Permanent magnet motor topology hierarchy optimization design method based on variable air gap domain and motor

Abstract

The invention discloses a variable air gap domain-based permanent magnet motor topology hierarchy optimization design method and a motor, and the method comprises the steps: firstly determining an electromagnetic performance layer optimization target, constructing a geometric parameter model, carrying out the dimensionality reduction of a high-dimensional parameter, and constructing a first hierarchy optimization sub-model I; performing geometric parameter optimization on the stator and rotor basic domain by combining a multi-objective optimization algorithm to obtain an optimized motor I; analyzing and optimizing the characteristic electromagnetic force of the motor I under different working conditions to serve as a second-level optimization target, constructing a second-level optimization sub-model II, performing small-scale online topological optimization in combination with a multi-target genetic algorithm, performing secondary design on a motor air gap design domain, obtaining a sample data set, constructing a parallel agent model, and constructing a second-level optimization sub-model II; then large-scale off-line topological optimization is carried out, and remodeling of a motor air gap design domain is achieved; and performing simulation evaluation on the final optimized motor. According to the method, on the premise that the vehicle permanent magnet driving electromagnetic performance is kept, electromagnetic vibration noise under multiple working conditions is suppressed, and meanwhile optimization efficiency is remarkably improved.

Description

Technical Field

[0001] The present invention relates to a permanent magnet motor topology level optimization design method based on a variable air gap domain and a motor, belonging to the technical field of permanent magnet motors. Background Art

[0002] In new energy vehicles, the drive motor, power battery, and vehicle electronic control are the three key technologies in the electrical system of new energy vehicles. Given that the power density of current power batteries is unlikely to be significantly improved in the short term, optimizing the performance of the vehicle drive motor and its control system has become key to improving vehicle performance. Permanent magnet motors are widely used due to their high power density and high efficiency. For permanent magnet synchronous motors, the rotor generally adopts an internal multi-layer magnetic steel arrangement. By increasing the motor's saliency, the torque density of the motor is further improved. At the same time, leakage magnetic channels are designed to increase the motor's speed regulation range. However, this design also makes the motor rotor magnetic circuit more complex. At the same time, local magnetic saturation under different operating conditions introduces a large number of additional harmonics, which causes motor vibration and noise problems.

[0003] Generally speaking, the vibration noise of permanent magnet synchronous motors (PMSMs) stems from two main sources: radial housing vibration caused by radial electromagnetic forces in the motor air gap, and gearbox shaft vibration caused by motor torque pulsation. To improve motor housing vibration or shaft vibration, researchers have proposed methods to improve motor torque pulsation and electromagnetic vibration, including altering the motor pole slot ratio and winding arrangement, optimizing stator slot skew, rotor pole skew, and stator-rotor topology. For example, Chinese Publication No. CN114915222A discloses a motor noise optimization method based on auxiliary slots in the rotor magnets, and Chinese Publication No. CN219643773U discloses a multi-skewed slot motor rotor for noise and vibration reduction. However, current methods for suppressing motor noise typically rely on adjusting the motor's structural parameters or topology, thereby adjusting the air gap magnetic field distribution, improving the motor's electromagnetic force and torque pulsation, and ultimately achieving motor noise suppression. However, adjusting the stator-rotor topology often alters the stator-rotor magnetic circuit, impacting basic motor performance such as torque and efficiency. With the continuous development of multi-physics simulation and optimization technology, multi-objective optimization methods are used to improve the comprehensive performance of motors, including vibration and noise. For example, the document with Chinese publication number CN114818166A discloses a switched reluctance motor vibration reduction and noise reduction optimization design method, which is used to improve the vibration and noise of the switched reluctance motor. However, existing multi-objective optimization methods often improve the electromagnetic characteristics and vibration and noise characteristics of the motor by adjusting the topological structure of the motor stator and rotor parameters. However, there is often a high degree of coupling between torque, efficiency objectives and electromagnetic force objectives. Reducing vibration characteristics is often accompanied by a decrease in motor torque density and efficiency. In particular, considering the operating characteristics of automotive drive motors under variable working conditions, traditional multi-objective optimization is difficult to obtain an effective motor design solution. Summary of the Invention

[0004] This invention aims to suppress the vibration and noise characteristics of automotive permanent magnet motors under multiple operating conditions while maintaining their basic electromagnetic performance. This method proposes a topology-level optimization method for permanent magnet motors based on a variable air gap domain. This method builds on the traditional stator and rotor topology design of permanent magnet motors by introducing the concept of an air gap domain, dividing the motor topology into two parts: the stator and rotor basic domain and the air gap design domain. This approach suppresses electromagnetic noise under multiple operating conditions while maintaining basic electromagnetic performance.

[0005] To achieve the above object, the present invention adopts a technical solution: a method for topology level optimization of a permanent magnet motor based on a variable air gap domain, comprising the following steps:

[0006] Step 1: Initial Motor Design: Based on the automotive drive permanent magnet motor application scenario, determine the motor's design requirements and specifications. Based on these requirements, select the motor, fix some key motor parameters, and determine the basic motor structure. Then, based on the motor's design specifications, determine the geometric design parameters of the motor's stator and rotor, as well as the performance to be optimized.

[0007] Step 2: Optimization of the first-level stator and rotor basic domain parameters: Through local sensitivity analysis, determine the sensitivity of the initial design parameters to the design objectives, select the top 7 sensitivity parameters as high-sensitivity design parameters based on the comprehensive sensitivity, and construct the first-level optimization sub-model I:

[0008]

[0009] Among them, T orque (x) is the average output torque, T orqueRipple (x) is the torque ripple, C oreloss (x), C opperloss (x) are the iron loss and copper loss under rated working conditions, J rate is the current density under rated working conditions, V pm1 、V pm2 are the volumes of a single inner and outer permanent magnet, respectively.

[0010] The optimization sub-model I is solved based on the Kriging surrogate model and the NSGA-II genetic algorithm. In the obtained non-dominated solution set, the optimal design point is selected as the final design parameter of the basic geometric structure of the motor stator and rotor, and the optimized motor I is obtained.

[0011] Step 3. Obtain the radial and tangential electromagnetic force waveforms of the optimized motor I under rated operating conditions through the finite element analysis method, and obtain the two groups of zero-order characteristic radial electromagnetic forces with the highest amplitude and the group of zero-order characteristic tangential electromagnetic forces with the highest amplitude based on the two-dimensional Fourier decomposition; obtain the characteristic electromagnetic force change characteristics under different loads through the finite element analysis method; divide the operating conditions of the drive motor into two typical operating conditions of low speed and heavy load OC1 and high speed and light load OC2, and select the characteristic electromagnetic forces under the two typical operating conditions as the second-level optimization target according to the characteristic electromagnetic force change characteristics to suppress the electromagnetic vibration noise problem under multiple operating conditions.

[0012] Step 4, second-level air gap design domain topology optimization: Determine the motor air gap design domain and perform topology optimization based on the normalized Gaussian basis function mesh (NGnet) within the specified design domain;

[0013] First, construct the second-level optimization sub-model II:

[0014]

[0015] Where T OC1ini 、T OC2ini are the average torques of the initial motor under two typical working conditions; T OC1op 、T OC2op are the average torques of the optimized motor under two typical working conditions; ω0, ω1, ..., ω 21 are the weight coefficients of the 22 Gaussian basis functions.

[0016] Then, a Python script was used to call Maxwell to perform online topology optimization, obtaining 600 data sets. A parallel agent model was constructed using the uniformly distributed data sets, and large-scale offline topology optimization was performed to obtain the Pareto solution set of the last generation of the population. The optimal solution was selected, and the corresponding Gaussian basis function weight coefficient was obtained. This was then brought into the Python script to generate the optimized motor II after air gap domain optimization, which is the final optimized motor.

