High-torque low-vibration built-in permanent magnet synchronous motor optimization method

Abstract

The invention relates to the field of optimization design of permanent magnet synchronous motors, in particular to a high-torque low-vibration built-in permanent magnet synchronous motor optimization method which comprises the following steps: S1, calculating motor electromagnetic field distribution through a finite element method, and determining a key noise source of a motor according to a calculation result; s2, defining an optimization target and constraint conditions, constructing an optimization mathematical model, performing sensitivity analysis on design variables by adopting a random forest algorithm, and dividing design parameters into a strong sensitive level, a middle sensitive level and a weak sensitive level; s3, constructing a multi-layer perceptron neural network agent model for the strong sensitive level parameters, and performing multi-objective optimization by adopting a second-generation non-dominated sorting genetic algorithm; s4, aiming at medium sensitive level parameters, adopting a response surface model for proxy modeling, and performing optimization in combination with a multi-target particle swarm optimization algorithm; s5, aiming at the weak sensitive level parameters, adopting a parameterized scanning method, combining weights of different targets through a weighted summation method, and comprehensively optimizing a target function; and S6, optimizing results are integrated and verified.

Description

Technical Field

[0001] This invention relates to the field of permanent magnet synchronous motor optimization design, specifically to an optimization method for a high-torque, low-vibration built-in permanent magnet synchronous motor. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs) are widely used in electric vehicles, rail transportation, and aerospace due to their high power density, high efficiency, and good speed regulation performance. Integrated permanent magnet synchronous motors (IPMSMs), with their excellent field weakening capability and high torque density, are particularly widely used in new energy vehicle drive systems. However, under complex operating conditions, motors generate wide-frequency electromagnetic excitation. When radial electromagnetic force harmonics interact with the natural frequencies of the motor structure, electromagnetic vibration and noise occur, severely affecting the motor's NVH performance and lifespan. Existing methods mostly focus on improving motor torque or efficiency, while noise optimization is largely limited to single structural improvements, such as skewed slots, skewed poles, or magnetic bridges, lacking systematic multi-objective optimization. Traditional finite element-multiphysics coupled optimization is computationally intensive and time-consuming, making it difficult to meet engineering requirements. Therefore, there is an urgent need for an optimization design method that can balance motor torque performance and electromagnetic noise performance while significantly improving computational efficiency. Summary of the Invention

[0003] The main objective of this invention is to address the shortcomings of existing motor design processes, such as low computational efficiency and the difficulty in simultaneously achieving high torque performance and low vibration noise. This invention provides an optimization method for a high-torque, low-vibration built-in permanent magnet synchronous motor. By combining finite element analysis, random forest sensitivity grading, multilayer perceptron (MLP) neural network surrogate model, response surface model, and multi-objective optimization algorithm, this method achieves efficient modeling and optimization solving of the motor's multi-objective optimization process, improves global optimization capability, reduces the risk of getting trapped in local optima, and ensures that the motor's output torque and noise performance meet design requirements, thereby improving the overall performance and reliability of the system. To achieve the above objectives, this invention discloses an optimization method for a high-torque, low-vibration built-in permanent magnet synchronous motor, comprising the following steps: S1 calculates the electromagnetic field distribution of the motor using the finite element method to obtain electromagnetic force harmonics at different operating points; then it calculates the stator mode frequency and structural response, combines the electromagnetic excitation of the finite element method with the unit force wave response function to establish a hybrid model of electromagnetic vibration and noise of the motor, and determines the key noise sources of the motor based on the calculation results. S2 defines the optimization objective and constraints, which is to optimize the average output torque of the motor. Torque pulsation and key electromagnetic harmonic amplitude As the optimization objective, the set of design variables is defined as X, including stator slot size, slot width and height, permanent magnet thickness and width, and magnetic bridge thickness. An optimization mathematical model is constructed, and the random forest algorithm is used to perform sensitivity analysis on the design variables. Based on the analysis results, the design parameters are divided into strong sensitivity level, medium sensitivity level, and weak sensitivity level. S3 constructs a multilayer perceptron neural network surrogate model for highly sensitive parameters and employs a second-generation non-dominated sorting genetic algorithm for multi-objective optimization; S4 employs a response surface methodology for proxy modeling of medium-sensitivity parameters, combined with a multi-objective particle swarm optimization algorithm for optimization. For parameters with weak sensitivity, S5 employs a parametric scanning method and calculates the comprehensive optimization objective function by combining the weights of different objectives through a weighted summation method.

[0004] S6 comprehensive optimization results and verification.

