A quick electromagnetic-thermal coupling optimization design method for permanent magnet motor considering demagnetization

Abstract

This invention discloses a rapid electromagnetic-thermal coupling optimization method for permanent magnet motors considering demagnetization. Specifically, it involves finding an initial topology for the permanent magnet motor that satisfies the temperature rise condition after electromagnetic-thermal coupling calculations, and obtaining the average temperature rise of each component of the motor. This temperature rise is then fixed in the parametric electromagnetic model of the motor, and the motor efficiency is calculated using LHS sampling and finite element method. or With the degree of demagnetization D m Furthermore, a demagnetization penalty term is introduced into the efficiency calculation. D m Mapped to a 01 array, denoted as D m_01 Using efficiency data and demagnetization data D m_01 The Kriging response surface and the decision tree response surface were trained separately. Using demagnetization as a constraint, the motor efficiency was optimized and validated using NAGA-II. If the accuracy of the efficiency response surface was insufficient, it was trained and optimized again to enhance its predictive ability in the high-efficiency region of the motor. Electromagnetic-thermal coupling calculations were performed using the highest-efficiency topology predicted by the efficiency response surface to obtain the efficiency. or re .like or re If the requirements are met, the optimization ends; if not, a new round of optimization begins, and this cycle continues.

Description

Technical Field

[0001] This invention belongs to the field of motor technology, and in particular relates to a fast electromagnetic thermal coupling optimization design method for permanent magnet motors that takes into account demagnetization. Technical Background

[0002] Permanent magnet motors have been widely used in various industries due to their excellent performance and high energy density. However, the frequent problems such as overheating and demagnetization that occur during application indicate that multiphysics-based coupled optimization design has become a necessary step in permanent magnet motor design. However, traditional multiphysics optimization is too time-consuming, and there is an urgent need for a universal and rapid optimization design method.

[0003] Currently, the mainstream methods for solving the electromagnetic and thermal fields of electric motors are broadly divided into the finite element method (FEM) and the equivalent network method. The FEM's advantage lies in its versatility; accurate simulation results can be obtained using FEM software. However, the solution time is quite long, especially for certain motors that require solving three-dimensional models, considering PWM harmonic losses, or simulating the motor startup process. The excessively long FEM simulation time is unacceptable for motor optimization, making optimization design extremely difficult. On the other hand, the equivalent network method, due to its lack of versatility, has a high application threshold, requiring significant time and experience to build a network model for a specific motor, making widespread application difficult.

[0004] The application of surrogate models is becoming increasingly widespread. Their combination with motor simulation models has greatly accelerated the process of optimizing motor size parameters. Surrogate models only need to learn from a small amount of sample data to efficiently capture the complex mapping relationship between motor size design parameters and motor performance, reducing the computational cost required for optimization. Surrogate models based on Kriging response surfaces have a simple and clear structure, making them easy to combine with other optimization models. Their high-precision predictive ability and accurate estimation of output response variance make them a commonly used tool for motor size optimization design.

[0005] In the field of motor design, 0-1 decision models, by discretizing the output response into 0 or 1, are often used to handle classification problems with constraints, such as whether a motor is demagnetizing. XGBoost (Extreme Gradient Boosting) and LightGBM (Light Gradient Boosting Machine) are two widely used decision tree-based boosting methods. The core idea of ​​the XGBoost model is to continuously build new trees to correct the errors of previous models and use the second derivative for optimization, thereby improving the model's fitting ability. LightGBM, on the other hand, optimizes the splitting process of the decision tree using a histogram-based method, greatly reducing memory consumption and computational costs. XGBoost and LightGBM, as 0-1 decision models, have broad application prospects in the field of motor design and can efficiently solve classification problems.

[0006] In summary, considering the multiphysics conditions, in order to accelerate the optimization process of permanent magnet motors and improve the versatility of the methods, innovative methods are urgently needed to achieve more efficient and accurate optimization design. Summary of the Invention

[0007] Purpose of the invention: In order to solve the problems existing in the prior art, the present invention provides a fast electromagnetic thermal coupling optimization design method for permanent magnet motors that takes into account demagnetization.

[0008] Technical solution: This invention discloses a rapid electromagnetic thermal coupling optimization design method for permanent magnet motors that takes demagnetization into account, specifically as follows:

[0009] Finite element method (FEM) calculations for electromagnetic-thermal coupling (ET-T) were performed on a permanent magnet motor (PMM), denoted as ET01. A PMM topology design satisfying the temperature rise condition was found, and the average temperature rise of each component of the motor was obtained. The temperature rise calculated in ET01 was fixed in the parameterized electromagnetic model of the motor, and the optimization parameters and optimization range were determined.