[0017] Step 5: Verification of the optimization scheme: Perform finite element comparative simulation evaluation on the electromagnetic performance and vibration and noise characteristics under multiple working conditions of the optimized motor II and the initial motor.

[0018] Beneficial effects of the present invention:

[0019] 1. This paper proposes a topology-level optimization design method for permanent magnet motors based on variable air gap domains. This method, building on the traditional stator and rotor topology design of permanent magnet motors, introduces the concept of air gap domain design, dividing the motor topology into two parts: the stator and rotor basic domain and the air gap design domain. At the first level, the stator and rotor basic domain is optimized to improve fundamental electromagnetic performance, such as motor torque. At the second level, the air gap design domain undergoes secondary design. By using topology optimization to adjust the air gap design domain topology and, consequently, its air gap permeance, this method suppresses electromagnetic noise while maintaining the motor's fundamental electromagnetic performance.

[0020] 2. In the topology optimization process of the second-level variable air gap domain, the present invention combines the changing characteristics of the characteristic electromagnetic force under different working conditions, selects different characteristic electromagnetic force amplitudes under typical working conditions as optimization targets, and uses the topology optimization method based on the normalized Gaussian basis function grid to perform secondary design of the motor air gap design domain, thereby achieving effective suppression of vibration noise under multiple working conditions of the permanent magnet motor.

[0021] 3. At the first level, this invention uses local sensitivity analysis to reduce the dimensionality of high-dimensional design parameter variables, improving the optimization efficiency of the first-level stator and rotor basic design domain. At the second level, a small-scale online topology optimization is used to collect a uniform data set, construct a parallel agent model, and then perform a large-scale offline topology optimization, further reducing the optimization efficiency of the second level. These two aspects effectively improve the optimization efficiency of permanent magnet motor noise suppression under multiple operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flowchart for the specific implementation of the proposed permanent magnet motor topology hierarchical optimization based on variable air gap domain.

[0023] Figure 2 This is a three-dimensional topology diagram of the hairpin winding permanent magnet flat wire motor and a planar structure diagram of its stator and rotor according to an embodiment of the present invention.

[0024] Figure 3 1 is a parametric model of the stator and rotor of the motor according to an embodiment of the present invention and a diagram of the selected geometric design parameters of the stator and rotor.

[0025] Figure 4 This is a heat map of the local sensitivity analysis results of the stator and rotor design parameters of the motor according to an embodiment of the present invention.

[0026] Figure 5 This is a flowchart for solving the specific solution of the stator and rotor basic domain geometric parameter optimization sub-model I according to an embodiment of the present invention.

[0027] Figure 6 This is a diagram showing the optimization results of the first-level stator and rotor basic domain optimization model I according to an embodiment of the present invention.

[0028] Figure 7 The radial and tangential electromagnetic force waveforms and harmonic distribution diagrams of the motor 1 are optimized for the embodiment of the present invention.

[0029] Figure 8 The figure shows the amplitude distribution diagram of the radial electromagnetic forces with different characteristics of the motor 1 under all working conditions, which is optimized in an embodiment of the present invention.

[0030] Figure 9 Graph showing the mapping relationship between topology optimization design parameters and variable air gap permeability according to an embodiment of the present invention.

[0031] Figure 10 This is a flow chart of solving the optimization sub-model II for small-scale online topology optimization in the air gap design domain according to an embodiment of the present invention.

[0032] Figure 11 A comparison chart of six sets of parallel agent models established in an embodiment of the present invention and finite element simulation results.

[0033] Figure 12 This is a flow chart of solving the optimization sub-model II for large-scale offline topology optimization of the air gap design domain according to an embodiment of the present invention.

[0034] Figure 13 This is a diagram of the air gap domain topology iteration process during large-scale offline topology optimization according to an embodiment of the present invention.

[0035] Figure 14 The Pareto front and optimal motor air gap domain structure diagram obtained by large-scale offline optimization in an embodiment of the present invention.

[0036] Figure 15 Comparison of the electromagnetic characteristics of the motor before and after optimization according to the embodiment of the present invention.

[0037] Figure 16 This figure compares the vibration characteristics of the motor under multiple working conditions before and after optimization according to the embodiment of the present invention.

[0038] Figure 17 This figure compares the noise characteristics of the motor under multiple working conditions before and after optimization according to the embodiment of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below with reference to the accompanying drawings and examples.

[0040] Example: Research on the optimization design of multi-operating conditions noise suppression for a 48-slot 8-pole hairpin winding permanent magnet flat wire motor. Figure 1 As shown, Figure 1 A flowchart of a topology level optimization design method for a permanent magnet motor with a variable air gap domain provided in an embodiment of the present invention to suppress electromagnetic noise under multiple working conditions.

[0041] Figure 2This is a topological diagram of the flat-wire permanent magnet motor in an embodiment. In the figure, 1 is the hairpin flat-wire winding, 2 is the motor stator, 3 is the motor rotor, 4 is the NdFeB permanent magnet, 5 is the motor bearing, and 6 is the housing. The motor in this embodiment of the present invention is a 48-slot, 8-pole hairpin permanent magnet flat-wire motor. Unlike the traditional method of inserting Litz coil windings through the stator slot openings, this motor uses hairpin flat-wire windings that are inserted directly from the end of the core and staggered. This enhances the winding's mechanical strength while reducing the stator slot opening width, improving electromagnetic performance. This method also effectively shortens the end winding length, thereby reducing copper loss.

[0042] The stator slots utilize a rectangular structure to maximize slot fill rate and significantly improve heat dissipation. The rotor utilizes a double-layer V-shaped permanent magnet arrangement, with rectangular magnetic barriers introduced between the second layer to reduce self-leakage magnetic flux and maximize permanent magnet utilization. Furthermore, to reduce the rotor's moment of inertia, triangular and trapezoidal weight-reducing holes are designed near the shaft.

[0043] The specific implementation steps are as follows:

[0044] S1: Based on the basic performance requirements of automotive drive motors, the optimization targets for the first level, the geometric parameters of the stator and rotor basic design domain, are determined. For automotive drive motors, on the one hand, they face a variety of complex and variable operating conditions, placing higher demands on the motor's torque quality. Therefore, the average output torque and torque ripple under the motor's rated operating conditions are set as the targets to be optimized. On the other hand, to improve the endurance of electric vehicles and further extend their cruising range, the iron loss and copper loss under the motor's rated operating conditions are set as the targets to be optimized. Furthermore, the cost is limited by the current density and the amount of permanent magnets used during the optimization process.