[0005] Preferably, the method for calculating the stator modal frequency and structural response in step S1 is as follows: linear stator structure of the motor The response formula for a unit force wave is as follows:

[0006] in, The force wave order is unit force wave, The frequency response function of the prototype stator structure is... The r-th order unit force wave response of the prototype stator structure. The stator mode frequency.

[0007] Preferably, the optimized mathematical model constructed in step S2 is as follows: ,in Indicates the output torque. This indicates motor torque pulsation. Indicates the key order of electromagnetic force. This indicates that the primary objective of optimization is to maximize the average torque. Represents a vector consisting of the minimized objective. Its constraints are: .

[0008] Preferably, the specific steps for performing sensitivity analysis on the design variables using the random forest algorithm in step S2 are as follows: (1) Definition Let be the accuracy of the j-th prediction. To improve accuracy after shuffling the d-th feature, The importance score for feature d is given by the following evaluation formula: ), where M represents the total number of predictions; (2) Divide the design parameters into strong sensitivity level: permanent magnet thickness, permanent magnet width, slot width and height, medium sensitivity level: slot height, magnetic bridge thickness, slot bottom radius and weak sensitivity level: such as slot width, slot core width.

[0009] Preferably, the comprehensive optimization objective function in step S5 is as follows:

[0010] The optimal solution is obtained by scanning.

[0011] This invention achieves the following beneficial effects through the above technical solution: 1. Significantly improved computational efficiency: By combining parameter sensitivity grading with a multi-layer surrogate model, the computational complexity of multiphysics field optimization design for motors is effectively reduced. Specifically, optimization of highly sensitive parameters significantly improves torque performance, while optimization of medium and weakly sensitive parameters plays a crucial role in improving noise performance and shortening computation time. 2. Significant optimization effects: Under the premise of ensuring structural stability, this invention achieves maximization of output torque, effective suppression of torque ripple, and significant reduction of electromagnetic noise, especially excelling in torque ripple and noise amplitude control. In summary, this invention proposes a motor optimization design method based on sensitivity grading and a multi-layer surrogate model, breaking through the limitations of traditional optimization techniques and providing new ideas and practical applications for the design of high-performance motors, possessing both high academic value and promising engineering prospects. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0013] Figure 1 It is a flowchart for multi-objective optimization.

[0014] Figure 2 This is a schematic diagram of the stator slot optimization parameters, including slot opening height, shoulder height, slot width height, slot opening width, slot center width, slot bottom width, and slot bottom radius.

[0015] Figure 3 This is a schematic diagram of rotor structure optimization parameters, including the width of the magnetic bridge, the distance between the magnet shafts, the thickness of the rotor magnets, and the width of the rotor magnets.

[0016] Figure 4 This is an analysis chart of the results of the structural parameter sensitivity calculation.

[0017] Figure 5This is a comparison of torque optimization effects at different stages. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example Refer to the instruction manual. Figure 1 This invention provides an optimization method for a high-torque, low-vibration built-in permanent magnet synchronous motor. The specific implementation steps of this invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0020] As per the instruction manual Figure 2-3 The preliminary optimization parameters selected in this invention include slot height, shoulder height, slot width-to-height, slot width, slot center width, slot bottom width, slot bottom radius, magnetic bridge width, magnet shaft distance, rotor magnet thickness, and rotor magnet width. Specific initial values ​​and ranges for the motor structural parameters are shown in Table 1.

[0021] Table 1 shows the initial values ​​and ranges of structural parameters for the example motor:

[0022] Figure 1 The following is the optimization flowchart proposed in this invention. To ensure the optimization of motor performance, a multi-layered surrogate model and a sensitivity hierarchical optimization strategy are adopted during the implementation process. The detailed steps are as follows: This invention first uses the finite element method (FEM) to electromagnetically model the motor, calculate the electromagnetic field distribution, and identify electromagnetic force waves under various operating conditions. Based on this, the natural modal frequencies and vibration response of the stator structure are calculated using the unit force wave response analysis (ULRA) method. Combining these two methods forms a coupled response model of the motor's electromagnetic force and vibration. This hybrid modeling method effectively identifies key electromagnetic force harmonics generated during motor operation, particularly the 4th-order 14th harmonic radial electromagnetic force that interacts with the stator modal frequencies and causes noise. This modeling approach provides fundamental data for subsequent optimization.