[0010] Latin hypercube sampling (LHS) is used to obtain a matrix of experimental points, Parameter_X, with a certain number of groups. The k-th experimental point is denoted as x. k1 ,x k2 ,x k3 …x kN The efficiency η(x) of the motor at each experimental point was calculated using finite element simulation. k1 ,x k2 ,x k3 …x kN ) and the degree of demagnetization of the magnet D m A penalty term D is introduced into the motor efficiency calculation. m This makes the efficiency η train (Dm ,x k1 ,x k2 ,x k3 …x kN Then, determine the degree of motor demagnetization D. m Mapped to a 01 array, denoted as D m_01 .

[0011] The Kriging response surface was trained using motor efficiency data and a normalized experimental point matrix. Demagnetization data of the magnets were used. m_01 The decision tree response surface XGB and LGBM are trained with a normalized experimental point matrix, and the one with higher prediction accuracy is selected as the response surface for demagnetization data prediction. The efficiency output by the Kriging response surface and the demagnetization status output by the decision tree response surface are used instead of finite element calculations. With the degree of demagnetization as a constraint, the motor efficiency η is optimized using a non-dominated genetic algorithm with an elitist strategy (NSGA-II).

[0012] After optimization, the P individuals with the highest efficiency are selected for finite element analysis (FEM) to verify the Kriging response surface prediction accuracy. If the accuracy is insufficient, the P individuals from the FEM calculation are merged with the original experimental points, and the efficiency response surface and demagnetization response surface are retrained and optimized again. This process is repeated to strengthen the prediction capabilities of the Kriging response surface and decision tree response surface in the high-efficiency region of the motor, until the efficiency response surface meets the prediction accuracy requirements. Finally, a motor topology with the highest efficiency value output by the efficiency response surface that does not demagnetize is obtained, denoted as η. max Finite element electromagnetic-thermal coupling calculations were performed on the motor topology. This electromagnetic-thermal coupling was denoted as ET02, and the efficiency η was obtained. re If η re If the optimization efficiency requirement is met, the optimization ends. If the requirement is not met, the temperature of each component of the motor needs to be fixed at the temperature calculated by ET02, and Latin hypercube sampling (LHS) is performed again to redetermine the optimization parameters and optimization range and start optimization again. This process is repeated until the optimization efficiency target is achieved.

[0013] Furthermore, the motor model was parameterized in detail, and a completely new motor topology can be obtained by modifying certain specific parameter values.

[0014] Furthermore, the experimental point matrix Parameter_X obtained by the Latin hypercube sampling method has a dimension of M×N, where M is the number of samples sampled by the Latin hypercube sampling method and N is the number of parameter values ​​contained in each group of experimental points, resulting in the sampling matrix shown below.

[0015]

[0016] Where, x MNThis refers to a parameterized value of the motor, where N is the number of motor topology design parameters, and x... 11 ~x 1N Each set of parameterized values ​​corresponds to one motor design topology. Similarly, there are a total of M sets of parameterized values, corresponding to M motor topology designs.

[0017] Furthermore, the degree of demagnetization D of the magnet. m This refers to the degree of demagnetization at the point where the surface or interior of all magnets in a permanent magnet motor exhibits the greatest degree of demagnetization. The demagnetization degree value can be obtained using the field calculator in finite element software, and it is a continuous value between 0 and 1, defined as follows.

[0018]

[0019] Among them, B r B represents the remanence of the magnet after permanent demagnetization. r0 D represents the remanence of the magnet before permanent demagnetization. m A value of 1 indicates no demagnetization, D m A value of 0 indicates complete demagnetization. If D... m If there is any data that is not 1, then set it to zero, and put the processed data into D. m_01 middle.

[0020] Furthermore, the input power P1 enters the motor, and then the stator winding consumes the power; the copper loss of the stator winding is denoted as p. copper The power consumed by the stator core, and the stator iron loss, are denoted as p. stator The energy is then transferred to the air gap, denoted as electromagnetic power P. EM Then the energy is transferred to the rotor, which consumes a portion of the energy. The rotor iron loss is denoted as p. rotor The rotor cage bar copper loss is denoted as p. bar The eddy current loss of the magnet is denoted as p. PM Finally, energy is transferred to the shaft, minus the wind friction loss p. windloss Thus, the actual output power P2 is obtained. Electromagnetic power P EM As shown below,

[0021]

[0022] Among them, T EM Let n be the electromagnetic torque and n be the motor speed.

[0023] The input power P1 is shown below.

[0024] P1 = P EM +p copper +p stator

[0025] Among them, P EMp is the electromagnetic power. copper For stator winding copper loss, p stator This is recorded as stator iron loss.

[0026] The output power P2 is shown below.

[0027] P2 = P EM -p bar -p rotor -p PM -p windloss

[0028] Among them, P EM p is the electromagnetic power. bar For rotor cage bar copper loss, p rotor For rotor iron loss, p windloss For fan losses, p PM This represents the eddy current loss of the magnet. It should be noted that rotor windage loss and bearing loss are negligible.

[0029] The efficiency η is shown below.

