Motor parameter prediction and optimization method based on finite element simulation and neural network
By combining finite element simulation and BP neural network, the efficient, accurate prediction and optimization of motor parameters are achieved, and the problems of low accuracy and low efficiency in traditional methods are solved, which improves the performance and efficiency of motor design.
Patent Information
- Application Number
- CN202510357406.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-25
- Publication Date
- 2025-08-01
AI Technical Summary
In traditional motor design, the parameter calculation accuracy and low efficiency are difficult to meet the needs of high-performance motors. In particular, the impact of stator notch width on cogging torque, no-load back potential and load output torque is difficult to accurately predict and optimize.
Combining finite element simulation and BP neural network, through data acquisition and preprocessing, a BP neural network model is constructed, and the notch width is optimized by genetic algorithms to achieve efficient prediction and optimization of motor performance parameters.
It improves the accuracy and efficiency of motor parameter prediction, reduces design costs, provides efficient and reliable optimization solutions, and meets the high-performance needs of motor design.
Smart Images
Figure CN120409091A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of motor design and optimization, and in particular to a motor parameter prediction and optimization method based on finite element simulation and neural network. Background Art
[0002] Currently, traditional motor design relies primarily on two methods for parameter calculation: formulas and finite element simulation. Formulas, due to their simplified assumptions, suffer from low accuracy in practical applications, making them inadequate for meeting the demands of high-performance motors. While finite element simulation offers high accuracy, it suffers from low efficiency and time consumption.
[0003] In motor design, stator slot width is a key parameter that directly affects multiple performance indicators of the motor, including cogging torque, no-load back EMF, and loaded output torque.
[0004] Cogging torque is an important indicator to measure the smoothness of motor operation. Larger cogging torque will cause torque fluctuations during motor operation, affecting the overall performance.
[0005] The no-load back EMF reflects the power generation capacity of the motor and is of great significance for optimizing the motor performance.
[0006] Load output torque is a key parameter for evaluating the output capacity of a motor, and directly affects the working efficiency and load capacity of the motor.
[0007] Traditional design methods for analyzing the impact of slot width on these performance indicators typically require extensive simulation and trial calculations, resulting in low efficiency and high costs. Furthermore, the complex relationship between slot width and other performance indicators makes accurate prediction and optimization difficult using simple formulas or empirical formulas. Therefore, balancing calculation accuracy and efficiency is crucial in motor optimization design. Summary of the Invention
[0008] In order to solve at least one of the above technical problems, a motor optimization method that takes into account both the calculation accuracy and efficiency of motor parameters is developed. This application provides a motor parameter prediction and optimization method based on finite element simulation and neural network.
[0009] On the one hand, the present application provides a motor parameter prediction and optimization method based on finite element simulation and neural network, comprising the following steps:
[0010] S1 data acquisition and finite element simulation
[0011] Model the permanent magnet synchronous motor, set the inherent structural parameters of the motor, and obtain a permanent magnet synchronous motor simulation model; then perform parameter setting and simulation operation, data storage and preprocessing;
[0012] Construction and Training of S2 BP Neural Network Model
[0013] Design the BP neural network structure, determine that the number of nodes in the input layer is 1, corresponding to the slot width; set the number of nodes in the hidden layer; set the number of nodes in the output layer to 3, corresponding to cogging torque, no-load back electromotive force, and load output torque respectively; and perform network initialization and training, model verification and optimization, and test set verification;
[0014] S3 Optimize Slot Width
[0015] According to the design goal, define the objective function and select the optimization algorithm; initialize the search range of the slot width parameters; use the BP neural network model to predict the candidate slot widths and calculate the objective function values; update the slot width parameters according to the rules of the optimization algorithm to find the optimal solution; and output the optimized best slot width and the corresponding cogging torque, no-load back electromotive force, and load output torque;
[0016] S4 Result Verification and Application
[0017] Carry out simulation verification and experimental verification, and analyze and evaluate the optimization results.
