissipation components with lower cost and smaller footprint are selected, provided that heat dissipation requirements are met.

10. A heat dissipation component selection system based on three-dimensional simulation and PINN collaborative optimization, implementing the method described in any one of claims 1-9, characterized in that, include: The simulation unit constructs a three-dimensional model of the heat dissipation system, including heat dissipation components, and simulates the three-dimensional model according to different operating conditions of the heat dissipation components. The data acquisition and processing unit acquires operating condition and simulation output data to establish sample data. The neural network unit constructs a physical information neural network model, trains the model based on sample data, and obtains a heat dissipation performance prediction model after training. The selection unit deploys a heat dissipation performance prediction model, inputs the operating condition data of the heat dissipation component under test to obtain heat dissipation performance prediction data, and selects the optimal heat dissipation component by comprehensively considering heat dissipation performance, cost and space constraints.

Images