[0045] S2: Establishing a parameterized model of the embodiment: Selecting 7 stator design parameters and 13 rotor design parameters to establish a permanent magnet motor stator and rotor geometric parameterized model. Figure 3 : 1 is the distance between the inner and outer permanent magnets h5, 2 is the height of the inner permanent magnet h4, 3 is the width of the rectangular magnetic barrier h1, 4 is the length of the inner permanent magnet l1, 5 is the length of the outer permanent magnet l2, 6 is the width of the inner permanent magnet t1, 7 is the width of the outer permanent magnet t2, 8 is the length of the rectangular magnetic barrier w2, 9 is the width of the outer reinforcement rib w1, 10 is the width of the inner permanent magnet magnetic bridge w5, 11 is the width of the outer permanent magnet magnetic bridge w4, 12 is the angle a1 of the inner permanent magnet, 13 is the angle a2 of the outer permanent magnet, 14 is the width t3 of the stator rectangular slot, 15 is the width w3 of the stator slot, 16 is the thickness h3 of the stator tooth, 17 is the length h2 of the rectangular slot, 18 is the inclination angle a3 of the stator tooth, and 19 is the outer radius R of the stator. so , 20 is the inner radius R of the stator siAccording to the initial design, the basic specifications and initial values ​​of the design parameters of the motor are shown in Table 1:

[0046] Table 1

[0047]

[0048]

[0049] Due to the large number of motor design parameters, the optimization of high-dimensional design variables generally has common problems such as convergence difficulties and low optimization efficiency. Therefore, in order to solve the problem of many stator and rotor design parameters at this level, the present invention uses the Latin Hypercube Sampling (LHS) method to select the design variables x1~x1. 18 300 uniform and non-repetitive sample points were randomly selected within the determined 18-dimensional design space, and the target sizes corresponding to all sample points were calculated using ANSYS Maxwell simulation software. The sensitivity of each design parameter to the four optimization targets was then calculated using the local sensitivity analysis method. The calculation formula is as follows:

[0050]

[0051] Among them, S i (i=1,2,3,4) is the local sensitivity of the design parameter to the i-th target, F(x0) is the function value corresponding to the design parameter reference value x0, F(x0±Δxi) is the function value corresponding to the design parameter reference value x0 with the disturbance ±Δxi added, Δx i is the design parameter x i The increment of Δx is based on the design parameter fluctuation method. i The value is 10% of the design parameter baseline value, and the sensitivity of the corresponding target is calculated. The sensitivity analysis results of each design parameter are as follows: Figure 4 As shown in the figure. After obtaining the sensitivity of each design parameter to the four targets, the weighted sum of the absolute values ​​of the four sensitivities was calculated by assigning the same weight coefficient of 0.25. The 18 weighted sensitivities were ranked, and the top seven design parameters were selected as highly sensitive design parameters. The seven highly sensitive design parameters include two stator structural parameters: stator rectangular slot width t3 and stator rectangular slot length h2; and five rotor structural parameters: inner layer permanent magnet angle a1, outer layer permanent magnet angle a2, inner layer permanent magnet height h4, inner and outer layer permanent magnet spacing h5, and inner layer permanent magnet length l1.

[0052] S3: After performing local sensitivity analysis and dimensionality reduction on the high-dimensional design parameters and identifying seven highly sensitive design parameters, the geometric parameters of the first-level stator and rotor basic design domain will be optimized. First, considering issues such as motor heat dissipation, cost, and design parameter conflicts, the motor constraints are determined, including:

[0053] First, considering that the flat wire motor of the embodiment is pre-cooled with oil, the armature current density is limited to 15A / mm 2 ;

[0054] Secondly, considering the cost of the permanent magnet flat wire motor of the embodiment, in the process of optimization design, the amount of single NdFeB permanent magnets in the inner and outer layers is set to a fixed value, wherein the amount of single NdFeB permanent magnets in the inner layer is 9945mm 3 ; The outer layer single NdFeB permanent magnet usage is 3705mm 3 ;

[0055] Combining the four motor optimization objectives determined in step S1, the seven motor design parameters determined in step S2, and the constraints set in step S3 above, the motor optimization sub-model I can be obtained:

[0056]

[0057] Where: T orqueRipple (x) is the torque ripple, T orque (x) is the average output torque, C oreloss (x), C opperloss (x) Iron loss and copper loss under rated working conditions, J rate is the current density under rated working conditions, V pm1 、V pm2 are the volumes of a single inner and outer permanent magnet respectively. Figure 5 This is a specific solution flow chart for the first-level stator and rotor basic domain optimization sub-model I according to an embodiment of the present invention, which specifically includes the following sub-steps:

[0058] S31. Perform Latin hypercube sampling within the 7-dimensional design space of the highly sensitive parameters h2, t3, a1, a2, h4, h5, and l1 of the embodiment to obtain a total of 1000 sample points. Calculate the average output torque, torque ripple, iron loss, and copper loss of all sample points at the rated operating point using ANSYS Maxwell.

[0059] S32. Construct a Kriging model based on the sample points obtained in S31. The root mean square error (RMSE) is used to evaluate the model accuracy. When the RMSE is less than 1%, the surrogate model accuracy is considered to meet subsequent optimization requirements.

[0060] S33, randomly initialize the first generation population. Generate N motor individuals (design parameter combinations) in the 7-dimensional design space through random functions, where N represents the number of samples, and try to make these individuals spread throughout the entire design space to facilitate subsequent optimization. Based on the Kriging proxy model constructed in S32, calculate the optimization target C of the population individuals in turn. oreloss (x), C opperloss (x), T orqueRipple(x), T orque (x).

[0061] S34, motor optimization target C obtained by S33 oreloss (x), C opperloss (x), T orqueRipple (x), T orque (x) is used as the individual fitness, and the NSGA-II algorithm is used to perform non-dominated sorting on the population, dividing individuals into multiple Pareto levels (dominated frontier layers). Individuals with lower non-dominated levels have higher priorities. Individuals within the same non-dominated level are sorted based on crowding distance, with greater crowding (i.e., the sparser the individuals in the target space), giving them higher priorities. Subsequently, the parent population and the equivalent sub-generation population generated by the genetic operator are merged into a population of 2N individuals. Based on the above sorting rules, N individuals are selected to form the next generation population, maintaining a constant population size while improving overall quality and diversity.

[0062] S35. Evolutionarily update the N individuals obtained in S34, and generate a new generation of offspring through operations such as crossover, mutation, and selection.

[0063] S36. Determine whether the optimization has converged by determining whether the difference between the two iterations meets the set accuracy or whether the number of iterations has reached the maximum number of iterations. If not, merge the parent and child populations and repeat the non-dominated sorting and evolution operations of S33-S35. Otherwise, the optimization ends, the Pareto front solution set is obtained, and the optimal geometric parameter combination is selected.

[0064] The specific parameter settings of the NSGA-II algorithm corresponding to the above S33-S36 in the present invention are as follows: the population size is 50, the crossover probability is 0.9, the mutation probability is 1 / 6, and the maximum number of iterations is 100.

[0065] Figure 6 The results of optimizing the geometric parameters of the first-level stator and rotor basic design domain in the embodiment of the present invention are given. The Pareto front solution set and corresponding design parameter combinations of the four optimization objectives in the parallel coordinate system are given. The comparison of the motor design parameters before and after optimization is shown in the following table:

[0066] Table 2

[0067]

[0068] This level determines the basic structure of the motor's stator and rotor, and an optimized motor I with improved electromagnetic performance such as torque and efficiency is obtained.

[0069] S4: This step is used to obtain the characteristic electromagnetic force related to the electromagnetic noise characteristics of the motor under different working conditions, which is used as the optimization target for the second-level variable air gap domain topology optimization. First, the radial and tangential electromagnetic force waveforms of the optimized motor I under rated working conditions are obtained through ANSYS Maxwell software simulation. Then, the time-space waveforms of the radial and tangential electromagnetic forces are decomposed by the two-dimensional Fourier decomposition method (FFT2D) to obtain the corresponding time-space harmonic distribution, such as Figure 7 As shown. Figure 7 (a) is the radial electromagnetic force waveform and its harmonic distribution, Figure 7 (b) is the tangential electromagnetic force waveform and its harmonic distribution. Considering the zero-order force wave that has the greatest impact on the motor noise, the radial electromagnetic force P with the highest amplitude, spatial order of 0 and time frequency of 6 is selected. r0s6t and the radial electromagnetic force P with spatial order 0 and temporal frequency 12 r0s12t As the characteristic radial electromagnetic force to characterize the radial vibration of the motor, the tangential electromagnetic force P with the highest amplitude and spatial order of 0 and time frequency of 12 is selected. t0s12t As the characteristic tangential electromagnetic force that characterizes the axial vibration of the motor.