[0018] Optionally, in S1, the parameter setting and simulation operation are as follows: set the slot width as a variable parameter, and keep other motor parameters unchanged; perform multiple groups of simulation calculations for different slot widths to obtain the corresponding cogging torque, no-load back electromotive force, and load output torque data;
[0019] The data storage and preprocessing are as follows: normalize the obtained data and divide it into a training set, a validation set, and a test set.
[0020] Optionally, in S2, activation functions are selected for the hidden layer and the output layer; set the loss function as mean square error or root mean square error to measure the difference between the predicted value and the true value;
[0021] Optionally, in S2, the network initialization and training are as follows: randomly initialize the network weights and biases; use the backpropagation algorithm for network training, adjust the weights and biases to minimize the loss function; set the training parameters to ensure network convergence;
[0022] The model verification and optimization are as follows: use the validation set to monitor the training process; adjust the network structure and training parameters to optimize the model performance;
[0023] The test set verification is as follows: verify the prediction accuracy and generalization ability of the trained BP neural network model on the test set.
[0024] Optionally, in S3, the objective function is defined as follows,
[0025] Minimize cogging torque: f(w) = CCM(w);
[0026] Maximize no-load back electromotive force: f(w) = EMF(w);
[0027] Maximize load output torque: f(w) = TORQUE(w);
[0028] Multi-objective optimization: f(w) = \alphaEMF(w) + \betaCCM(w) + \gammaTORQUE(w).
[0029] Optionally, in the step S3, the optimization algorithm adopts a genetic algorithm.
[0030] Optionally, in the step S4, the simulation verification is: substituting the optimal slot width into the finite element simulation model to verify the consistency between the predicted result and the simulation result;
[0031] The experimental verification is: fabricating a motor sample with the optimal slot width and conducting actual tests to verify the optimization effect;
[0032] The analysis of the optimization result is: comparing the performance parameters before and after optimization to evaluate whether the optimization effect meets the design objectives; analyzing the influence of the optimized slot width on the motor performance to further optimize the design.
[0033] In a second aspect, the present application provides an application of the above-mentioned motor parameter prediction and optimization method based on finite element simulation and neural network in the field of motor design.
[0034] In summary, the present invention utilizes finite element simulation technology to obtain data of cogging torque, no-load back electromotive force, and load output torque at different slot widths, and constructs a data set; on this basis, uses a BP neural network to learn and train the data set, and establishes a non-linear mapping relationship model between the slot width and the motor performance parameters. On the basis of realizing prediction, combined with an optimization algorithm, quickly find the optimal slot width to meet the design objectives (such as minimizing cogging torque, maximizing no-load back electromotive force or load output torque). This method effectively combines the advantages of finite element simulation and machine learning, improves the accuracy and efficiency of parameter prediction, and provides an efficient and reliable optimization scheme for motor design. Description of the Drawings
[0035] Figure 1 It is the Ansys electronics simulation model of the permanent magnet synchronous motor in the embodiment;
[0036] Figure 2 It is the multi-group cogging torque data diagram of the permanent magnet synchronous motor in the embodiment;
[0037] Figure 3 It is the structure diagram of the BP neural network in the embodiment;
[0038] Figure 4 The error histogram for analyzing the training results in the embodiment;
[0039] Figure 5 The mean squared error graph for analyzing the training results in the embodiment;
[0040] Figure 6 The regression result graph for the training in the embodiment;
[0041] Figure 7 The graph of multiple sets of no-load back electromotive force data of the permanent magnet synchronous motor in the embodiment;
[0042] Figure 8 The graph of multiple sets of load output torque data of the permanent magnet synchronous motor in the embodiment;
[0043] Figure 9 The interface graph for predicting the motor parameters of the permanent magnet synchronous motor in the embodiment;
[0044] Figure 10 The interface graph for optimizing the motor parameters of the permanent magnet synchronous motor in the embodiment;
[0045] Figure 11 The flow chart of the embodiment. Detailed implementation manners
[0046] The method for predicting and optimizing motor parameters based on finite element simulation and neural network is an innovative method that combines finite element simulation technology and artificial intelligence large model technology. At present, the motor parameters obtained by the ordinary formula method have the problem of low calculation accuracy, while the motor parameters obtained by using finite element simulation face problems such as low efficiency and long time consumption. The technical solution of this application uses machine learning technology, through a large number of motor optimization parameter trainings, deeply learns the operation laws and performance characteristics of the motor, so as to realize the accurate prediction and optimization of motor parameters, taking into account the accuracy and efficiency of parameter calculation.
