Axial flux motor temperature prediction method based on equivalent thermal circuit method calculation

Abstract

The invention discloses an axial flux motor temperature prediction method based on equivalent thermal circuit method calculation, and the method comprises the following steps: 1, motor thermal circuit unit division: decomposing a motor into a stator unit, a rotor unit, an air gap unit and a housing unit, and constructing an equivalent thermal network through a node-branch method; 2, calculating a multi-physics coupling heat source; 3, thermal resistance parameter dynamic calculation and CFD coupling optimization: calculating conduction thermal resistance, convection thermal resistance and contact thermal resistance; and 4, parameter self-calibration based on machine learning: constructing a PINN model according to a physical law and known physical parameters, introducing a Bayesian optimization algorithm on the basis of the PINN model, adjusting and optimizing the parameters of the PINN model by using Bayesian optimization, designing a reinforcement learning algorithm, and carrying out self-calibration on the parameters of the PINN model. Enabling the reinforcement learning algorithm to dynamically adjust parameters according to the real-time operation data of the system and the prediction result of the PINN model; 5, constructing and solving a heat balance equation; and step 6, visualizing and verifying the temperature field.

Description

Technical Field

[0001] The present invention relates to the technical field of axial flux motor temperature prediction, and in particular to a method for predicting the temperature of an axial flux motor based on an equivalent thermal circuit method. Background Art

[0002] Axial flux motors (AFMs) have attracted widespread attention and application in fields such as electric vehicles, wind power generation, and aerospace due to their high power density, small size, light weight, compact structure, and excellent dynamic performance. As a motor's operating time increases, its internal temperature continues to rise. Excessively high temperatures can have a serious impact on the motor's operating efficiency, service life, and even safety. Accurately predicting and evaluating a motor's temperature distribution and temperature rise is crucial for ensuring stable operation and extending its service life. The equivalent thermal circuit method, a simplified thermal analysis method, has been widely used in engineering practice. It simplifies the complex motor model into a lumped parameter thermal circuit model, utilizes the principle of thermoelectric analogy, and employs formulas from circuit theory to solve the thermal problem, thereby predicting and evaluating the motor's temperature.

[0003] Contact thermal resistance is another important parameter in motor thermal analysis. It reflects the thermal resistance generated by poor contact between different components. In practical engineering applications, contact thermal resistance calibration typically requires experimental data. This calibration process can require extensive experimental equipment and manpower, resulting in high project implementation costs. Therefore, a temperature prediction method for axial flux motors based on the equivalent thermal circuit method is proposed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a temperature prediction method for an axial flux motor based on an equivalent thermal circuit method.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A method for predicting the temperature of an axial flux motor based on an equivalent thermal circuit method includes the following steps: Step 1: Motor thermal circuit unit division: Based on the axial flux motor topology, the motor is decomposed into stator unit, rotor unit, air gap unit, and housing unit. The node-branch method is used to construct an equivalent thermal network. Each node represents a temperature point, and each branch represents thermal resistance and heat capacity, forming a complete thermal circuit model. Step 2: Multi-physics coupled heat source calculation: Calculate the copper loss of the stator winding, the iron loss of the stator / rotor core, and the eddy current loss of the permanent magnets. Map the electromagnetic losses to the corresponding thermal circuit nodes according to their spatial distribution to establish a heat source matrix. Step 3: Dynamic calculation of thermal resistance parameters and CFD coupling optimization: Calculate the conduction thermal resistance (between components), convection thermal resistance (air gap / housing), and contact thermal resistance (assembly interface) based on the material thermal conductivity and component size. Introduce a temperature correction factor to reflect the temperature dependence of the thermal conductivity. Combined with CFD computational fluid dynamics methods, optimize the convection thermal resistance model. Calculate the convection thermal resistance based on the convection heat transfer coefficient and heat dissipation area. The convection heat transfer coefficient is dynamically calculated based on forced air cooling or natural cooling conditions and verified and optimized through CFD simulation. Calibrate the contact admittance coefficient through pressure-contact admittance experiments, and then calculate the contact thermal resistance. Step 4: Parameter self-calibration based on machine learning: Based on the laws of physics and known physical parameters, a PINN model is constructed. Based on the PINN model, a Bayesian optimization algorithm is introduced to tune the parameters of the PINN model using Bayesian optimization. A reinforcement learning algorithm is designed so that the reinforcement learning algorithm dynamically adjusts the parameters based on the system's real-time operating data and the PINN model's prediction results. Step 5: Construct and solve the heat balance equation: Based on the principle of energy conservation, construct the heat balance equation for each node, including the heat source term, thermal resistance term, and heat capacity term. Use the implicit Euler method to discretize the equation, construct a sparse matrix, and use a high-performance computing platform for parallel solution to improve computational efficiency. Step 6: Temperature Field Visualization and Verification: Generate a 3D temperature cloud map and mark the hotspot locations to facilitate intuitive analysis of the motor temperature distribution. Verify the accuracy of the prediction results through infrared thermal imaging, embedded thermocouples, and fiber optic temperature measurement. At the same time, use CFD simulation results for auxiliary verification to ensure the reliability of the prediction results.