[0023] Based on hybrid modeling, this invention defines the optimization objective of the motor and determines the constraints. The optimization objective of the motor includes average torque. Torque pulsation Electromagnetic force harmonic amplitude ,in:

[0024] To reduce the complexity of optimization calculations, this invention employs the random forest algorithm to perform sensitivity analysis on multiple design parameters of the motor. The random forest algorithm can calculate the degree of influence of each parameter on the optimization objective, defining... Let be the sensitivity score of the Dth feature. The prediction accuracy for the original dataset. Let M be the prediction accuracy after the i-th feature is shuffled, and M be the number of decision trees. Specifically, Random Forest calculates the influence score of each feature by integrating multiple decision trees using the following formula:

[0025] Based on the instruction manual Figure 4 Sensitivity analysis results selected slot width and height, rotor magnet thickness, and rotor magnet width as highly sensitive parameters. Slot opening height, slot bottom radius, and magnetic bridge width were selected as medium-sensitive parameters. Slot center width and slot bottom width were selected as low-sensitive parameters. Since the influence of slot bottom width, slot width and height, and magnet shaft distance on all objective functions does not exceed 5%, these three structural parameters are considered irrelevant to the optimization objective and are filtered out in subsequent calculations. The specific hierarchical optimization steps are as follows: Step S1: For highly sensitive parameters, this invention establishes a multilayer perceptron (MLP) neural network surrogate model and combines it with a non-dominated sorting genetic algorithm (NSGA II) for global optimization, as follows: S1.1, MLP surrogate model construction and training: Based on the selected parameters, a multilayer perceptron (MLP) neural network surrogate model is constructed, containing an input layer, multiple hidden layers, and dual output layers (corresponding to average torque and torque ripple, respectively). Its inputs are 3 slot widths and heights, 10 rotor magnet thicknesses, and 11 rotor magnet widths; the output is the average output torque. Torque pulsation and key electromagnetic harmonic amplitude The training process aims to minimize the mean squared error loss function. :

[0026] In the formula, These are the weighting coefficients for each objective item. , and represent the true value and predicted value of the i-th sample, respectively.

[0027] S1.2 is a multi-objective optimization based on NSGA-II, using an MLP surrogate model as the objective function evaluator. The optimization objectives are set as maximizing average torque and minimizing torque ripple. The NSGA-II algorithm population is initialized. During iteration, the surrogate model is used to quickly predict the objective values ​​of individuals in the population, performing non-dominated sorting, crowding calculation and selection, crossover, and mutation operations. An elite retention strategy is used to generate a new generation of the population. The process iterates until the maximum number of generations is reached, outputting the Pareto optimal solution set.

[0028] S1.3, Result Verification and Decision-Making: Key solutions are selected from the Pareto solution set and accurately verified through finite element analysis. If the error between the prediction and calculation results is less than 2%, the optimization is confirmed to be effective. The designer selects the final parameter combination from the verified solutions based on performance preferences, fixing it as the highly sensitive optimization result.

[0029] Step S2: For the medium-sensitivity parameters, this invention uses Response Surface Modeling (RSM) and MOPSO for optimization, as follows: S2.1, Response Surface Model Establishment: A quadratic polynomial response surface model is established for each performance index. This model form can effectively characterize the nonlinear relationship and interaction between parameters and performance. Taking average torque as an example, its complete quadratic model form is as follows:

[0030] In the formula, For constant terms, The coefficient of the linear term, The coefficient of the quadratic term, For the interaction term coefficient.

[0031] S2.2, Model Validation and Improvement: First, calculate the coefficient of determination. and adjusting the coefficient of determination ,Require To further verify the model's predictive ability, several additional sets of verification sample points can be randomly generated, and finite element simulation and RSM model prediction can be performed respectively, with their relative errors calculated. If the average relative error is less than 5%, the model accuracy is considered to meet the optimization requirements.

[0032] S2.3, Optimal Solution Decision and Verification. A final solution is selected from the Pareto optimal solution set. For example, if the application scenario has extremely high requirements for torque smoothness, then a preferred solution can be selected. The minimum acceptable solution was selected. Finally, the selected optimal parameter combination was substituted into the finite element model for accurate simulation verification to confirm that its performance indicators were consistent with the prediction results of the response surface model and met all design constraints.

[0033] The optimization results at this stage show that, based on the optimization of the highly sensitive level, the average torque of the motor has been further increased by 5.6%, and the torque ripple has been reduced by 16.8%.

[0034] Step S3: For weakly sensitive parameters, this invention employs a parametric scan and weighted summation method for optimization. The optimization result is obtained by scanning the value range of all weakly sensitive parameters and then performing a weighted summation. A parametric scan is performed on Bs0, and then the three objective functions are normalized. The normalized function... , , We summed the values ​​with weights of 0.4, 0.2, and -0.4 respectively. In this optimization step, the motor torque ripple was further reduced by 28.3%, and the average torque was further increased by 1.7%.