[0030]

[0031] Furthermore, a penalty term D is introduced into the motor efficiency calculation. m The detailed calculation expression is as follows:

[0032] η train =η(x) k1 ,x k2 ,x k3 …x kN )-D coef ×(1-D m )

[0033] Where, η train To provide training data for the Kriging response surface, D coef The demagnetization penalty parameter is designed to reduce the probability of selecting demagnetized individuals when the Kriging response surface predicts high-efficiency individuals.

[0034] Furthermore, a normalized experimental point matrix is ​​used to establish the Kriging response surface model and the decision tree model, and the normalization formula is shown below.

[0035]

[0036] Where, x train x represents the elements in the normalized experimental point matrix. mn x represents the elements in the unnormalized experimental point matrix. max With x min These are the upper and lower limits of a certain parameter during LHS sampling, respectively.

[0037] Furthermore, both the decision tree models XGB and LGBM employ K-folding and random search to train their model parameters. n-folding training involves dividing the dataset into n parts, with parts 1 through n-1 used for training and part n for validation, recording the model accuracy. Then, parts 1 and 3 through n are used for training again, and part 2 for validation, again recording the accuracy. This process continues until the average accuracy of the decision tree model under these model parameters is obtained. The average accuracy under each model parameter generated by random search is then compared, and the highest accuracy is selected for model prediction. Similarly, 5-folding training divides the dataset into 5 parts, with parts 1 through 4 used for training and part 5 for validation, recording the model accuracy. Then, parts 1 and 3 through 5 are used for training again, and part 2 for validation, again recording the accuracy. This process continues until the average accuracy of the decision tree model under these model parameters is obtained. The average accuracy under each model parameter generated by random search is then compared, and the highest accuracy is selected for model prediction, demonstrating good model generalization ability.

[0038] Furthermore, the NSGA-II optimization algorithm is employed to output D using a decision tree response surface. m_01 The output is a constraint condition. The decision tree output value is 0 or 1. The constraint condition is: if the decision tree output is 0, it means that the magnet is demagnetized, and the optimization objective function motor efficiency is set to infinitesimal. If the decision tree output is 1, it means that the magnet is not demagnetized, and the optimization objective function motor efficiency is the Kriging efficiency response surface output.

[0039] Furthermore, after NSGA-II optimization, P individuals with the highest efficiency were selected for finite element verification calculations to calculate the error between the actual and predicted motor efficiency. The calculation formula is shown below.

[0040]

[0041] The actual motor efficiency is η. cal The predicted motor efficiency is η. pre The error is e y If the error is less than p%, the Kriging efficiency response surface prediction accuracy is considered sufficient.

[0042] Further, the permanent magnet motor undergoes electromagnetic-thermal coupling calculations. Specifically, under the motor's operating ambient temperature, the electromagnetic field of the motor is simulated using the finite element method (FEM) to obtain the loss values ​​of each component. These loss values ​​are then converted into heat generation values ​​using the FEM, and the motor's thermal field is simulated to obtain the stable temperatures of each component. Based on these temperatures, the material properties affected by temperature are modified in the electromagnetic field simulation for a second step. The loss values ​​are then converted back into heat generation values, and the motor's thermal field is simulated again to obtain the stable temperatures of each component. These temperatures are compared one by one with the temperatures obtained in the previous step. If the error exceeds 5% (an empirical parameter), another iteration is required. If the accuracy requirements are met, the iteration ends.

[0043] Beneficial effects:

[0044] 1. This invention provides a practical, rapid electromagnetic-thermal coupling optimization design method for permanent magnet motors that considers demagnetization. This method is applicable to optimization scenarios involving permanent magnet motors that require consideration of electromagnetic-thermal coupling calculations (motor application conditions are temperature-sensitive, magnets are prone to demagnetization, or high simulation accuracy is required), whether motor demagnetization occurs, long electromagnetic simulation times, and high motor efficiency requirements. Optimal application scenarios include: joint optimization of the electromagnetic and thermal fields of permanent magnet motors using 3D modeling; joint optimization of the electromagnetic and thermal fields of permanent magnet motors considering PWM harmonic losses; and joint optimization of the electromagnetic and thermal fields of asynchronous permanent magnet synchronous motors considering the startup process. Traditional optimization methods struggle to handle these application scenarios. This method minimizes the number of finite element calculations, significantly accelerates the optimization simulation process, and provides a solution for the aforementioned design scenarios.

[0045] 2. This invention provides a general optimization method that can still achieve optimization even by removing certain conditions in the optimal application scenario, allowing for flexible changes to the application scenario. For example, if the electromagnetic simulation time for a certain permanent magnet motor is long, demagnetization needs to be examined, and high motor efficiency is required, this design method can still be used. Only the electromagnetic thermal coupling process needs to be removed to complete the optimization of this motor.

[0046] 3. This invention employs an adaptive response surface update method to increase the prediction accuracy of the high-efficiency region during motor design. Considering electromagnetic-thermal coupling simulation, it achieves a highly efficient motor topology design while minimizing the possibility of demagnetization in the optimized permanent magnet motor. Individuals meeting all requirements can be selected from the optimized Pareto front for further simulation, avoiding demagnetization verification failures after obtaining highly efficient individuals and significantly accelerating the optimization design process.

