A method and device for optimizing motor topology based on prediction coefficient analysis
Through the prediction coefficient analysis method of Latin hypercube sampling and polynomial regression, the size design of the V-type internal permanent magnet synchronous motor is optimized, the problem of balancing electromagnetic performance and structural strength is solved, and efficient and accurate motor optimization is achieved.
Patent Information
- Application Number
- CN202410476955.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-19
- Publication Date
- 2025-09-19
- Estimated Expiration
- 2044-04-19
AI Technical Summary
The existing technology has difficulty in effectively balancing electromagnetic performance and structural strength when optimizing the size design of a V-shaped internal permanent magnet synchronous motor. In particular, under high power density, the rotor magnetic isolation bridge is subjected to large centrifugal stress and has insufficient magnetic isolation performance.
The prediction coefficient analysis method of Latin hypercube sampling and polynomial regression is adopted. The variable response is evaluated and the prediction coefficient is calculated through the finite element solver to determine the key dimensional parameters affecting the motor performance. Combined with optimized sampling, the optimal parameter combination is obtained to optimize the motor size design.
The sampling efficiency and accuracy of motor design are improved, the correlation of sample points is reduced, the electromagnetic performance and structural strength of the motor are optimized, and the design requirements are met.
Smart Images

Figure CN118296765B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of motor structure and topology optimization, and in particular to a prediction coefficient analysis method based on Latin hypercube sampling and polynomial regression. The present invention relates to a method and apparatus for parameter sensitivity analysis of motor topology optimization. Background Art
[0002] With the global popularity of new energy vehicles, market demands for higher performance in drive motors are increasing. V-type interior permanent magnet synchronous motors, as one such type, offer advantages such as improved starting performance, mechanical properties, a wider speed range, and higher power density. However, as the power density of drive motors continues to increase, so too does the maximum speed requirement. This means that the rotor magnetic isolation bridge must withstand greater centrifugal stress while also ensuring sufficient magnetic isolation. To balance electromagnetic performance and structural strength, a multi-physical variable sensitivity analysis was conducted on the motor's dimensional design. To this end, a motor structure and topology optimization method and device based on Coefficient of Prognosis (CoP) analysis was designed. Traditional sampling methods require a large number of sampling points and extensive computation to obtain a relatively accurate distribution space of input variables and their correlation. Using Latin Hypercube Sampling (LHS) can ensure more reasonable, accurate, and representative sampling points, minimize sample correlation, reduce the number of sampling points, and achieve a more accurate distribution space of input variables. Based on the distribution space of these variables, a polynomial regression model is constructed. These variables are then evaluated using the Moving Least Squares Approximation (MLS) algorithm of the finite element solver to determine the model response for each variable and calculate the coefficient of prediction (CoP). The metamodel with the largest CoP is then selected as the optimal metamodel, and a corresponding variable space is established for each approximate response model. Finally, the approximate model estimates the variance-based first-order and total effect sensitivity indices to quantify the importance of individual input variables, i.e., the prediction coefficients of individual variables. The prediction coefficients of individual variables can be used to determine which of the various motor dimensional parameters have the greatest impact on the electromagnetic performance and structural strength. Combined with the sample space obtained through optimized sampling after prediction coefficient analysis, optimal variable combinations can be obtained, thereby improving the motor's dimensional design to balance electromagnetic performance and structural strength. Optimized sampling after prediction coefficient analysis improves sampling efficiency and accuracy, resulting in more reasonable samples and minimized correlation between sample points. Finally, these optimal parameter combinations are returned to the finite element analysis software for torque and stress calculations to verify improvement. Summary of the Invention
[0003] The present invention aims to overcome the above-mentioned shortcomings of the prior art and provides a method and device for optimizing the motor topology structure based on prediction coefficient analysis.