[0006] The above further includes: Furthermore, the stator unit includes a winding, an iron core, and an insulating layer, and the corresponding thermal characteristic parameters of the stator unit include copper loss, iron loss, and contact thermal resistance; the rotor unit includes a permanent magnet, a rotor yoke, and a bearing; the corresponding thermal characteristic parameters of the rotor unit include eddy current loss and conduction thermal resistance; the air gap unit includes air / cooling medium; the corresponding thermal characteristic parameters of the air gap unit include convection thermal resistance and radiation thermal resistance; the shell unit includes a casing and heat dissipation fins; the corresponding thermal characteristic parameters of the shell unit include forced convection coefficient and surface radiation coefficient.

[0007] Furthermore, the calculation formula of the copper loss of the stator winding is expressed as ,in, , considering the temperature-resistivity coupling effect, the calculation formula of the stator / rotor core iron loss is expressed as , using the improved Steinmetz model, the coefficients are calibrated by measured data, and the calculation formula of the eddy current loss of the permanent magnet is expressed as ,in is the number of permanent magnet segments, The conductivity is the electromagnetic loss, which is mapped to the corresponding heat path node according to the spatial distribution, and the heat source matrix is established. .

[0008] Furthermore, the calculation formula of the thermal conductivity resistance (between components) is expressed as ,in, , introduce the temperature correction factor of material thermal conductivity .

[0009] Furthermore, the calculation formula of the convection thermal resistance (air gap / housing) is expressed as ,in, is the convective heat transfer coefficient (W / m²·K), which reflects the efficiency of heat exchange between the fluid and the solid surface. is the heat dissipation area (m²), that is, the area of contact between the fluid and the solid surface, and the convection heat transfer coefficient Depending on the cooling conditions (such as forced air cooling or natural cooling) and the characteristics of the fluid and solid surface (such as flow rate, temperature, surface roughness, etc.), we need to dynamically calculate according to the specific cooling conditions , forced air cooling , Reynolds number Real-time calculation. After calculating the convection thermal resistance, CFD simulation is used to verify and optimize the model. CFD simulation simulates the flow and heat transfer process of the fluid on the heat dissipation surface, thereby providing an accurate value of the convection heat transfer coefficient. CFD simulation steps: Establishing geometric model: According to the actual size and shape of the motor, establish geometric model in CFD software; Meshing: Meshing the geometric model to generate the computational mesh for simulation; Set boundary conditions: Set boundary conditions based on cooling conditions (such as the speed and temperature of forced air cooling); Run simulation: Start CFD simulation to simulate the flow and heat transfer process of fluid on the heat dissipation surface; Result analysis: Extract the convective heat transfer coefficient from the simulation results and compare it with the results previously calculated using the empirical formula. If there are any differences, adjust the constants in the empirical formula or improve the geometric model, and then rerun the simulation for optimization.

[0010] Furthermore, the calculation formula of the contact thermal resistance (assembly interface) is expressed as , through the pressure-contact admittance test standard .