[0047] This method can predict the key parameters and optimization schemes in motor design, combine factors such as motor type, structure, material, and thermal characteristics, identify potential problems in advance, realize the optimal design of the motor, and reduce the trial-and-error cost. It has the advantages of high efficiency, high operation accuracy, and reduced production cost, can provide reliable design guidance in the industrial field, and bring new ideas and methods to motor design and manufacturing. Generally speaking, the method for predicting and optimizing motor parameters based on finite element simulation and neural network provides an efficient and accurate method for calculating the equipment parameters of motors, which helps to promote the innovation and development of motor technology.
[0048] The following further describes this application in detail with reference to the drawings and embodiments. Specific embodiments
[0050] In this embodiment, a BP neural network is used to explore the relationship between the stator slot opening width and the cogging torque, realizing single-variable and single-objective prediction. On this basis, multi-variable and single-objective, multi-variable and multi-objective predictions are realized to optimize the motor parameters and improve the motor efficiency.
[0051] Therefore, first, we need to perform finite element simulation on the permanent magnet synchronous motor to obtain the relationship data between the stator slot opening width and the cogging torque.
[0052] The cogging torque is the torque generated by the interaction between the permanent magnet and the iron core when the windings of the permanent magnet motor are not energized, and is caused by the tangential component of the interaction force between the permanent magnet and the armature teeth. When there is relative movement between the stator and the rotor, the magnetic conductance between the armature teeth in the pole arc part of the permanent magnet remains basically unchanged, so the magnetic field around these armature teeth also remains basically unchanged. However, in a small area composed of one or two armature teeth corresponding to both sides of the permanent magnet, the magnetic conductance changes greatly, causing a change in the magnetic field energy storage, thereby generating the cogging torque. The cogging torque is thus defined as the negative derivative of the magnetic field energy W of the motor when it is not energized with respect to the relative position angle α between the stator and the rotor, where α is the angle between the center line of a specified tooth and the center line of a specified permanent magnet pole:
[0053]
[0054] Now make the following assumptions:
[0055] (1) The magnetic permeability of the iron core is infinite, i.e., μ0 = ∞;
[0056] (2) Ignore magnetic leakage and edge effects;
[0057] (3) The magnetic properties of the motor are the same and evenly distributed.
[0058] The expression of the cogging torque can be derived:
[0059]
[0060] L α is the axial length of the motor; R r and R s are the inner diameter of the motor stator and the outer diameter of the rotor; i f is the excitation current of the motor; G n and B m NL are the Fourier decomposition coefficients of the air-gap flux density function.
[0061] From this formula, it can be seen that there are many factors affecting the cogging torque of the motor. To reduce the cogging torque, the following three aspects can be considered:
[0062] (1) Change the pole parameters
[0063] The method of changing the magnetic pole parameters is to change the amplitude of Bm which plays a major role in the cogging torque, so as to weaken the cogging torque. Such methods mainly include changing the pole arc coefficient of the magnetic pole, using permanent magnets with unequal thicknesses, magnetic pole offset, skewed poles, combination of unequal pole arc coefficients, etc.
[0064] (2) Changing the armature parameters
[0065] Changing the armature parameters can change the amplitude of Gn which plays a major role in the cogging torque, thereby weakening the cogging torque. Such methods mainly include changing the slot opening width, changing the shape of the teeth, unequal slot opening widths, skewed slots, opening auxiliary slots, etc.
[0066] (3) Reasonably selecting the number of armature slots and poles
[0067] By reasonably selecting the number of armature slots and poles, the order and magnitude of Bm and Gn which play a major role in the cogging torque are changed, thereby weakening the cogging torque.