[0004] The present invention aims to optimize the size of a motor and improve sampling efficiency and the accuracy of a computational metamodel by using prediction coefficient analysis. The ultimate goal is to optimize the electromagnetic performance and structural strength of the motor to meet design requirements. Taking a V-type interior permanent magnet synchronous motor as an example, the present invention solves the above-mentioned problem by adopting a technical solution: a sensitivity analysis method and device for motor structure and topology optimization based on prediction coefficient analysis, characterized in that Latin hypercube sampling is used to obtain a sample set with a smaller number of samples, more reasonable accuracy, and representativeness, and less correlation between variables, to form a sample distribution space of variables. Polynomial regression is established based on the distribution space of these variables, and these variables are evaluated using the moving least squares (MLS) algorithm of a finite element solver to determine the model response of each variable and calculate the prediction coefficient (CoP). Next, the metamodel with the largest prediction coefficient is selected as the optimal metamodel, and a corresponding variable space is established for each approximate response model. Finally, the approximate model is used to estimate the first-order and total effect sensitivity index based on the variance to quantify the importance of a single input variable, that is, the prediction coefficient of the single variable. The prediction coefficients of individual variables can be used to determine which of the various motor dimensional parameters have the greatest impact on the motor's electromagnetic performance and structural strength. Combined with the sample space obtained through optimized sampling after prediction coefficient analysis, optimal variable combinations can be obtained, allowing for improved motor dimensional design that balances electromagnetic performance and structural strength. Optimized sampling after prediction coefficient analysis improves sampling efficiency and accuracy, resulting in more reasonable samples and minimizing correlation between sample points. Finally, these optimal parameter combinations are returned to the finite element analysis software for torque and stress calculations to verify improvement. Latin Hypercube Sampling (LHS) is a stratified sampling technique that approximates random sampling from a multivariate parameter distribution. Each hyperplane perpendicular to the axis contains at most one sample, and it also forces the simulation to sample "outlying events" to ensure that these "outlying events" are accurately represented in the simulation output. Polynomial regression is a regression analysis method that can be used to fit data exhibiting nonlinear relationships between independent and dependent variables. Unlike simple linear regression, which can only fit linear relationships, polynomial regression incorporates polynomial functions to fit data, allowing it to accommodate more complex data patterns. The advantage of the moving least squares (MLS) method is that no training is required before computing the approximation points; only weighting factors and local polynomial coefficients need to be computed at each point, which makes the method very fast compared to other approximation techniques.
[0005] Latin hypercube sampling (LHS) was first performed within a preset range on the various uncertain parameters (dimensional parameters of the V-type permanent magnet synchronous motor) that affect the minimum torque ripple, maximum average torque, and maximum stress of a V-type interior permanent magnet synchronous motor. This yielded a variable sample and sample space. Polynomial regression was then established and these variables were evaluated using algorithms such as moving least squares (MLS). The model response of each variable was determined and the coefficient of prediction (CoP) was calculated. The meta-model with the largest prediction coefficient was then selected as the optimal meta-model, and a corresponding variable space was established for each approximate response model. Finally, the approximate model was used to estimate the first-order and total effect sensitivity indices based on variance to quantify the importance of individual input variables, i.e., the prediction coefficients of individual variables. Based on the CoP values of the individual variables, the influence weights of the various uncertain parameters (dimensional parameters of the V-type permanent magnet synchronous motor) that affect the minimum torque ripple, maximum average torque, and maximum stress of the V-type interior permanent magnet synchronous motor were analyzed. Finally, the optimal parameter combination based on the influence weights, obtained through finite element analysis, was used to simulate and calculate the minimum torque ripple, maximum average torque, and maximum stress of the V-type interior permanent magnet synchronous motor.
[0006] A first aspect of the present invention provides a motor topology optimization method based on prediction coefficient analysis, which specifically includes the following steps:
[0007] S1. Determine the optimization target based on the design requirements, then scan the space of the motor's topology-related dimensions (i.e., input variables) using the Latin hypercube sampling method and establish a polynomial regression.
[0008] S2, evaluating these variables using a finite element solver, determining the model response for each variable and calculating the prediction coefficients;
[0009] S3. The meta-model with the largest prediction coefficient is selected as the optimal meta-model, and the corresponding variable space is established for each approximate response model. The approximate model is then used to estimate the first-order and total effect sensitivity indices based on variance to quantify the importance of individual input variables, i.e., the prediction coefficients of individual variables.
[0010] S4. After analyzing the prediction coefficients of individual variables, the sampling is optimized again to obtain some optimal variable combination parameters. The finite element method is used to calculate the motor under the parameter combination structure to determine whether the performance meets the optimization requirements. If so, the optimization is completed; otherwise, return to step S1.