[0011] Furthermore, the specific steps of the parameter self-calibration based on machine learning are: Data preparation and preprocessing: Collect historical data and real-time operation data, including sensor data such as temperature, pressure, flow, and related physical parameters, and clean, denoise, and normalize the data to ensure data quality and consistency; Constructing PINN model: According to the physical laws and known physical parameters, for heat conduction, Fourier's heat conduction law is used, which is expressed as ,in, is the heat flux density, is the thermal conductivity, is the temperature gradient; for thermal convection, Newton's cooling law is used, which is expressed as ,in, is the heat dissipation power, is the convective heat transfer coefficient, is the heat dissipation area, is the motor surface temperature, The ambient temperature is used to construct a PINN model. The input is the motor's operating parameters (such as current, voltage, and speed), and the output is the motor's temperature distribution or the temperature at a specific point. When constructing the model, the laws of physics are embedded as constraints into the neural network's loss function. Bayesian optimization: Based on the PINN model, the Bayesian optimization algorithm is introduced. The steps of Bayesian optimization are: defining the objective function, which is the prediction error of the PINN model; setting the prior function, which is constructed based on Gaussian process regression (GPR), which represents the prediction of the objective function without any observation data. The formula of the Gaussian process regression is ,in, is the mean function, is the covariance function (also called the kernel function); select the acquisition function, which is used to determine the location of the next evaluation point. It balances the relationship between exploration (finding new possibilities) and utilization (sampling in known high-performance areas). The acquisition functions include UCB (Upper Confidence Bound), PI (Probability of Improvement), and EI (Expected Improvement). The formula for UCB is ,in, is the predicted mean, is the forecast standard deviation, is the adjustment parameter, the formula of PI is ,in, is the currently known maximum value. is the adjustment parameter, is the normal cumulative distribution function, and the formula for EI is ,in, , is the normal cumulative distribution function, is a normal probability density function; in each iteration, the acquisition function is used to select the next evaluation point, and the value of the objective function at that point is calculated. The posterior distribution of the Gaussian process regression model is updated to include the new observation data, and this process is repeated until the predetermined number of iterations is reached or a stopping condition is met; Introducing reinforcement learning: Based on Bayesian optimization and the PINN model, the Q-Network (DQN) algorithm is used to design a reinforcement learning controller for adjusting the thermal resistance parameters of an axial flux motor. The state space is defined: the state space includes parameters such as the motor's temperature distribution, current, and voltage, and some key parameters are selected as the dimensions of the state space; the action space is defined: the action space is the adjustment range of the thermal resistance parameter, which is divided into several discrete values, and one of the values is selected as the action; constructing a neural network: a DQN neural network is constructed, whose input is the state of the environment (i.e., parameters such as the motor's temperature distribution, current, and voltage), and whose output is the Q value of each possible action; training the neural network: historical data and the prediction results of the PINN model are used to train the DQN neural network. During the training process, the intelligent agent will select an action (i.e., adjust the thermal resistance parameter) according to the current strategy and observe the environmental state and reward (i.e., prediction error). This data is then used to update the weights of the neural network, achieving closed-loop optimization: at each time step, the agent uses the PINN model to predict the temperature distribution of the motor and selects the next action based on the prediction (i.e., adjusting the thermal resistance parameter). The new state is then input into the DQN neural network to obtain the Q value for the next action. This process continues until a predetermined number of iterations is reached or a stopping condition is met.

[0012] System deployment and verification: Integrate the trained PINN model, Bayesian optimization algorithm, and reinforcement learning algorithm into the parameter self-calibration system. Deploy the parameter self-calibration system in the actual system, and conduct verification and testing. Based on the test results, make necessary adjustments and optimizations to the system.

[0013] Furthermore, the node heat balance equation is expressed as , is the node heat capacity, is the inter-node thermal resistance, As the heat source, the implicit Euler method is used to discretize the equation and construct a sparse matrix: ,in, is the thermal conductivity matrix, and the PETSc library is used for parallel solution to accelerate the calculation.

[0014] The present invention has the following beneficial effects: 8. In the present invention, by combining physical laws and known physical parameters to construct a PINN model, it can more accurately reflect the physical phenomena in motor thermal analysis, improve the accuracy and reliability of the prediction results, introduce a Bayesian optimization algorithm to tune the parameters of the PINN model, and efficiently search for the optimal parameter combination, further improving the prediction performance of the model. A reinforcement learning algorithm is designed to dynamically adjust the parameters based on the system's real-time operating data and the prediction results of the PINN model. This dynamic adjustment mechanism enables parameter self-calibration to adapt to different working environments and conditions, enhancing the adaptability and robustness of parameter self-calibration. Parameter self-calibration based on machine learning realizes automatic parameter calibration, reduces manual intervention and the number of experiments, and thus reduces the cost of parameter calibration. This has important economic significance for the field of motor thermal analysis, which requires a large amount of experimental data and manpower investment.