[0070] The variation characteristics of the characteristic radial electromagnetic force under different loads are further analyzed. First, the weak magnetic current I d , load current I q Full factor sampling was performed within the range, with each factor taking 10 levels to form 100 groups of working condition samples. The characteristic radial electromagnetic force P under each working condition was obtained by simulation using ANSYS Maxwell software. r0s6t and P r0s12t The amplitude size is calculated by using the cubic spline interpolation method to construct the characteristic electromagnetic force amplitude distribution map within the full working range, such as Figure 8 As shown. Figure 8 (a) It can be seen that P r0s6t It is greatly affected by the load current, and its amplitude increases with the q-axis current I q Continuously rising due to the increase of Figure 8 (b) It can be seen that P r0s12t As the load current increases, the weak magnetic current I d The injection of the radial electromagnetic force P has a significant increasing trend in the high-speed region, indicating that the electromagnetic force harmonics also need to be effectively suppressed in high-speed weak magnetic field. Therefore, low-speed heavy-load OC1 and high-speed light-load OC2 are selected as typical working conditions, and the radial electromagnetic force P is selected as the radial electromagnetic force P. r0s6t 、P r0s12t As the vibration and noise suppression target of the low-speed and heavy-load working condition OC1, the radial electromagnetic force P is selected. r0s12t and the tangential electromagnetic force P t0s12tAs the target of noise and vibration suppression in high-speed and light-load operating condition OC2.

[0071] S5: Based on S4, the characteristic electromagnetic force targets under different working conditions are extracted, and the variable air gap permeability ΔΛ(ω0,ω1,...,ω n ) and topology optimization design parameters ω0,ω1,...,ω n In this step, the NSGA-II genetic algorithm is used to calculate the characteristic electromagnetic force of the low-speed and heavy-load working condition OC1 extracted in S4: P r0s6t and P r0s12t , characteristic electromagnetic force of high speed and light load condition OC2: P r0s12t and P t0s12t The purpose of performing small-scale online optimization is to extract a uniformly distributed sample data set containing 600 sets of design parameters and their corresponding optimization targets. This data set is then used to train a parallel agent model with satisfactory accuracy, in preparation for step S6. Therefore, the small-scale online topology optimization sub-model II of this embodiment of the present invention is as follows:

[0072]

[0073] Where T OC1ini 、T OC2ini are the average torques of the initial motor under two typical working conditions; T OC1OP 、T OC2OP are the average torques of the optimized motor under two typical working conditions; ω0, ω1, ..., ω 21 are the weight coefficients of the 22 Gaussian basis functions.

[0074] Figure 10 This is a flowchart for solving the small-scale online optimization of the second-level variable air gap domain optimization sub-model II according to an embodiment of the present invention, which specifically includes the following sub-steps:

[0075] S51. First, select the topology optimization design area. In order to reduce the number of iterations and improve the optimization efficiency, considering the symmetry of the initial motor, only 1 / 16 of the motor air gap domain is selected for local optimization design. The specific motor air gap design domain is as follows: Figure 9 As shown in (a), the design area is divided into 3000 air gap micro-grids of uniform size by 100 equal divisions in the tangential direction and 30 equal divisions in the radial direction. Figure 9 As shown in (b), the position of each air gap microgrid is determined by the corresponding position function n(x,y), where x is the tangential coordinate and y is the radial coordinate. The size of this grid determines its material properties, thus affecting the air gap permeance harmonic distribution across the entire air gap region. The specific calculation rules for this microgrid value are as follows:

[0076] S511, place 22 two-dimensional Gaussian basis functions uniformly covering the entire design area at the radial 1 / 3 and 2 / 3 of the area along the tangential direction of the motor rotor, that is, Figure 9 (b) The circle. By using formula (6), we can calculate the 22 sets of values ​​G of each microgrid n(x,y) under 22 two-dimensional Gaussian basis functions. i (x,y)(i=0,1,2,…,21);

[0077]

[0078] S512, normalize the 22 groups of values ​​using formula (7), and give each normalized value b i (x,y) assigns the corresponding weight coefficient ω i ;

[0079]

[0080] S513, calculate the final weighted value M of the microgrid n(x,y) by formula (8) n (x, y), that is, the value of each microgrid is obtained by multiplying the corresponding 22 normalized Gaussian basis function values ​​by the corresponding weight coefficient and summing them;

[0081]

[0082] S514. Finally, the material properties are determined by comparing the value with the given reference value ξ. The rule is shown in formula (9):

[0083]

[0084] where x i and y i is the central coordinate of the i-th Gaussian basis function, and σ is the standard deviation of the two-dimensional Gaussian basis function. Its size determines the expansion range of each two-dimensional Gaussian basis function. Selecting an appropriate σ value can make the transition between each basis function smoother, that is, there will be no continuous micro-grid material property mutations, effectively avoiding the generation of iron core suspension structure, thereby significantly improving the processability of the generated topology.

[0085] When assigning different weight coefficients ω0, ω1, ..., ω to 22 Gaussian basis functions 21 When the weight coefficients are updated, the corresponding microgrid material properties will also change accordingly. The genetic algorithm NSGA-II is used to guide the iteration of the weight coefficients, so that the structure of the air gap topology is continuously optimized. The corresponding air gap permeance Λ(θ) is also continuously updated with the 22 weight coefficients to form a variable air gap permeability ΔΛ(ω0,ω1,...,ω 21 ).

[0086] S52. Process the generated Gaussian basis function weight coefficient combinations in sequence based on the Python script: assign corresponding material properties to the corresponding air gap microgrid numbers according to the material criteria, call ANSYS Maxwell to assign the same material properties to the microgrids with the same numbers in the software, and realize the generation of the motor air gap domain topology from the Gaussian basis function weight coefficients.

[0087] S53, randomly initialize the first generation population. Generate N motor individuals (design parameter combinations) in the 22-dimensional design space through random functions, where N represents the number of samples, and try to make these individuals spread throughout the entire design space to facilitate subsequent optimization. Then, through the sample motors generated by finite element simulation, calculate the optimization target P of the population individuals in turn. r0s6t (x), P r0s12 (x)(OC1),P r0s12t (x)(OC2) and P t0s12t (x), constraint target T OC1 、T OC2 .

[0088] S54, the motor optimization target P obtained in S53 r0s6t (x), P r0s12 (x)(OC1),P r0s12t (x)(OC2) and P t0s12t (x) is used as individual fitness, and individuals within the population are sorted in a non-dominated hierarchical manner. The constrained non-dominated hierarchical sorting follows the following rules: 1) feasible solutions take precedence over infeasible solutions; 2) infeasible solutions are sorted from low to high according to the total violation of the two constraint objectives (CV); and 3) feasible solutions are sorted according to Pareto dominance. Based on the hierarchical sorting results, individuals are ranked, with lower non-dominated ranks receiving higher priority. Individuals with the same non-dominated rank are further ranked based on the degree of crowding between individuals, with greater distances between individuals receiving higher priority. Subsequently, the parent population and the equivalent subpopulation generated by the genetic operator are merged into a population of 2N individuals. Based on the aforementioned sorting rules, N individuals are selected to form the next generation population, maintaining a constant population size while improving overall quality and diversity.

[0089] S55. Evolutionarily update the N individuals obtained in S54, and generate a new generation of offspring through operations such as crossover, mutation, and selection.