[0047] 4. The method of this invention is highly scalable, and the optimization objective can be extended to multiple objectives or other single objectives, exhibiting high flexibility and diverse application scenarios. Two examples are given: ① Single optimization objective: optimizing the peak torque of the permanent magnet motor; ② Multiple optimization objectives: simultaneously optimizing the peak torque, torque ripple, and efficiency of the permanent magnet motor. Attached Figure Description

[0048] Figure 1 This is a flowchart of the present invention.

[0049] Figure 2 A comparison of the topology diagrams of the asynchronous starting permanent magnet synchronous motor before and after optimization is provided for the example.

[0050] Figure 3 The diagram shows a comparison of the demagnetization degree of a permanent magnet synchronous motor before and after asynchronous start-up in the example. Detailed Implementation

[0051] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0052] Please see Figure 1-3 This invention provides a rapid electromagnetic thermal coupling optimization design method for permanent magnet motors that takes demagnetization into account, comprising the following steps:

[0053] S1: Perform finite element electromagnetic thermal coupling calculation on the permanent magnet motor, denoted as ET01; find a permanent magnet motor topology design that meets the temperature rise condition through this calculation, and obtain the average temperature rise of each component of the motor; fix the temperature rise obtained by ET01 in the parameterized electromagnetic model of the motor, and determine the optimization parameters and optimization range.

[0054] S2: The experimental point matrix Parameter_X is obtained by Latin hypercube sampling with a certain number of groups. The k-th experimental point is denoted as x. k1 ,x k2 ,x k3 …x kN The efficiency η(x) of the motor at each experimental point was calculated using finite element simulation. k1 ,x k2 ,x k3 …x kN ) and the degree of demagnetization of the magnet D m ;

[0055] S3: Incorporate the degree of demagnetization D of the magnet into the motor efficiency calculation. m The associated penalty term makes the efficiency η train (D m ,x k1 ,x k2 ,xk3 …x kN Then, determine the degree of motor demagnetization D. m Mapped to a 01 array, denoted as D m_01 The Kriging response surface was trained using motor efficiency data and a normalized experimental point matrix; demagnetization data of magnets were used. m_01 The decision tree response surfaces XGB and LGBM are trained with the normalized experimental point matrix, and the one with higher prediction accuracy is selected as the demagnetization degree prediction response surface.

[0056] S4: The efficiency output by the Kriging response surface and the demagnetization status output by the decision tree response surface are used instead of finite element calculations. With the degree of demagnetization as a constraint, a non-dominated genetic algorithm with an elite strategy is used to optimize the motor efficiency η. The P individuals with the highest efficiency are selected for finite element calculations to verify the prediction accuracy of the Kriging response surface. If the accuracy does not meet the requirements, the individuals that do not demagnetize after finite element calculations are merged with the original experimental points, and the efficiency response surface and demagnetization response surface are trained and optimized again. This process is repeated to strengthen the prediction ability of the Kriging response surface and the decision tree response surface in the high-efficiency region of the motor until the efficiency response surface meets the prediction accuracy.

[0057] S5: The motor topology that yields the highest efficiency value in the efficiency response surface output that meets the requirements and does not demagnetize, denoted as η. max Finite element electromagnetic-thermal coupling calculations were performed on the motor topology. This electromagnetic-thermal coupling was denoted as ET02, and the efficiency η was obtained. re If η re If the optimization efficiency requirement is met, the optimization ends. If the requirement is not met, the temperature of each component of the motor is fixed at the temperature calculated by ET02, and Latin hypercube sampling is performed again to redetermine the optimization parameters and optimization range and start optimization again. This process is repeated until the optimization efficiency target is achieved.

[0058] Furthermore, the motor model was parameterized in detail, and a completely new motor topology can be obtained by modifying certain specific parameter values.

[0059] Furthermore, the experimental point matrix Parameter_X obtained by the Latin hypercube sampling method has a dimension of M×N, where M is the number of samples sampled by the Latin hypercube sampling method and N is the number of parameter values ​​contained in each group of experimental points, resulting in the sampling matrix shown below.

[0060]

[0061] Where, x MN This refers to a parameterized value of the motor, where N is the number of motor topology design parameters, and x... 11 ~x 1NEach set of parameterized values ​​corresponds to one motor design topology. Similarly, there are a total of M sets of parameterized values, corresponding to M motor topology designs.

[0062] Furthermore, the degree of demagnetization D of the magnet. m This refers to the degree of demagnetization at the point where the surface or interior of all magnets in a permanent magnet motor exhibits the greatest degree of demagnetization. The demagnetization degree value can be obtained using the field calculator in finite element software, and it is a continuous value between 0 and 1, defined as follows.