[0015] 9. In this invention, by combining CFD with thermal analysis, the turbulent flow state in the motor air gap can be more accurately simulated, thereby optimizing the convective thermal resistance model. This coupled calculation provides more detailed flow information, helps reduce deviations in the convective thermal resistance calculation, and improves the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a step diagram of a temperature prediction method for an axial flux motor based on an equivalent thermal circuit method proposed by the present invention; Figure 2 Schematic diagram of the equivalent thermal network in the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0018] See also Figure 1 As shown, the present invention is a method for predicting the temperature of an axial flux motor based on the equivalent thermal circuit method, comprising the following steps: Step 1: Motor thermal circuit unit division: Based on the axial flux motor topology, the motor is decomposed into stator unit, rotor unit, air gap unit, and housing unit. The node-branch method is used to construct an equivalent thermal network. Each node represents a temperature point, and each branch represents thermal resistance and heat capacity, forming a complete thermal circuit model. Step 2: Multi-physics coupled heat source calculation: Calculate the copper loss of the stator winding, the iron loss of the stator / rotor core, and the eddy current loss of the permanent magnets. Map the electromagnetic losses to the corresponding thermal circuit nodes according to their spatial distribution to establish a heat source matrix. Step 3: Dynamic calculation of thermal resistance parameters and CFD coupling optimization: Calculate the conduction thermal resistance (between components), convection thermal resistance (air gap / housing), and contact thermal resistance (assembly interface) based on the material thermal conductivity and component size. Introduce a temperature correction factor to reflect the temperature dependence of the thermal conductivity. Combined with CFD computational fluid dynamics methods, optimize the convection thermal resistance model. Calculate the convection thermal resistance based on the convection heat transfer coefficient and heat dissipation area. The convection heat transfer coefficient is dynamically calculated based on forced air cooling or natural cooling conditions and verified and optimized through CFD simulation. Calibrate the contact admittance coefficient through pressure-contact admittance experiments, and then calculate the contact thermal resistance. Step 4: Parameter self-calibration based on machine learning: Based on the laws of physics and known physical parameters, a PINN model is constructed. Based on the PINN model, a Bayesian optimization algorithm is introduced to tune the parameters of the PINN model using Bayesian optimization. A reinforcement learning algorithm is designed so that the reinforcement learning algorithm dynamically adjusts the parameters based on the system's real-time operating data and the PINN model's prediction results. Step 5: Construct and solve the heat balance equation: Based on the principle of energy conservation, construct the heat balance equation for each node, including the heat source term, thermal resistance term, and heat capacity term. Use the implicit Euler method to discretize the equation, construct a sparse matrix, and use a high-performance computing platform for parallel solution to improve computational efficiency. Step 6: Temperature Field Visualization and Verification: Generate a 3D temperature cloud map and mark the hotspot locations to facilitate intuitive analysis of the motor temperature distribution. Verify the accuracy of the prediction results through infrared thermal imaging, embedded thermocouples, and fiber optic temperature measurement. At the same time, use CFD simulation results for auxiliary verification to ensure the reliability of the prediction results.

[0019] In one embodiment, the stator unit includes a winding, an iron core, and an insulating layer, and the corresponding thermal characteristic parameters of the stator unit include copper loss, iron loss, and contact thermal resistance. The rotor unit includes a permanent magnet, a rotor yoke, and a bearing. The corresponding thermal characteristic parameters of the rotor unit include eddy current loss and conduction thermal resistance. The air gap unit includes air / cooling medium. The corresponding thermal characteristic parameters of the air gap unit include convection thermal resistance and radiation thermal resistance. The shell unit includes a casing and heat dissipation fins. The corresponding thermal characteristic parameters of the shell unit include forced convection coefficient and surface radiation coefficient.

[0020] In one embodiment, the calculation formula of the copper loss of the stator winding is expressed as: ,in, , considering the temperature-resistivity coupling effect, the calculation formula of the stator / rotor core iron loss is expressed as , using the improved Steinmetz model, the coefficients are calibrated by measured data, and the calculation formula of the eddy current loss of the permanent magnet is expressed as ,in is the number of permanent magnet segments, The conductivity is the electromagnetic loss, which is mapped to the corresponding heat path node according to the spatial distribution, and the heat source matrix is established. .

[0021] In one embodiment, the calculation formula of the conductive thermal resistance (between components) is expressed as: ,in, , introduce the temperature correction factor of material thermal conductivity .