[0090] S56. Save all sample point data in the population, including 22 sets of design parameters and their corresponding 4 optimization objectives and 2 constraint objective data.

[0091] S57: Determine whether the optimization has converged by determining whether the difference between the two iteration results meets the set accuracy or whether the number of iterations has reached the maximum number of iterations. If not, merge the parent and child populations and repeat the non-dominated sorting and evolution operations of S53-S56; otherwise, the optimization ends.

[0092] S58. After the optimization iteration is completed, the sample data sets in all populations are output. Each population contains 30 sets of optimization objectives and constraint target data sets obtained by 22-dimensional Gaussian function weight coefficients and their corresponding motor topology simulations. The maximum number of iterations is 20, so there are a total of 600 sets of sample data sets.

[0093] The specific parameter settings of the NSGA-II algorithm corresponding to the above S53-S58 in the present invention are as follows: the population size is 30, the crossover probability is 0.9, the mutation probability is 1 / 6, and the maximum number of iterations is 20.

[0094] S59. A parallel agent model was constructed using the 600 datasets obtained. Specifically, the LightGBM method was used for optimization and the random forest (RF) method was used for constraint. The root mean square error (RMSE) and the coefficient of determination (R) were used to calculate the optimal target. 2 To evaluate the model accuracy, when the root mean square error is less than 1% and the coefficient of determination is greater than 0.90, the surrogate model accuracy is considered to meet the subsequent optimization requirements. At the same time, in order to further verify the model prediction accuracy, 50 sets of design parameter combinations are generated by the Latin Hypercube Sampling (LHS) machine, and the prediction results are compared with the finite element simulation results. Figure 11 As shown in the figure, the maximum error of the prediction model is within 5%.

[0095] S6: The characteristic electromagnetic force of the low-speed and heavy-load condition OC1 extracted in S4 is calculated using the NSGA-II genetic algorithm: P r0s6t and P r0s12t ; Characteristic electromagnetic force of high-speed and light-load working condition OC2: P r0s12t and P t0s12t , and perform large-scale offline multi-objective optimization. The selected design parameters are the same as S5, which are 22 Gaussian basis function weight coefficients. Therefore, the large-scale offline topology optimization sub-model II of the embodiment of the present invention is consistent with the optimization sub-model II of S5:

[0096]

[0097] Figure 12 This is a flowchart of solving the optimization sub-model II for large-scale offline topology optimization according to an embodiment of the present invention, which specifically includes the following sub-steps:

[0098] S61. Randomly initialize the first generation population. Generate N motor individuals (design parameter combinations) in the 22-dimensional design space through random functions, where N represents the number of samples, and try to make these individuals spread throughout the entire design space to facilitate subsequent optimization. Based on the parallel agent model built in S5, use the LightGBM model to calculate the optimization target P of the population individuals. r0s6t (x), P r0s12 (x)(OC1),P r0s12t (x)(OC2) and P t0s12t (x); Use the RF model to calculate the constraint target T of the population individuals OC1 、T OC2 .

[0099] S62, motor optimization target P obtained by S61 r0s6t (x), P r0s12 (x)(OC1),P r0s12t (x)(OC2) and P t0s12t (x) is used as individual fitness, and individuals within the population are sorted in a non-dominated hierarchical manner. The constrained non-dominated hierarchical sorting follows the following rules: 1) feasible solutions take precedence over infeasible solutions; 2) infeasible solutions are sorted from low to high according to the total violation of the two constraint objectives (CV); and 3) feasible solutions are sorted according to Pareto dominance. Based on the hierarchical sorting results, individuals are ranked, with lower non-dominated ranks receiving higher priority. Individuals with the same non-dominated rank are further ranked based on the degree of crowding between individuals, with greater distances between individuals receiving higher priority. Subsequently, the parent population and the equivalent subpopulation generated by the genetic operator are merged into a population of 2N individuals. Based on the aforementioned sorting rules, N individuals are selected to form the next generation population, maintaining a constant population size while improving overall quality and diversity.

[0100] S63. Evolutionarily update the N individuals obtained in S62, and generate a new generation of offspring through operations such as crossover, mutation, and selection.

[0101] S64: Determine whether the optimization has converged by checking whether the difference between the two iterations meets the set accuracy or whether the number of iterations has reached the maximum number of iterations. If not, merge the parent and child populations and repeat the non-dominated sorting and evolution operations in S61-S63. Otherwise, the optimization ends and the Pareto front solution set is obtained. Figure 13The iterative process of the air gap design domain topology is presented. In the first 20 generations of populations, the stator and rotor edge topology within the air gap design domain continuously changed at different locations, resulting in an irregular air gap between the stator and rotor. Simultaneously, due to the influence of the electromagnetic force target, the air gap spacing became larger, and the motor torque performance decreased. As the iterations progressed, the air gap design domain topology gradually stabilized from the 30th to the 50th generation. Constrained by the average torque constraint, the stator and rotor air gap gradually shrank and became fixed at the rotor edge, and the optimization process converged.

[0102] S65. Draw a scatter plot of the last five generations of samples against the target and mark the Pareto frontier as follows: Figure 14 As shown. Select appropriate sample points on the Pareto front as the final optimized motor, and output the corresponding two-dimensional Gaussian function optimal weight coefficient combination. By establishing a parallel proxy model instead of online finite element simulation to optimize the Gaussian function weight coefficient, the air gap design domain air gap permeability is changed, so that the noise optimization efficiency of the motor is greatly improved. Taking the simulation of a single working condition data set for 8 minutes as an example, the time to complete the large-scale optimization is only about 7.7 days (including model training time). Compared with relying entirely on online optimization, the optimization time is shortened by 72.2%.

[0103] The specific parameter settings of the NSGA-II algorithm corresponding to the above S61-S64 in the present invention are as follows: the population size is 50, the crossover probability is 0.9, the mutation probability is 1 / 6, and the maximum number of iterations is 100.

[0104] S7: The two-dimensional Gaussian basis function weight coefficient combination corresponding to the optimal motor air gap domain topology obtained in S6 is input into the Python script, and ANSYS Maxwell is called through the Python script to generate the corresponding air gap domain topology to obtain the final optimized motor complete topology.

[0105] S8: A multi-physics simulation model was constructed using ANSYS Maxwell and ANSYS Workbench to compare and evaluate the electromagnetic performance and vibration and noise characteristics of the above motor design under multiple operating conditions.

[0106] Figure 15 The electromagnetic performance of the motor before and after optimization of the embodiment of the present invention is compared, where (a) is the comparison of the air gap flux density waveform and its harmonics under the high-speed and light-load working condition OC2 of the motor before and after optimization. The 11th and 13th harmonics that originally rose due to the influence of the weak magnetic current Id are effectively suppressed, so that the characteristic radial electromagnetic force harmonic P under this working condition is r0s12tThe amplitude of the radial air gap magnetic flux fundamental wave has decreased, while the radial air gap magnetic flux fundamental wave has increased slightly compared with that before optimization, which effectively ensures the torque output of the optimized motor. (b) The left figure shows the torque waveform comparison, and the right figure is the corresponding Fourier transform result. By comparing the torque harmonics of the motor before and after optimization, it can be observed that the torque fundamental wave of the optimized motor is slightly improved compared with the initial motor, while the higher harmonics, especially the dominant 12th harmonic, are significantly reduced. Therefore, under two typical working conditions, the output torque is better than the initial motor, and the torque pulsation is significantly improved. Under low-speed and heavy-load conditions, the torque pulsation is reduced from the original 26% to 16%; under the high-speed and light-load conditions of the initial motor, the electromagnetic force harmonic P t0s12t For the optimization goal, the torque pulsation was significantly suppressed. Compared with the initial motor, the pulsation was reduced from 30% to 20%, a reduction of 33.33%. (c) The characteristic radial electromagnetic force P of the motor before and after optimization r0s12t Amplitude Map under all conditions. Under low speed and heavy load condition OC1, the initial motor P r0s12t is 4.42kPa. With the injection of weak magnetic current Id, P r0s12t The amplitude rises to 7.29kPa, and the motor with variable air gap topology optimization has a P value of 0. r0s12t is 3.35kPa, which is 24.2% lower than the initial motor. At the same time, P r0s12t The optimized Map shows that P in the high-speed weak magnetic area r0s12t The high-amplitude region greater than or equal to 6 kPa was reduced by 47.24% compared to before optimization, effectively suppressing the characteristic electromagnetic force. Table 3 details the comparison of the motor's electromagnetic performance before and after optimization.