[0063]

[0064] Among them, B r B represents the remanence of the magnet after permanent demagnetization. r0 D represents the remanence of the magnet before permanent demagnetization. m A value of 1 indicates no demagnetization, D m A value of 0 indicates complete demagnetization. If D... m If there is any data that is not 1, then set it to zero, and put the processed data into D. m_01 middle.

[0065] Furthermore, the input power P1 enters the motor, and then the stator winding consumes the power; the copper loss of the stator winding is denoted as p. copper The power consumed by the stator core, and the stator iron loss, are denoted as p. stator The energy is then transferred to the air gap, denoted as electromagnetic power P. EM Then the energy is transferred to the rotor, which consumes a portion of the energy. The rotor iron loss is denoted as p. rotor The rotor cage bar copper loss is denoted as p. bar The eddy current loss of the magnet is denoted as p. PM Finally, energy is transferred to the shaft, minus the wind friction loss p. windloss Thus, the actual output power P2 is obtained. Electromagnetic power P EM As shown below,

[0066]

[0067] Among them, T EM Let n be the electromagnetic torque and n be the motor speed.

[0068] The input power P1 is shown below.

[0069] P1 = P EM +p copper +p stator

[0070] Among them, P EM p is the electromagnetic power. copper For stator winding copper loss, p stator This is recorded as stator iron loss.

[0071] The output power P2 is shown below.

[0072] P2 = P EM -p bar -p rotor -p PM -p windloss

[0073] Among them, P EM p is the electromagnetic power. bar For rotor cage bar copper loss, p rotor For rotor iron loss, p windloss For fan losses, p PM This represents the eddy current loss of the magnet. It should be noted that rotor windage loss and bearing loss are negligible.

[0074] The efficiency η is shown below.

[0075]

[0076] Furthermore, a penalty term D is introduced into the motor efficiency calculation. m The detailed calculation expression is as follows:

[0077] η train =η(x) k1 ,x k2 ,x k3 …x kN )-D coef ×(1-D m )

[0078] Where, η train To provide training data for the Kriging response surface, D coef The demagnetization penalty parameter is designed to reduce the probability of selecting demagnetized individuals when the Kriging response surface predicts high-efficiency individuals.

[0079] Furthermore, a normalized experimental point matrix is ​​used to establish the Kriging response surface model and the decision tree model, and the normalization formula is shown below.

[0080]

[0081] Where, x train x represents the elements in the normalized experimental point matrix. mn x represents the elements in the unnormalized experimental point matrix. max With x min These are the upper and lower limits of a certain parameter during LHS sampling, respectively.

[0082] Furthermore, both the decision tree models XGB and LGBM employ K-folding and random search to train their model parameters. n-folding training involves dividing the dataset into n parts, with parts 1 through n-1 used for training and part n for validation, recording the model accuracy. Then, parts 1 and 3 through n are used for training again, and part 2 for validation, again recording the accuracy. This process continues until the average accuracy of the decision tree model under these model parameters is obtained. The average accuracy under each model parameter generated by random search is then compared, and the highest accuracy is selected for model prediction. Similarly, 5-folding training divides the dataset into 5 parts, with parts 1 through 4 used for training and part 5 for validation, recording the model accuracy. Then, parts 1 and 3 through 5 are used for training again, and part 2 for validation, again recording the accuracy. This process continues until the average accuracy of the decision tree model under these model parameters is obtained. The average accuracy under each model parameter generated by random search is then compared, and the highest accuracy is selected for model prediction, demonstrating good model generalization ability.

[0083] Furthermore, the NSGA-II optimization algorithm is employed to output D using a decision tree response surface. m_01 The output is a constraint condition. The decision tree output value is 0 or 1. The constraint condition is: if the decision tree output is 0, it means that the magnet is demagnetized, and the optimization objective function motor efficiency is set to infinitesimal. If the decision tree output is 1, it means that the magnet is not demagnetized, and the optimization objective function motor efficiency is the Kriging efficiency response surface output.

[0084] Furthermore, after NSGA-II optimization, P individuals with the highest efficiency were selected for finite element verification calculations to calculate the error between the actual and predicted motor efficiency. The calculation formula is shown below.

[0085]

[0086] The actual motor efficiency is η. cal The predicted motor efficiency is η. pre The error is e y If the error is less than p%, the Kriging efficiency response surface prediction accuracy is considered sufficient.

[0087] Further, the permanent magnet motor undergoes electromagnetic-thermal coupling calculations. Specifically, under the motor's operating ambient temperature, the electromagnetic field of the motor is simulated using the finite element method (FEM) to obtain the loss values ​​of each component. These loss values ​​are then converted into heat generation values ​​using the FEM, and the motor's thermal field is simulated to obtain the stable temperatures of each component. Based on these temperatures, the material properties affected by temperature are modified in the electromagnetic field simulation for a second step. The loss values ​​are then converted back into heat generation values, and the motor's thermal field is simulated again to obtain the stable temperatures of each component. These temperatures are compared one by one with the temperatures obtained in the previous step. If the error exceeds 5% (an empirical parameter), another iteration is required. If the accuracy requirements are met, the iteration ends.