[0011] The Latin hypercube sampling method described in step S1 includes: firstly determining the sampling dimension n and the sampling size m according to the variable dimension, where m=(n+1) 2; Secondly, for the ath dimension (a = 1, 2, 3...n), divide it into m randomly distributed sampling point intervals with equal spacing, randomly select a number from an interval as a sample value, and then delete the interval from the ath dimension list. Repeat this operation in other dimensions to obtain a sample set X with n sample values. i (i=1,2,3...m), each sample set X i That is a parameter combination, and finally m parameter combinations can be obtained. This makes each sample value of each dimension appear only once, to ensure the uniformity and efficiency of sampling; finally, the obtained sample set X i Arrange them by columns to obtain a Latin hypercube sample matrix with n rows and m columns.
[0012] The polynomial regression in step S1 includes: generating a variable set x i That is, the sample set X i The model output set y i , which is an approximate value and the error term ε i The sum can be expressed by the following formula:
[0013]
[0014] where p(x) is a second-order coupled polynomial basis:
[0015]
[0016] is a vector containing the unknown regression coefficients. These coefficients are usually estimated from a given set of sampled support points by assuming independent errors with equal variance at each point. Using matrix notation, the resulting least squares solution is shown below:
[0017]
[0018] Where P is the matrix of the basis polynomials containing the support point samples, and y is the vector of support point values.
[0019] The model response of the variable in step S2 includes: According to formula 1 and formula 3, the final approximate function can be obtained:
[0020]
[0021] The diagonal matrix W(x) contains the weighting function value corresponding to each support point. The distance-dependent weighting function w=w(||xx) has been introduced. i ||). Use a Gaussian weighting function:
[0022]
[0023] The influence radius D directly affects the approximation error, and α is a numerical constant. Samples closer to the center point, i.e., samples with smaller influence radius D, have greater weight and a greater impact on the results.
[0024] The prediction coefficient in step S2 is defined by the following formula:
[0025]
[0026] Among them SS T is the total variation of the output:
[0027] is the sum of squared prediction errors:
[0028] The prediction coefficient of the single variable in step S3 can be defined by the following formula:
[0029]
[0030] in are the first-order and total effect sensitivity indices:
[0031]
[0032] where V(Y|X ~i ) are all model inputs (excluding X i ) causes the variance of Y.
[0033] The optimized sampling after the prediction coefficient analysis of a single variable in step S4 can improve the efficiency and accuracy of sampling, making the samples more reasonable and minimizing the correlation between sample points.
[0034] A second aspect of the present invention relates to a motor topology optimization device based on prediction coefficient analysis, comprising a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the motor topology optimization method based on prediction coefficient analysis of the present invention.
[0035] A third aspect of the present invention relates to a computer-readable storage medium, characterized in that a program is stored thereon, and when the program is executed by a processor, the motor topology structure optimization method based on prediction coefficient analysis of the present invention is implemented.
[0036] The present invention quantifies the uncertainty of model outputs to determine the degree to which changes in dimensional parameters affect the electromagnetic performance and structural strength of a motor. The method includes: first, scanning the space of input variables (i.e., the dimensions related to the motor's topology) based on a Latin hypercube sampling method and establishing a polynomial regression; second, evaluating these variables using a finite element solver to determine the model response of each variable and calculate the prediction coefficient; then, selecting the metamodel with the largest prediction coefficient as the optimal metamodel and establishing a corresponding variable space for each approximate response model; and finally, using the approximate model to estimate the first-order and total effect sensitivity indices based on variance to quantify the importance of individual input variables (i.e., the prediction coefficients of the individual variables). Using the prediction coefficients of the individual variables, it is possible to determine which of the various dimensional parameters have the greatest impact on the electromagnetic performance and structural strength of the motor. Combined with the sample space obtained by Latin hypercube sampling, optimal variable parameter combinations can be obtained, thereby improving and optimizing the motor's dimensional design to balance electromagnetic performance and structural strength. Optimized sampling after prediction coefficient analysis can improve sampling efficiency and accuracy, making the samples more reasonable and minimizing the correlation between sample points.
[0037] The advantages of the present invention are that it can effectively calculate and predict the minimum torque fluctuation, maximum average torque and maximum stress of the motor, and provide a reference basis for motor design and motor operation optimization. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a schematic diagram of the original model of the V-type interior permanent magnet synchronous motor and its optimized parameters applied to the present invention.
[0039] Figure 2 It is a schematic diagram of the parameter combination 1 model of the present invention.
[0040] Figure 3 It is a schematic diagram of the parameter combination 2 model of the present invention.