[0107] Table 3

[0108] Design indicators Initial Motor Optimizing motors <![CDATA[P r0s6t_OC1 (kPa)]]> 6.84 3.28 <![CDATA[P r0s12t_OC1 (kPa)]]> 4.42 3.35 <![CDATA[P r0s12t_OC2 (kPa)]]> 7.29 3.17 <![CDATA[P t0s12t_OC2 (kPa)]]> 2.39 0.82 OC1 average torque (Nm) 115.11 118.05 OC2 average torque (Nm) 74.65 77.15 OC1 torque ripple (%) 26 16 OC2 torque ripple (%) 30 20

[0109] Figure 16 This is a comparison of the vibration characteristics of the motor before and after optimization of the embodiment of the present invention, where (a) (b) (c) are the comparison of the motor system displacement cloud maps at the zero-order 6th frequency and 12th frequency under the low-speed and heavy-load working condition OC1 and the zero-order 12th frequency under the high-speed and light-load working condition OC2 before and after optimization. Table 4 shows the comparison of vibration acceleration at the corresponding characteristic frequencies under two typical working conditions. It can be seen that under the low-speed and heavy-load and high-speed and light-load working conditions, the optimized motor obtained by the variable air gap domain topology optimization has better vibration acceleration and displacement than the initial motor. (d) The vibration equivalent radiation power level waterfall diagram under the low-speed and heavy-load working condition OC1 is given respectively. Due to the P under the low-speed and heavy-load working condition r0s6t and P r0s12tThe vibration of the initial motor at the 6th and 12th frequencies was effectively suppressed, and the equivalent radiated power at the 12th frequency under OC1 was reduced from the original 52.15dB to 45.68dB.

[0110] Table 4

[0111]

[0112] Figure 17 The noise characteristics of the motor before and after optimization of the embodiment of the present invention are compared. The noise distribution results of the motor before and after optimization on a 1m cylindrical model are given. It can be seen that the sound pressure level of the original structure at 12 times the frequency under low speed and heavy load OC1 is 71.069dB, and the sound pressure level of the motor after variable air gap domain topology optimization is 68.010dB; the sound pressure level of the two models at 12 times the frequency under high speed and light load OC2 is the highest, and the sound pressure level is reduced from the original 84.653dB to 75.203B, and the noise suppression effect is obvious.

[0113] Based on the above optimization design method, an embodiment of the present invention further includes a flat wire permanent magnet synchronous motor obtained by the above optimization design method.

[0114] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. Any equivalent methods or changes that do not deviate from the technology of the present invention should be included in the scope of protection of the present invention.

Claims

1. A topology optimization design method for a permanent magnet motor based on a variable air gap domain, characterized in that: Based on the stator and rotor topology of a permanent magnet motor, the air gap domain design concept is introduced. The motor topology is divided into the stator and rotor basic domain and the air gap design domain. This approach suppresses electromagnetic vibration noise under multiple motor operating conditions while maintaining basic electromagnetic performance. The process includes the following steps: Step 1, Initial Motor Design: Determine the motor design requirements and specifications based on the automotive drive permanent magnet motor application scenario. Based on the motor design requirements, select the motor and determine the basic motor structure. Then, based on the motor design specifications, determine the geometric design parameters of the motor stator and rotor and their optimization targets. Step 2: Optimization of the first-level stator and rotor basic domain parameters: Determine the sensitivity of the initial design parameters to the optimization target through local sensitivity analysis. Select the top 7 highly sensitive design parameters based on the comprehensive sensitivity. Combined with the motor constraints, construct the first-level optimization sub-model I. The first-level optimization sub-model I is solved based on the Kriging surrogate model and the NSGA-II genetic algorithm. The optimal design point is selected from the obtained non-dominated solution set as the final design parameters of the basic geometric structure of the motor stator and rotor to obtain the optimized motor I. Step 3: Obtain the radial and tangential electromagnetic force waveforms of the optimized motor I under rated working conditions by finite element analysis, and obtain the two sets of zero-order characteristic radial electromagnetic forces P with the highest amplitude based on two-dimensional Fourier decomposition. r0s6t 、P r0s12t and a set of zero-order characteristic tangential electromagnetic forces P with the highest amplitude t0s12t Finite element analysis was used to obtain the characteristic electromagnetic force variations under different loads. The drive motor operating conditions were divided into two typical operating conditions: low-speed, heavy-load OC1 and high-speed, light-load OC2. The characteristic electromagnetic forces under these two typical operating conditions were determined as the second-level optimization targets to suppress electromagnetic vibration noise issues under multiple operating conditions. Step 4, topology optimization of the second-level air gap design domain: Determine the motor air gap design domain, perform topology optimization based on the normalized Gaussian basis function grid within the design domain, and construct the second-level optimization sub-model II; Step 5: Solve the second-level optimization sub-model II: Call Maxwell through a Python script to perform online topology optimization and obtain 600 data sets. Use the 600 data sets to build a parallel agent model and perform large-scale offline topology optimization to obtain the Pareto solution set of the last generation of the population. Select the optimal solution and obtain the corresponding Gaussian basis function weight coefficient. Enter it into the Python script to generate the optimized motor II after air gap domain optimization. This motor is the final optimized motor.

2. The method for topology optimization design of a permanent magnet motor based on a variable air gap domain according to claim 1, characterized in that: In step S1, the basic structure of the motor adopts a 48-slot 8-pole hairpin winding permanent magnet flat wire motor, which adopts hairpin flat wire winding, is directly inserted from the end of the iron core and staggered and stacked. The stator slot adopts a rectangular slot structure, and the rotor structure adopts a double-layer V-shaped permanent magnet arrangement. A rectangular magnetic barrier is set between the second layer of V-shaped permanent magnets to reduce self-leakage magnetic flux and improve the utilization rate of the permanent magnets. In order to reduce the rotational inertia of the rotor, triangular and trapezoidal weight-reducing holes are designed near the shaft.

3. The method for topology optimization design of a permanent magnet motor based on a variable air gap domain according to claim 2, characterized in that: The geometric design parameters of the motor stator and rotor in step S1 specifically include: The distance between the inner and outer permanent magnets is h5, the height of the inner permanent magnet is h4, the width of the rectangular magnetic barrier is h1, the length of the inner permanent magnet is l1, the length of the outer permanent magnet is l2, the width of the inner permanent magnet is t1, the width of the outer permanent magnet is t2, the length of the rectangular magnetic barrier is w2, the width of the outer reinforcement rib is w1, the width of the inner permanent magnet magnetic bridge is w5, the width of the outer permanent magnet magnetic bridge is w4, the angle between the inner permanent magnet is a1, the angle between the outer permanent magnet is a2, the width of the stator rectangular slot is t3, the width of the stator slot is w3, the width of the stator tooth is h3, the length of the rectangular slot is h2, the inclination angle of the stator tooth is a3, and the outer radius of the stator is R. so , stator inner radius R si ; The target parameters to be optimized include: average output torque, torque ripple, iron loss and copper loss under rated working conditions of the motor.