[0088] Specific example of applying this patent: The design process of an asynchronous starting permanent magnet synchronous motor is as follows. Figure 1 As shown, its characteristics are: electromagnetic simulation needs to consider its startup process, and electromagnetic simulation takes a long time; electromagnetic thermal coupling needs to be considered (the motor application conditions are sensitive to temperature). Its design specifications are shown in Table 1.

[0089] Table 1 Design parameter requirements for asynchronous starting permanent magnet synchronous motors

[0090] Design Requirements index Axial length limit 390mm (including casing) Radial length limit 275mm (excluding shaft extension length) Rated speed 1500rpm Rated power 5.5kW Efficiency requirements 91.9％ Cooling method Self-fan cooling insulation 155(F) Stator winding allowable temperature rise 80K Maximum casing temperature 90℃ Ambient temperature 25℃ Demagnetizing magnets Not allowed

[0091] According to the patent flowchart, an initial motor topology is first selected, and electromagnetic thermal coupling calculations are performed to ensure that the temperature rise requirements are met. The motor topology is as follows: Figure 2 As shown in (a), the electromagnetic thermal coupling calculation results of the initial design are shown in Table 2.

[0092] Table 2 Electromagnetic and thermal coupling calculation results of initial topology design

[0093]

[0094] Calculations show that under 5.5kW operating conditions, η = 86.99%, which does not meet the efficiency requirements. Furthermore, the magnets demagnetized during motor startup. Figure 3 As shown in (a).

[0095] Then, the temperature rise of each component of the motor is fixed, and the inner radius R of the stator is selected. Si Permanent magnet angle PMtheta, permanent magnet length L PM Motor air gap g, permanent magnet width W PM The spacing between the permanent magnet and the rotor slot (middle_rib) and the thickness of the magnetic bridge between half of the permanent magnet (PM_FB) were used as optimization parameters and sampled using the LHS sampling method. The optimization parameters are shown in Table 3.

[0096] Table 3 Initial Motor Parameter Optimization Range

[0097] parameter unit lower limit value Upper limit <![CDATA[Inner radius R of stator Si > mm 61 75 <![CDATA[Permanent Magnet Angle PM theta > Deg 77 83 <![CDATA[Length L of the permanent magnet PM > mm 22 28 motor air gap g mm 0.75 1.80 <![CDATA[Width W of the permanent magnet PM > mm 5 13 The distance between the permanent magnet and the rotor slot is middle_rib mm 12 25 1 / 2 permanent magnet inter-magnet bridge thickness PM_FB mm 0.8 1.5

[0098] Finite element analysis was performed on the sampling points to calculate the degree of demagnetization D at each sampling point. m Calculate the efficiency η, with a penalty term D. m motor efficiency η train The degree of demagnetization D m Mapped to 01 array D m_01 After normalizing the LHS sampled data, the motor efficiency η is used. train The Kriging response surface is trained using normalized sampled data, and a neural network is used to find the optimal model parameters for the Kriging response surface. A 0-1 array D is used. m_01 The decision tree response surfaces XGB and LGBM were trained with normalized sampled data. The model parameters were trained using K-folding and random search, and the model with high accuracy was selected as the decision tree response surface.

[0099] The NSGA-II optimization algorithm is used for optimization, and the decision tree response surface output D is used for optimization. m_01 The output is a constraint condition. The decision tree output value is 0 or 1. The constraint condition is: if the decision tree output is 0, it means that the magnet is demagnetized, and the optimization objective function motor efficiency is set to infinitesimal. If the decision tree output is 1, it means that the magnet is not demagnetized, and the optimization objective function motor efficiency is the Kriging efficiency response surface output.

[0100] After NSGA-II optimization, P individuals with the highest efficiency were selected for finite element verification calculations. The error between the actual motor efficiency and the predicted motor efficiency was calculated. If the error was less than 0.05%, the Kriging efficiency response surface prediction accuracy was considered sufficient. In this case, the predicted high-efficiency individual was reliable, had a high probability of not demagnetizing, and was closer to the global optimum. If the error did not meet the requirements, the solution of the newly solved finite element method needed to be added to the training set to enhance the predictive ability of the response surface in the high-efficiency region of the motor. The response surface was then trained again, followed by optimization, and this process was repeated until the accuracy requirements were met. This design underwent three rounds of training, and the results are shown in Table 4. It can be seen that the prediction error gradually decreased, and the optimal individual (highest efficiency) output by the finite element calculation was closer to the global optimum.

[0101] Table 4. Iterative training process of Kriging response surface

[0102] Response surface iterations Optimal individual prediction value (%) of response surface methodology Finite element analysis output values ​​(%) Prediction error (%) 1 91.995 91.683 0.312 2 92.092 92.216 0.124 3 92.726 92.687 0.039

[0103] A high-efficiency motor topology design is selected from the Pareto front, such as... Figure 2 As shown in (b). Furthermore, this design prevents demagnetization during motor startup, as... Figure 3As shown in (b), the ambient temperature was set to 25℃, and the electromagnetic thermal coupling calculation was performed again. The final electromagnetic thermal coupling calculation results are compared with those before optimization, as shown in Table 5.