[0022] In one embodiment, the calculation formula of the convection thermal resistance (air gap / housing) is expressed as: ,in, is the convective heat transfer coefficient (W / m²·K), which reflects the efficiency of heat exchange between the fluid and the solid surface. is the heat dissipation area (m²), that is, the area of contact between the fluid and the solid surface, and the convection heat transfer coefficient Depending on the cooling conditions (such as forced air cooling or natural cooling) and the characteristics of the fluid and solid surface (such as flow rate, temperature, surface roughness, etc.), we need to dynamically calculate according to the specific cooling conditions , forced air cooling , Reynolds number Real-time calculation. After calculating the convection thermal resistance, CFD simulation is used to verify and optimize the model. CFD simulation simulates the flow and heat transfer process of the fluid on the heat dissipation surface, thereby providing an accurate value of the convection heat transfer coefficient. CFD simulation steps: Establishing geometric model: According to the actual size and shape of the motor, establish geometric model in CFD software; Meshing: Meshing the geometric model to generate the computational mesh for simulation; Set boundary conditions: Set boundary conditions based on cooling conditions (such as the speed and temperature of forced air cooling); Run simulation: Start CFD simulation to simulate the flow and heat transfer process of fluid on the heat dissipation surface; Result analysis: Extract the convective heat transfer coefficient from the simulation results and compare it with the results previously calculated using the empirical formula. If there are any differences, adjust the constants in the empirical formula or improve the geometric model, and then rerun the simulation for optimization.

[0023] In one embodiment, the calculation formula of the contact thermal resistance (assembly interface) is expressed as: , through the pressure-contact admittance test standard .

[0024] In one embodiment, the specific steps of the parameter self-calibration based on machine learning are: Data preparation and preprocessing: Collect historical data and real-time operation data, including sensor data such as temperature, pressure, flow, and related physical parameters, and clean, denoise, and normalize the data to ensure data quality and consistency; Constructing PINN model: According to the physical laws and known physical parameters, for heat conduction, Fourier's heat conduction law is used, which is expressed as ,in, is the heat flux density, is the thermal conductivity, is the temperature gradient; for thermal convection, Newton's cooling law is used, which is expressed as ,in, is the heat dissipation power, is the convective heat transfer coefficient, is the heat dissipation area, is the motor surface temperature, The ambient temperature is used to construct a PINN model. The input is the motor's operating parameters (such as current, voltage, and speed), and the output is the motor's temperature distribution or the temperature at a specific point. When constructing the model, the laws of physics are embedded as constraints into the neural network's loss function. Bayesian optimization: Based on the PINN model, the Bayesian optimization algorithm is introduced. The steps of Bayesian optimization are: defining the objective function, which is the prediction error of the PINN model; setting the prior function, which is constructed based on Gaussian process regression (GPR), which represents the prediction of the objective function without any observation data. The formula of the Gaussian process regression is ,in, is the mean function, is the covariance function (also called the kernel function); select the acquisition function, which is used to determine the location of the next evaluation point. It balances the relationship between exploration (finding new possibilities) and utilization (sampling in known high-performance areas). The acquisition functions include UCB (Upper Confidence Bound), PI (Probability of Improvement), and EI (Expected Improvement). The formula for UCB is ,in, is the predicted mean, is the forecast standard deviation, is the adjustment parameter, the formula of PI is ,in, is the currently known maximum value. is the adjustment parameter, is the normal cumulative distribution function, and the formula for EI is ,in, , is the normal cumulative distribution function, is a normal probability density function; in each iteration, the acquisition function is used to select the next evaluation point, and the value of the objective function at that point is calculated. The posterior distribution of the Gaussian process regression model is updated to include the new observation data, and this process is repeated until the predetermined number of iterations is reached or a stopping condition is met; Introducing reinforcement learning: Based on Bayesian optimization and the PINN model, the Q-Network (DQN) algorithm is used to design a reinforcement learning controller for adjusting the thermal resistance parameters of an axial flux motor. The state space is defined: the state space includes parameters such as the motor's temperature distribution, current, and voltage, and some key parameters are selected as the dimensions of the state space; the action space is defined: the action space is the adjustment range of the thermal resistance parameter, which is divided into several discrete values, and one of the values is selected as the action; constructing a neural network: a DQN neural network is constructed, whose input is the state of the environment (i.e., parameters such as the motor's temperature distribution, current, and voltage), and whose output is the Q value of each possible action; training the neural network: historical data and the prediction results of the PINN model are used to train the DQN neural network. During the training process, the intelligent agent will select an action (i.e., adjust the thermal resistance parameter) according to the current strategy and observe the environmental state and reward (i.e., prediction error). This data is then used to update the weights of the neural network, achieving closed-loop optimization: at each time step, the agent uses the PINN model to predict the temperature distribution of the motor and selects the next action based on the prediction (i.e., adjusting the thermal resistance parameter). The new state is then input into the DQN neural network to obtain the Q value for the next action. This process continues until a predetermined number of iterations is reached or a stopping condition is met.