4. The method for topology optimization design of a permanent magnet motor based on a variable air gap domain according to claim 3, characterized in that: The seven highly sensitive design parameters in step S2 include two stator structural parameters: stator rectangular slot width t3 and stator rectangular slot length h2; five rotor structural parameters: inner layer permanent magnet angle a1, outer layer permanent magnet angle a2, inner layer permanent magnet height h4, inner and outer layer permanent magnet spacing h5, and inner layer permanent magnet length l1; The motor restriction conditions include limiting the armature current density to 15A / mm 2 The width of the insulation layer in the stator slot is set to 1.5mm; the amount of a single NdFeB permanent magnet in the inner layer is 9945mm 3 ; The outer layer single NdFeB permanent magnet usage is 3705mm 3 ; Based on the target to be optimized, highly sensitive design parameters and constraints, the first-level optimization sub-model I is constructed as follows: x=[h2,t3,a1,a2,h4,h5,l1] Among them, T orqueRipple (x), T orque (x), C oreloss (x), C opperloss (x) are the torque ripple, average output torque, iron loss and copper loss under rated working conditions, respectively. rate is the current density under rated working conditions, V pm1 、V pm2 are the volumes of a single inner and outer permanent magnet, respectively.

5. The method for topology optimization design of a permanent magnet motor based on a variable air gap domain according to claim 4, characterized in that: The method for solving the first-level optimization sub-model I and obtaining the optimized motor I in step S2 is as follows: S21. Perform Latin hypercube sampling in the 7-dimensional design space of the highly sensitive parameters h2, t3, a1, a2, h4, h5, and l1 to obtain 1000 sample points. Calculate the average output torque, torque ripple, iron loss, and copper loss of all sample points at the rated operating point using ANSYS Maxwell. S22. Construct a Kriging proxy model based on the sample points obtained in S21; wherein, the root mean square error is used to evaluate the model accuracy. When the root mean square error is less than 1%, it is considered that the accuracy of the proxy model meets the subsequent optimization requirements; S23, randomly initialize the first generation population; generate N motor individuals through random functions in the 7-dimensional design space, and let these individuals spread throughout the entire design space to facilitate subsequent optimization; based on the Kriging proxy model constructed in S22, calculate the optimization target C of the population individuals in turn oreloss (x), C opperloss (x), T orqueRipple (x), T orque (x); S24, motor optimization target C obtained by S23 oreloss (x), C opperloss (x), T orqueRipple (x), T orque (x) As individual fitness, individuals in the population are sorted by non-dominance, and the individuals are ranked according to the sorting results, where individuals with lower non-dominance ranks have higher priorities; for individuals with the same non-dominance rank, further ranking is performed based on the crowding degree between individuals, where individuals with greater distance have higher priorities. Subsequently, the parent population and the equal-subgeneration population generated by the genetic operator are merged into a group containing 2N individuals. Based on the above sorting rules, N individuals are selected to form the next generation population to maintain a constant population size and improve overall quality and diversity. S25, perform evolutionary updates on the N individuals obtained in S24, and generate a new generation of offspring through crossover, mutation, and selection operations; S26. Determine whether the optimization has converged by judging whether the difference between the results of the two iterations meets the set accuracy or whether the number of iterations reaches the maximum number of iterations. If not, merge the parent and child populations and re-perform the non-dominated sorting and evolution operations of S23-S25. Otherwise, the optimization ends, the Pareto front solution set is obtained, and the optimal geometric parameter combination is selected to obtain an optimized motor with improved torque and efficiency. The specific parameter settings of the NSGA-II algorithm corresponding to S23-S26 are as follows: population size is 50, crossover probability is 0.9, mutation probability is 1 / 6, and the maximum number of iterations is 100.

6. The method for topology optimization design of a permanent magnet motor based on a variable air gap domain according to claim 5, characterized in that: The method for obtaining the characteristic electromagnetic force variation characteristics under different loads by the finite element analysis method in step S3 is as follows: First, the weak magnetic current I d , load current I q Full factor sampling was performed within the range, with each factor taking 10 levels to form 100 groups of working condition samples. The characteristic radial electromagnetic force P under each working condition was obtained by simulation using ANSYS Maxwell software. r0s6t and P r0s12t Amplitude size, the characteristic electromagnetic force amplitude distribution map within the full working range is constructed by the cubic spline interpolation method, P r0s6t It is greatly affected by the load current, and its amplitude increases with the q-axis current I q The increase of P r0s12t As the load current increases, it gradually increases. At the same time, as the weak magnetic current I d The injection of the electromagnetic force has an obvious increasing trend in the high-speed region, indicating that the electromagnetic force harmonics should also be effectively suppressed at high speed and weak magnetic field.

7. The method for topology optimization design of a permanent magnet motor based on a variable air gap domain according to claim 6, characterized in that: The optimization target of the second level in step S3 is specifically: radial electromagnetic force P r0s6t 、P r0s12t As the target of noise suppression in low speed and heavy load condition OC1; radial electromagnetic force P r0s12t and the tangential electromagnetic force P t0s12t As the target of noise and vibration suppression in high-speed and light-load operating condition OC2.

8. The method for topology optimization design of a permanent magnet motor based on a variable air gap domain according to claim 7, characterized in that: The implementation method of step S4 is as follows: The variable air gap permeability ΔΛ(ω0,ω1,...,ω 21 ) and topology optimization design parameters ω0,ω1,...,ω 21 The NSGA-II genetic algorithm is used to associate the characteristic electromagnetic force P of the low-speed and heavy-load working condition OC1 obtained in step S3. r0s6t and P r0s12t , Characteristic electromagnetic force P of high speed and light load condition OC2 r0s12t and P t0s12t Perform small-scale online optimization, extract a uniformly distributed sample data set (including 600 sets of design parameters and their corresponding optimization targets), and build the second-level optimization sub-model II: x=[ω0,ω 1, ...oh 21 ] Where T OC1ini 、T OC2ini are the average torques of the initial motor under two typical working conditions; T OC1OP 、T OC2OP are the average torques of the optimized motor under two typical working conditions; ω0, ω1, ..., ω 21 Corresponding to the weight coefficients of 22 Gaussian basis functions.