[0104] Table 5. Changes in the calculation results of motor electromagnetic thermal coupling before and after optimization.

[0105]

[0106] After calculation, the optimized motor efficiency η after electromagnetic thermal coupling calculation is obtained. re =92.687%, meeting 91.9% of the design requirements, design complete.

[0107] The above embodiments are merely illustrative examples of the present invention and do not limit its scope of protection. Those skilled in the art can make partial changes to it. For example, electromagnetic thermal coupling optimization considering demagnetization can be performed on other types of permanent magnet motors, and electromagnetic thermal coupling optimization considering demagnetization can be performed on other time-consuming permanent magnet electromagnetic simulations. Any equivalent substitution in any form that conforms to the spirit of the invention falls within the scope of protection of the present invention.

Claims

1. A rapid electromagnetic thermal coupling optimization design method for permanent magnet motors considering demagnetization, characterized in that: Including the following steps: S1: Perform finite element electromagnetic thermal coupling calculation on the permanent magnet motor, denoted as ET01; find a permanent magnet motor topology design that meets the temperature rise condition through this calculation, and obtain the average temperature rise of each component of the motor; fix the temperature rise obtained by ET01 in the parameterized electromagnetic model of the motor, and determine the optimization parameters and optimization range. S2: The experimental point matrix Parameter_X is obtained by Latin hypercube sampling with a certain number of groups. The k-th experimental point is denoted as x. k1 , x k2 , x k3 …x kN The efficiency η(x) of the motor at each experimental point was calculated using finite element simulation. k1 , x k2 , x k3 …x kN ) and the degree of demagnetization of the magnet D m N represents the number of parameter values ​​included in each experimental group. S3: Incorporate the degree of demagnetization D of the magnet into the motor efficiency calculation. m The associated penalty term makes the efficiency η train (D m ,x k1 , x k2 , x k3 …x kN Then, determine the degree of motor demagnetization D. m Mapped to a 01 array, denoted as D m_01 The Kriging response surface was trained using motor efficiency data and a normalized experimental point matrix; demagnetization data of magnets were used. m_01 The decision tree response surface XGB and LGBM were trained with the normalized experimental point matrix, and the response surface model with higher prediction accuracy was selected as the response surface for predicting the degree of demagnetization. S4: The efficiency output by Kriging response surface and the demagnetization status output by decision tree response surface are used instead of finite element calculation; the motor efficiency η is optimized using a non-dominated genetic algorithm with an elite strategy, with the degree of demagnetization as a constraint. P individuals with the highest efficiency are selected for finite element calculation to verify the prediction accuracy of the Kriging response surface. If the accuracy does not meet the requirements, the individuals that do not demagnetize after finite element calculation are merged with the original experimental points, and the Kriging response surface and decision tree response surface are trained and optimized again. This process is repeated to strengthen the prediction ability of the Kriging response surface and decision tree response surface in the high-efficiency region of the motor until the Kriging response surface meets the prediction accuracy. S5: Obtain the motor topology that meets the requirements for the highest efficiency value of the Kriging response surface output and does not demagnetize. The highest efficiency is denoted as η. max ; Finite element electromagnetic thermal coupling calculations were performed on the motor topology. This electromagnetic thermal coupling is denoted as ET02, and the efficiency η was obtained. re If η re If the optimization efficiency requirement is met, the optimization ends. If the requirement is not met, the temperature of each component of the motor is fixed at the temperature calculated by ET02, and Latin hypercube sampling is performed again to redetermine the optimization parameters and optimization range and start optimization again. This process is repeated until the optimization efficiency target is achieved.

2. The rapid electromagnetic thermal coupling optimization design method for permanent magnet motors considering demagnetization according to claim 1, characterized in that: The experimental point matrix Parameter_X obtained by the Latin hypercube method has dimensions M×N, where M is the number of samples obtained by the Latin hypercube method and N is the number of parameter values ​​contained in each group of experimental points. The resulting sampling matrix is ​​shown below: ； Where, x MN This refers to a parameterized value of the motor, where N is the number of motor topology design parameters, and x... 11 ~x 1N Each set of parameterized values ​​corresponds to one motor topology design, and so on, for a total of M sets of parameterized values, corresponding to M motor topology designs.

3. The rapid electromagnetic thermal coupling optimization design method for permanent magnet motors considering demagnetization according to claim 1, characterized in that: Degree of demagnetization of the magnet D m This refers to the degree of demagnetization at the point where the surface or interior of all magnets in a permanent magnet motor exhibits the greatest degree of demagnetization. The demagnetization degree value is obtained using the field calculator in finite element software and ranges from 0 to 1 as a continuous value, defined as follows. ； Among them, B r B represents the remanence of the magnet after permanent demagnetization. r0 D represents the remanence of the magnet before permanent demagnetization. m A value of 1 indicates no demagnetization, D m If D is 0, it means complete demagnetization. m If there is any data that is not 1, then set it to zero, and put the processed data into D. m_01 middle.