[0068] In this embodiment, the cogging torque of the motor is changed by changing the stator slot opening width of the motor, and the relationship between the stator slot opening width and the cogging torque is explored.
[0069] Taking a permanent magnet synchronous motor as an example below, Table 1 shows the main parameters of the motor; the Ansys electronics simulation model is as Figure 1 shown.
[0070] Table 1
[0071]
[0072] The stator slot opening width Bs0 is set as a variable. By changing different stator slot opening widths, different cogging torques are obtained, and a large dataset of the relationship between the stator slot opening width and the cogging torque is obtained. Multiple groups of cogging torque data graphs are as Figure 2 shown.
[0073] Model training
[0074] In this embodiment, a bp neural network is used to explore the relationship between the stator slot opening width and the cogging torque, realize single-variable single-object prediction, and on this basis, realize multi-variable single-object and multi-variable multi-object predictions, optimize the motor parameters, and improve the motor efficiency. Therefore, on the basis of obtaining the dataset, theoretical learning and practical training of the bp neural network are carried out.
[0075] 1. Definition of bp neural network
[0076] The BP neural network is a multi-layer feedforward neural network based on the error backpropagation algorithm. It is a typical supervised learning model used to handle tasks such as classification and regression. The core idea of the BP neural network is to optimize the network parameters through the gradient descent method to minimize the error between the network output and the target value.
[0077] The BP neural network consists of an input layer, a hidden layer, and an output layer, and learns and optimizes through two processes: forward propagation and backpropagation, as Figure 3 shown in the structure diagram of the BP neural network.
[0078] 2. Training the model using the dataset
[0079] First, we summarize and organize the dataset obtained in the previous stage, as shown in Table 2. Then, we apply 200 groups of data to the neural network model for multiple rounds of training, and finally explore whether the expected results can be obtained, that is, whether the BP neural network can accurately predict the relationship between the stator slot opening width and the cogging torque based on the training.
[0080] Table 2
[0081]
[0082]
[0083] Apply the dataset to the neural network model for multiple rounds of training, analyze the training results, and the error histogram is as Figure 4 shown, and the mean squared error graph is as Figure 5 shown.
[0084] From Figure 4 it can be seen that the peak of the error histogram is relatively steep, and the errors are concentrated in a relatively small range, indicating that the prediction results of the model are relatively accurate and the errors are small. From Figure 5 it can be seen that the mean squared error is small. The smaller the mean squared error, the smaller the difference between the prediction of the model and the true value, and the higher the prediction accuracy of the model. 1.1663e-06 is a very small value, indicating that the model has a very good fitting effect on the training data and the errors can be almost ignored.
[0085] Then analyze the regression results of the training; the regression result graph is as Figure 6 shown, and it can be found that the fitting effects of both the training set and the test set are good, and the gap between the actual value of the target and the predicted value of the output is small. Therefore, we can consider that this training is relatively successful, and we can predict the relationship between the stator slot opening width and the cogging torque through this model.
[0086] Based on the realization of single-variable and single-objective prediction of the slot width and cogging torque, in order to achieve multi-objective prediction and coordinated optimization of the slot width, we continue to perform finite element simulation on the permanent magnet synchronous motor to obtain the relationship data between the stator slot width, no-load back electromotive force, and load output torque, such as Figure 7 , Figure 8 shown.
[0087] After summarizing and organizing the data set, it is applied to the BP neural network model for multiple rounds of training. By adjusting the parameters, the prediction results are made to conform to the actual model, realizing the single-objective and multi-variable prediction of the slot width for cogging torque, no-load back electromotive force, and load output torque. Visualize the final prediction results to form the motor parameter prediction interface diagram of the permanent magnet synchronous motor, such as Figure 9 shown.
[0088] To achieve the optimization of minimizing the cogging torque, maximizing the no-load back electromotive force, and load output torque at a certain slot width, we introduce the genetic algorithm. By establishing a multi-objective optimization function, the optimal slot width is found to form the motor parameter optimization interface diagram of the permanent magnet synchronous motor, such as Figure 10 shown.