[0035] The results of the three-stage structural parameter optimization are shown in Table 2.

[0036] Table 2 Initial and Optimal Structural Parameters

[0037] The optimization results of the three stages are shown in Table 3.

[0038] Table 3. Optimization results at each stage of hierarchical optimization

[0039] As shown in Table 3, the average torque of the motor was increased by 25% and the torque ripple was reduced by 55.7% after optimization using the non-dominated sorting genetic algorithm (NSGAⅡ).

[0040] In summary, as per the instruction manual appendix Figure 5 As shown in Table 3, compared to the initial design, the average torque of the motor after hierarchical optimization adopted in this invention is increased by 34.6%, torque ripple is reduced by 67.8%, and electromagnetic force harmonic amplitude is reduced by 13.7%. In addition, the calculation time is reduced from 4176.5 minutes to 1881.9 minutes, and the optimization efficiency is improved by approximately 54.9%.

[0041] S4 employs a response surface methodology for proxy modeling of medium-sensitivity parameters, combined with a multi-objective particle swarm optimization algorithm for optimization. S5 employs a parametric scanning method for weakly sensitive parameters and calculates the comprehensive optimization objective function by combining the weights of different objectives through a weighted summation method. The comprehensive optimization objective function in step S5 is as follows:

[0042] The optimal solution is obtained by scanning, enabling further optimization based on the results of the medium-sensitivity optimization.

[0043] S6 comprehensive optimization results and verification.

[0044] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An optimization method for a high-torque, low-vibration built-in permanent magnet synchronous motor, characterized in that, Includes the following steps: S1 calculates the electromagnetic field distribution of the motor using the finite element method to obtain electromagnetic force harmonics at different operating points; then it calculates the stator mode frequency and structural response, combines the electromagnetic excitation of the finite element method with the unit force wave response function to establish a hybrid model of electromagnetic vibration and noise of the motor, and determines the key noise sources of the motor based on the calculation results. S2 defines the optimization objective and constraints, which is to optimize the average output torque of the motor. Torque pulsation and key electromagnetic harmonic amplitude As the optimization objective, the set of design variables is defined as X, including stator slot size, slot width and height, permanent magnet thickness and width, and magnetic bridge thickness. An optimization mathematical model is constructed, and the random forest algorithm is used to perform sensitivity analysis on the design variables. Based on the analysis results, the design parameters are divided into strong sensitivity level, medium sensitivity level, and weak sensitivity level. S3 constructs a multilayer perceptron neural network surrogate model for highly sensitive parameters and employs a second-generation non-dominated sorting genetic algorithm for multi-objective optimization; S4 employs a response surface methodology for proxy modeling of medium-sensitivity parameters, combined with a multi-objective particle swarm optimization algorithm for optimization. S5 employs a parametric scanning method for weakly sensitive parameters and calculates the comprehensive optimization objective function by combining the weights of different objectives through a weighted summation method. S6 comprehensive optimization results and verification.

2. The optimization method for a high-torque, low-vibration built-in permanent magnet synchronous motor according to claim 1, characterized in that, The method for calculating the stator modal frequency and structural response in step S1 is as follows: linear stator structure of the motor The response formula for a unit force wave is as follows: in, The force wave order is unit force wave, The frequency response function of the prototype stator structure is... The r-th order unit force wave response of the prototype stator structure. The stator mode frequency.

3. The optimization method for a high-torque, low-vibration built-in permanent magnet synchronous motor according to claim 1, characterized in that, The optimized mathematical model constructed in step S2 is as follows: ， in Indicates the output torque. This indicates motor torque pulsation. Indicates the key order of electromagnetic force. This indicates that the primary objective of optimization is to maximize the average torque. Represents a vector consisting of the minimized objective. Its constraints are: .

4. The optimization method for a high-torque, low-vibration built-in permanent magnet synchronous motor according to claim 1, characterized in that, The specific steps for sensitivity analysis of the design variables using the random forest algorithm in step S2 are as follows: definition Let be the accuracy of the j-th prediction. To improve accuracy after shuffling the d-th feature, The importance score for feature d is given by the following evaluation formula: ), where M represents the total number of predictions; The design parameters are divided into four levels: strong sensitivity level: permanent magnet thickness, permanent magnet width, and slot width and height; medium sensitivity level: slot opening height, magnetic bridge thickness, and slot bottom radius; and weak sensitivity level: slot opening width and slot center width.

5. The optimization method for a high-torque, low-vibration built-in permanent magnet synchronous motor according to claim 1, characterized in that, The comprehensive optimization objective function in step S5 is as follows: The optimal solution is obtained by scanning.

Images