4. The rapid electromagnetic thermal coupling optimization design method for a permanent magnet motor considering demagnetization according to claim 1, characterized in that: Input power P1 enters the motor, and then the stator windings consume power; the copper loss of the stator windings is denoted as p. copper The power consumed by the stator core, and the stator iron loss, are denoted as p. stator The energy is then transferred to the air gap, denoted as electromagnetic power P. EM Then the energy is transferred to the rotor, which consumes a portion of the energy. The rotor iron loss is denoted as p. rotor The rotor cage bar copper loss is denoted as p. bar The eddy current loss of the magnet is denoted as p. PM Finally, the energy is transferred to the shaft, minus the wind friction loss p. windloss The actual output power P2 and electromagnetic power P are obtained. EM As shown below, ； Among them, T EM Where n is the electromagnetic torque and n is the motor speed; The input power P1 is shown below. ； Among them, P EM p is the electromagnetic power. copper For stator winding copper loss, p stator This is recorded as stator iron loss; The output power P2 is shown below. ； Among them, P EM p is the electromagnetic power. bar For rotor cage bar copper loss, p rotor For rotor iron loss, p windloss For wind friction loss, p PM This refers to the eddy current loss of the magnet; The efficiency η is shown below. 。 5. The rapid electromagnetic thermal coupling optimization design method for a permanent magnet motor considering demagnetization according to claim 1, characterized in that: Introducing a penalty term D in motor efficiency calculation m The detailed calculation expression is as follows: ； Where η is the efficiency of the motor, η train To provide training data for the Kriging response surface, D coef The demagnetization penalty parameter is designed to reduce the probability of selecting demagnetized individuals when the Kriging response surface predicts high-efficiency individuals.

6. The rapid electromagnetic thermal coupling optimization design method for a permanent magnet motor considering demagnetization according to claim 1, characterized in that: The Kriging response surface model and decision tree model are established using a normalized experimental point matrix. The normalization formula is shown below: ； Where, x train x represents the elements in the normalized experimental point matrix. mn x represents the elements in the unnormalized experimental point matrix. max With x min These are the upper and lower limits of a certain parameter during Latin hypercube sampling.

7. The rapid electromagnetic thermal coupling optimization design method for a permanent magnet motor considering demagnetization according to claim 1, characterized in that: Both decision tree models XGB and LGBM use K-folding and random search to train model parameters. n-folding training involves dividing the dataset into n parts, where the first to n-1 parts are used for model training and the nth part is used for model validation, recording the model accuracy. Then, the first part and the third to nth parts are used for model training, and the second part is used for model validation, recording the model accuracy again. This process is repeated until the average accuracy of the decision tree model under these model parameters is obtained. The average accuracy under each model parameter generated by random search is compared, and the highest accuracy is selected for model prediction.

8. The rapid electromagnetic thermal coupling optimization design method for a permanent magnet motor considering demagnetization according to claim 1, characterized in that: The NSGA-II optimization algorithm is used to output D using a decision tree response surface. m_01 The output is a constraint condition. The decision tree output value is 0 or 1. The constraint condition is: if the decision tree output is 0, it means that the magnet is demagnetized, and the optimization objective function motor efficiency is set to infinitesimal. If the decision tree output is 1, it means that the magnet is not demagnetized, and the optimization objective function motor efficiency is the Kriging response surface output.

9. The rapid electromagnetic thermal coupling optimization design method for a permanent magnet motor considering demagnetization according to claim 1, characterized in that: After NSGA-II optimization, P individuals with the highest efficiency were selected for finite element verification calculations. The error between the actual motor efficiency and the predicted motor efficiency was calculated, and the calculation formula is shown below. ； The actual motor efficiency is η. cal The predicted motor efficiency is η. pre The error is e y If the error is less than p%, the Kriging response surface prediction accuracy is considered sufficient.

10. The rapid electromagnetic thermal coupling optimization design method for a permanent magnet motor considering demagnetization according to claim 1, characterized in that: Electromagnetic-thermal coupling calculations for permanent magnet motors specifically involve: Simulating the motor's electromagnetic field using the finite element method (FEM) under the motor's operating ambient temperature to obtain the loss values ​​of each component; converting these loss values ​​into heat generation values ​​using the FEM to simulate the motor's thermal field and obtain the stable temperatures of each component; then, using these temperatures as a basis, modifying the material properties affected by temperature in the electromagnetic field, performing a second electromagnetic field simulation to obtain the loss values ​​again, converting them back into heat generation values, and simulating the motor's thermal field to obtain the stable temperatures of each component; comparing these temperatures with those obtained in the previous step, and if the error exceeds 5%, another iteration is required; if the accuracy requirements are met, the iteration ends.

Images