9. The method for topology optimization design of a permanent magnet motor based on a variable air gap domain according to claim 8, characterized in that: The implementation method of step S5 is as follows: S51. First, select the topology optimization design area. In order to reduce the number of iterations and improve the optimization efficiency, considering the symmetry of the initial motor, select 1 / 16 of the motor air gap domain for local optimization design. Perform a fine-grained subdivision operation on the design area with 100 equal divisions in the tangential direction and 30 equal divisions in the radial direction, dividing it into 3000 air gap microgrids of uniform size. The position of each air gap microgrid is determined by the corresponding position function n(x, y), where x is the tangential coordinate and y is the radial coordinate. The value of the microgrid affects the air gap permeance harmonic distribution of the entire air gap area. The specific calculation rules for the microgrid value are as follows: S511, place 22 two-dimensional Gaussian basis functions uniformly covering the entire design area at the radial 1 / 3 and 2 / 3 of the region along the tangential direction of the motor rotor, and calculate the 22 sets of values ​​G of each microgrid n(x, y) under the 22 two-dimensional Gaussian basis functions using formula (3): i (x,y)(i=0,1,2,…,21); S512, normalize the 22 groups of values ​​using formula (4), and give each normalized value b i (x,y) assigns the corresponding weight coefficient ω i ; S513, calculate the final weighted value M of the microgrid n(x,y) by formula (5) n(x,y) , that is, the value of each microgrid is obtained by multiplying the corresponding 22 normalized Gaussian basis function values ​​by the corresponding weight coefficient and summing them up; S514. Finally, the material properties are determined by comparing the value with the given reference value ξ. The rule is shown in formula (6): where x i and y i is the center coordinate of the i-th Gaussian basis function, σ is the standard deviation of the two-dimensional Gaussian basis function; When assigning different weight coefficients ω0, ω1, ..., ω to 22 Gaussian basis functions 21 When the weight coefficients are updated, the corresponding microgrid material properties will also change accordingly. The genetic algorithm NSGA-II is used to guide the iteration of the weight coefficients, so that the structure of the air gap topology is continuously optimized. The corresponding air gap permeance Λ(θ) is also continuously updated with the 22 weight coefficients to form a variable air gap permeability ΔΛ(ω0,ω1,...,ω 21 ); S52. Processing the generated Gaussian basis function weight coefficient combinations sequentially based on the Python script: assigning corresponding material properties to corresponding air gap microgrid numbers according to the material criterion, calling ANSYS Maxwell, assigning the same material properties to microgrids with the same numbers in the software, and realizing the generation of the motor air gap domain topology from the Gaussian basis function weight coefficients; S53, randomly initialize the first generation population, generate N motor individuals through random functions in the 22-dimensional design space, then generate sample motors through finite element simulation, and calculate the optimization target P of the population individuals in turn. r0s6t (x), P r0s12 (x)(OC1),P r0s12t (x)(OC2) and P t0s12t (x), constraint target T OC1 、T OC2 ; S54, the motor optimization target P obtained in S53 r0s6t (x), P r0s12 (x)(OC1),P r0s12t (x)(OC2) and P t0s12t (x) As individual fitness, the individuals in the population are sorted by non-dominated hierarchical order. The non-dominated hierarchical order considering constraints follows the following rules: 1) feasible solutions take precedence over infeasible solutions, 2) infeasible solutions are sorted from low to high according to the total violation degree CV of the two constraint targets, and 3) feasible solutions are sorted according to Pareto dominance. According to the hierarchical ordering results, the individuals are ranked, where individuals with lower non-dominated ranks have higher priorities. For individuals with the same non-dominated rank, they are further ranked according to the crowding degree between individuals, where individuals with greater distance have higher priorities. Subsequently, the parent population and the equal-quantum generation population generated by the genetic operator are merged into a group containing 2N individuals. Based on the above sorting rules, N individuals are selected to form the next generation population to keep the population size constant and improve the overall quality and diversity. S55, perform evolutionary updates on the N individuals obtained in S54, and generate a new generation of offspring through crossover, mutation, and selection operations; S56. Save all sample point data in the population, including 22 sets of design parameters and their corresponding 4 optimization objectives and 2 constraint objective data; S57. Determine whether the optimization has converged by judging whether the difference between the results of the two iterations meets the set accuracy or whether the number of iterations reaches the maximum number of iterations. If not, merge the parent and child populations and perform the non-dominated sorting and evolution operations of S53-S56 again; otherwise, the optimization ends. S58. After the optimization iteration is completed, the sample data sets in all populations are output. Each population contains 30 sets of 22-dimensional Gaussian function weight coefficients and their corresponding optimization objectives and constraint target data sets. The maximum number of iterations is 20, so there are a total of 600 sample data sets. Among them: The specific parameter settings of the NSGA-II algorithm corresponding to the above S53-S58 are as follows: population size is 30, crossover probability is 0.9, mutation probability is 1 / 6, and the maximum number of iterations is 20; S59, using the 600 sets of data sets obtained to build a parallel agent model, in which the optimization target uses the LightGBM method, the constraint target uses the random forest method, and the root mean square error RMSE and the determination coefficient R are used to calculate the optimal target. 2 To evaluate the model accuracy, when the root mean square error is less than 1% and the coefficient of determination is greater than 0.90, the surrogate model accuracy is considered to meet the subsequent optimization requirements. At the same time, in order to further verify the model prediction accuracy, 50 sets of design parameter combinations are randomly generated through LHS, and the prediction results are compared with the finite element simulation results. The maximum error of the prediction model is within 5%. S6: Using NSGA-II genetic algorithm, the characteristic electromagnetic force P of the low-speed and heavy-load condition OC1 is extracted. r0s6t and P r0s12t 、Characteristic electromagnetic force P of high speed and light load condition OC2 r0s12t and P t0s12t Perform large-scale offline multi-objective optimization; the selected design parameters are 22 Gaussian basis function weight coefficients. The large-scale offline topology optimization sub-model II is consistent with the optimization sub-model II in step 4: x=[ω0,ω 1, ...oh 21 ] The specific steps of solving the optimization sub-model II for large-scale offline topology optimization are as follows: S61, randomly initialize the first generation population; generate N motor individuals through random functions in the 22-dimensional design space, and use the LightGBM model based on the parallel agent model constructed in S59 to calculate the optimization target P of the population individuals. r0s6t 、P r0s12tOC1 、P r0s12tOC2 and P t0s12t ; Use the random forest RF model to calculate the constraint target T of the population individuals OC1 、T OC2 ; S62, motor optimization target P obtained by S61 r0s6t (x), P r0s12 (x)(OC1),P r0s12t (x)(OC2) and P t0s12t (x) As individual fitness, the individuals in the population are sorted by non-dominated hierarchical order. The non-dominated hierarchical order considering constraints follows the following rules: 1) feasible solutions take precedence over infeasible solutions, 2) infeasible solutions are sorted from low to high according to the total violation degree CV of the two constraint targets, and 3) feasible solutions are sorted according to Pareto dominance. According to the hierarchical ordering results, the individuals are ranked, where individuals with lower non-dominated ranks have higher priorities. For individuals with the same non-dominated rank, they are further ranked according to the crowding degree between individuals, where individuals with greater distance have higher priorities. Subsequently, the parent population and the equal-quantum generation population generated by the genetic operator are merged into a group containing 2N individuals. Based on the above sorting rules, N individuals are selected to form the next generation population to keep the population size constant and improve the overall quality and diversity. S63, perform evolutionary updates on the N individuals obtained in S62, and generate a new generation of offspring through crossover, mutation, and selection operations; S64: Determine whether the optimization has converged by judging whether the difference between the results of the two iterations meets the set accuracy or whether the number of iterations reaches the maximum number of iterations. If not, merge the parent and child populations and perform the non-dominated sorting and evolution operations of S61-S63 again; otherwise, the optimization ends and the Pareto front solution set is obtained. The specific parameter settings of the NSGA-II algorithm corresponding to S61-S64 are as follows: population size is 50, crossover probability is 0.9, mutation probability is 1 / 6, and the maximum number of iterations is 100; S65. Draw a scatter plot of the last five generations of samples corresponding to the target, mark the Pareto front, select appropriate sample points on the Pareto front as the final optimized motor, and output the corresponding two-dimensional Gaussian basis function optimal weight coefficient combination; S7: Input the two-dimensional Gaussian basis function weight coefficient combination corresponding to the final optimized motor air gap domain topology obtained in S65 into the Python script, call ANSYS Maxwell through the Python script to generate the corresponding air gap domain topology, and obtain the optimized motor II after gap domain optimization. This motor is the final optimized motor.

10. A permanent magnet motor for a vehicle, characterized in that: The motor is obtained by the permanent magnet motor topology level optimization design method based on variable air gap domain described in any one of claims 1-9.

Images