[0089] The above are all preferred embodiments of this application, and the protection scope of this application is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of this application should be covered within the protection scope of this application.
Claims
1. A method for predicting and optimizing motor parameters based on finite element simulation and neural network, characterized in that It includes the following steps: S1 Data acquisition and finite element simulation Model the permanent magnet synchronous motor, set the inherent structure parameters of the motor, and obtain the permanent magnet synchronous motor simulation model; And perform parameter setting, simulation operation, data storage and preprocessing; S2 BP neural network model construction and training Design the BP neural network structure, determine that the number of input layer nodes is 1, corresponding to the slot opening width; Set the number of hidden layer nodes; set the number of output layer nodes to 3, corresponding to cogging torque, no-load back electromotive force and load output torque respectively; and perform network initialization and training, model verification and optimization, and test set verification; S3 Optimize the slot opening width According to the design goal, define the objective function and select the optimization algorithm; initialize the search range of the slot opening width parameters; use the BP neural network model to predict the candidate slot opening widths and calculate the objective function values; according to the rules of the optimization algorithm, update the slot opening width parameters to find the optimal solution; and output the optimized best slot opening width and the corresponding cogging torque, no-load back electromotive force and load output torque; S4 Result verification and application Perform simulation verification and experimental verification, and analyze and evaluate the optimization results.
2. The method for predicting and optimizing motor parameters based on finite element simulation and neural network according to claim 1, wherein In the above S1, the parameter setting and simulation operation are as follows: set the slot opening width as a variable parameter and keep other motor parameters unchanged; perform multiple groups of simulation calculations for different slot opening widths to obtain the corresponding cogging torque, no-load back electromotive force and load output torque data; The data storage and preprocessing are as follows: normalize the obtained data and divide it into a training set, a validation set and a test set.
3. The motor parameter prediction and optimization method based on finite element simulation and neural network according to claim 1, wherein, In the above S2, activation functions are selected for the hidden layer and the output layer; set the loss function as mean square error or root mean square error to measure the difference between the predicted value and the true value.
4. The motor parameter prediction and optimization method based on finite element simulation and neural network according to claim 1, characterized in that In the above S2, the network initialization and training are as follows: randomly initialize the network weights and biases; use the backpropagation algorithm for network training to adjust the weights and biases to minimize the loss function; set the training parameters to ensure network convergence; The model verification and optimization are as follows: use the validation set to monitor the training process; adjust the network structure and training parameters to optimize the model performance; The test set verification is as follows: verify the prediction accuracy and generalization ability of the trained BP neural network model on the test set.
5. The motor parameter prediction and optimization method based on finite element simulation and neural network according to claim 1, wherein In the above S3, the objective function is defined as follows, Minimize the cogging torque: f(w) = CCM(w); Maximize the no-load back electromotive force: f(w) = EMF(w); Maximize the load output torque: f(w) = TORQUE(w); Multi-objective optimization: f(w) = \alphaEMF(w)+\betaCCM(w)+\gammaTORQUE(w).
6. The method for predicting and optimizing motor parameters based on finite element simulation and neural network according to claim 1, wherein In the above S3, the genetic algorithm is used as the optimization algorithm.
7. The motor parameter prediction and optimization method based on finite element simulation and neural network according to claim 1, characterized in that In the above S4, the simulation verification is as follows: substitute the optimal slot opening width into the finite element simulation model to verify the consistency between the predicted result and the simulation result; The experimental verification is as follows: fabricate a motor sample with the optimal slot opening width and conduct actual tests to verify the optimization effect; The analysis of the optimization results is as follows: compare the performance parameters before and after optimization, evaluate whether the optimization effect meets the design goal; analyze the influence of the optimized slot opening width on the motor performance and further optimize the design.
8. Application of the method for predicting and optimizing motor parameters based on finite element simulation and neural network as described in claim 1 in the field of motor design.