[0025] System deployment and verification: Integrate the trained PINN model, Bayesian optimization algorithm, and reinforcement learning algorithm into the parameter self-calibration system. Deploy the parameter self-calibration system in the actual system, and conduct verification and testing. Based on the test results, make necessary adjustments and optimizations to the system.

[0026] In one embodiment, the node heat balance equation is expressed as , is the node heat capacity, is the inter-node thermal resistance, As the heat source, the implicit Euler method is used to discretize the equation and construct a sparse matrix: ,in, is the thermal conductivity matrix, and the PETSc library is used for parallel solution to accelerate the calculation.

[0027] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A temperature prediction method for an axial flux motor based on an equivalent thermal circuit method, characterized in that: The following steps are involved: Step 1: Motor thermal circuit unit division: Based on the axial flux motor topology, the motor is decomposed into stator unit, rotor unit, air gap unit, and housing unit. The node-branch method is used to construct an equivalent thermal network. Each node represents a temperature point, and each branch represents thermal resistance and heat capacity, forming a complete thermal circuit model. Step 2: Multi-physics coupled heat source calculation: Calculate the copper loss of the stator winding, the iron loss of the stator / rotor core, and the eddy current loss of the permanent magnets. Map the electromagnetic losses to the corresponding thermal circuit nodes according to their spatial distribution to establish a heat source matrix. Step 3: Dynamic calculation of thermal resistance parameters and CFD coupling optimization: Calculate the conduction thermal resistance, convection thermal resistance, and contact thermal resistance based on the material thermal conductivity and component dimensions. Introduce a temperature correction factor to reflect the temperature dependence of the thermal conductivity. Combined with CFD computational fluid dynamics methods, optimize the convection thermal resistance model. Calculate the convection thermal resistance based on the convection heat transfer coefficient and heat dissipation area. The convection heat transfer coefficient is dynamically calculated based on forced air cooling or natural cooling conditions and verified and optimized through CFD simulation. Calibrate the contact admittance coefficient through pressure-contact admittance experiments to calculate the contact thermal resistance. Step 4: Parameter self-calibration based on machine learning: Based on the laws of physics and known physical parameters, a PINN model is constructed. Based on the PINN model, a Bayesian optimization algorithm is introduced to tune the parameters of the PINN model using Bayesian optimization. A reinforcement learning algorithm is designed so that the reinforcement learning algorithm dynamically adjusts the parameters based on the system's real-time operating data and the PINN model's prediction results. Step 5: Construct and solve the heat balance equation: Based on the principle of energy conservation, construct the heat balance equation for each node, including the heat source term, thermal resistance term, and heat capacity term. Use the implicit Euler method to discretize the equation, construct a sparse matrix, and solve it in parallel. Step 6: Temperature field visualization and verification: Generate a 3D temperature cloud map, mark the hotspot locations, and verify the accuracy of the prediction results. At the same time, use CFD simulation results for auxiliary verification.

2. The temperature prediction method of an axial flux motor based on the equivalent thermal circuit method according to claim 1, characterized in that: The stator unit includes a winding, an iron core, and an insulating layer. The corresponding thermal characteristic parameters of the stator unit include copper loss, iron loss, and contact thermal resistance. The rotor unit includes a permanent magnet, a rotor yoke, and a bearing. The corresponding thermal characteristic parameters of the rotor unit include eddy current loss and conduction thermal resistance. The air gap unit includes air / cooling medium. The corresponding thermal characteristic parameters of the air gap unit include convection thermal resistance and radiation thermal resistance. The shell unit includes a casing and heat dissipation fins. The corresponding thermal characteristic parameters of the shell unit include forced convection coefficient and surface radiation coefficient.

3. The temperature prediction method of an axial flux motor based on the equivalent thermal circuit method according to claim 1, characterized in that: The calculation formula of the copper loss of the stator winding is expressed as ,in, , considering the temperature-resistivity coupling effect, the calculation formula of the stator / rotor core iron loss is expressed as , the calculation formula of the eddy current loss of the permanent magnet is expressed as ,in is the number of permanent magnet segments, The conductivity is the electromagnetic loss, which is mapped to the corresponding heat path node according to the spatial distribution, and the heat source matrix is established. .

4. The temperature prediction method of an axial flux motor based on the equivalent thermal circuit method according to claim 1, characterized in that: The calculation formula of the thermal conductivity resistance is expressed as ,in, , introduce the temperature correction factor of material thermal conductivity .

5. The temperature prediction method of an axial flux motor based on the equivalent thermal circuit method according to claim 1, characterized in that: The calculation formula of the convection thermal resistance is expressed as ,in, is the convective heat transfer coefficient, is the heat dissipation area. After calculating the convection thermal resistance, CFD simulation is used to verify and optimize the model. CFD simulation simulates the flow and heat transfer process of the fluid on the heat dissipation surface, thereby providing an accurate value of the convection heat transfer coefficient. The CFD simulation steps are as follows: Establishing geometric model: According to the actual size and shape of the motor, establish geometric model in CFD software; Meshing: Meshing the geometric model to generate the computational mesh for simulation; Set boundary conditions: Set boundary conditions according to cooling conditions; Run simulation: Start CFD simulation to simulate the flow and heat transfer process of fluid on the heat dissipation surface; Result analysis: Extract the convective heat transfer coefficient from the simulation results and compare it with the results previously calculated using the empirical formula. If there are any differences, adjust the constants in the empirical formula or improve the geometric model, and then rerun the simulation for optimization.

6. The method for predicting the temperature of an axial flux motor based on the equivalent thermal circuit method according to claim 1, characterized in that: The calculation formula of the contact thermal resistance) is expressed as , through the pressure-contact admittance test standard .

7. The temperature prediction method of an axial flux motor based on the equivalent thermal circuit method according to claim 1, characterized in that: The specific steps of the parameter self-calibration based on machine learning are: Data preparation and preprocessing: Collect historical data and real-time operation data, and clean, denoise, and normalize the data; Constructing PINN model: According to the physical laws and known physical parameters, for heat conduction, Fourier's heat conduction law is used, which is expressed as ,in, is the heat flux density, is the thermal conductivity, is the temperature gradient; for thermal convection, Newton's cooling law is used, which is expressed as ,in, is the heat dissipation power, is the convective heat transfer coefficient, is the heat dissipation area, is the motor surface temperature, The ambient temperature is the ambient temperature. A PINN model is constructed with the motor's operating parameters as input and the motor's temperature distribution or the temperature at a specific point as output. When constructing the model, the laws of physics are embedded as constraints into the neural network's loss function. Bayesian optimization: Based on the PINN model, the Bayesian optimization algorithm is introduced. The steps of Bayesian optimization are as follows: define the objective function, which is the prediction error of the PINN model; set the prior function, which is constructed based on Gaussian process regression and represents the prediction of the objective function in the absence of any observation data; select the acquisition function, which is used to determine the location of the next evaluation point. The acquisition functions include UCB, PI, and EI; in each iteration, the acquisition function is used to select the next evaluation point and calculate the value of the objective function at that point. The posterior distribution of the Gaussian process regression model is updated to include the new observation data. This process is repeated until the predetermined number of iterations is reached or a stopping condition is met. Introducing reinforcement learning: Based on Bayesian optimization and the PINN model, a Q-Network algorithm is used to design a reinforcement learning controller for adjusting the thermal resistance parameters of the axial flux motor. System deployment and verification: Integrate the trained PINN model, Bayesian optimization algorithm, and reinforcement learning algorithm into the parameter self-calibration system, deploy the parameter self-calibration system in the actual system, and conduct verification and testing. Based on the test results, adjust and optimize the system.

8. The temperature prediction method of an axial flux motor based on the equivalent thermal circuit method according to claim 1, characterized in that: The node heat balance equation is expressed as , is the node heat capacity, is the inter-node thermal resistance, As the heat source, the implicit Euler method is used to discretize the equation and construct a sparse matrix: ,in, is the thermal conductivity matrix, and the PETSc library is used for parallel solution to accelerate the calculation.

Images