[0041] Figure 4 It is a flow chart of the method of the present invention. Specific implementation methods
[0042] The present invention is further described in detail below with reference to the accompanying drawings and through implementation cases. The implementation cases provided below for a V-type internal permanent magnet synchronous motor are explanations of the application of the present invention, and the present invention is not limited to the following cases.
[0043] Example 1
[0044] Reference Figure 4 This embodiment relates to a motor topology optimization method based on prediction coefficient analysis, which specifically includes the following steps:
[0045] S1. Determine the optimization target based on the design requirements, then scan the space of the motor's topology-related dimensions (i.e., input variables) using the Latin hypercube sampling method and establish a polynomial regression.
[0046] S2, evaluating these variables using a finite element solver, determining the model response for each variable and calculating the prediction coefficients;
[0047] S3. The meta-model with the largest prediction coefficient is selected as the optimal meta-model, and the corresponding variable space is established for each approximate response model. The approximate model is then used to estimate the first-order and total effect sensitivity indices based on variance to quantify the importance of individual input variables, i.e., the prediction coefficients of individual variables.
[0048] S4. After analyzing the prediction coefficients of individual variables, the sampling is optimized again to obtain some optimal variable combination parameters. The finite element method is used to calculate the motor under the parameter combination structure to determine whether the performance meets the optimization requirements. If so, the optimization is completed; otherwise, return to step S1.
[0049] The following example optimizes a V-type interior permanent magnet synchronous motor:
[0050] (1) See Figure 1 , a V-type interior permanent magnet synchronous motor model was established in the finite element analysis software and stress and electromagnetic force simulation settings were performed to obtain the parameters of the maximum equivalent stress and magnetic isolation performance, see Table 1;
[0051] Magnet_Thickness [mm] 6 Magnet_Step [mm] 1 Magnet_Lip_in [mm] 0.25 CenterBrige_Width [mm] 1.5 CenterBrige_Length [mm] 5 w1 [mm] 1 OuterBrige_Length [mm] 3 OuterBrige_Width [mm] 0.82 Yoke [mm] 11.7 Bs1 [mm] 5.1 Hs0 [mm] 1.3 Equivalent stress Max [MPa] 3.68426
[0052] Table 1 Motor size parameters and stress simulation results
[0053] (2) Set the variables to be optimized (motor size), optimization objectives (torque ripple, average torque) and constraints (maximum equivalent stress);
[0054] (3) Set the optimal value range of each variable, see Table 2;
[0055]
[0056]
[0057] Table 2 Optimized value range of variables
[0058] (4) Setting the optimization objectives, namely, minimum torque fluctuation, maximum average torque, and maximum equivalent stress no greater than 300 MPa;
[0059] min pk2pk(Torque)
[0060] max avg(Torque)
[0061] stmax(Equivalent stress)≤300
[0062] (5) Using the Latin hypercube sampling method and setting the number of sampling times to obtain a reasonably accurate sample space, and then establishing polynomial regression;
[0063] (6) Start the calculation and calculate the COP value of each variable according to the above-mentioned polynomial regression and moving least squares algorithm, see Table 3;
[0064]
[0065]
[0066] Table 3 COP values of various variables
[0067] (7) From the data in Table 3, it can be observed that among the variables affecting the average torque, Bs1 and Yoke have a larger influence weight; among the variables affecting the torque fluctuation, Bs1 and OuterBrige_Length have a larger influence weight; among the variables affecting the maximum stress, CenterBrige_Width and Magnet_Thickness have a larger influence weight. In the subsequent optimization settings, the finite element analysis software uses the Evolutionary Algorithm (EA) algorithm to focus on sampling and analyzing these variable parameters with high CoP values.
[0068] (8) In the finite element analysis software, two sets of optimal variable parameter combinations that meet the optimization objectives can be obtained. Parameter combination 1 is the maximum optimization result for torque fluctuation while ensuring that the maximum equivalent stress is no more than 300 MPa and the average torque remains unchanged. Parameter combination 2 is the maximum optimization result for average torque while ensuring that the maximum equivalent stress is no more than 300 MPa and the torque fluctuation remains unchanged. See Table 4 (parameter combination 1), Figure 2 (Parameter combination 1 model), Table 5 (Parameter combination 2), Figure 3 (parameter combination 2 model);
[0069]
[0070]
[0071] Table 4 Parameter combination 1
[0072] Magnet_Thickness [mm] 6.6 CenterBrige_Width [mm] 1.35 CenterBrige_Length [mm] 5.41 OuterBrige_Length [mm] 3.09 OuterBrige_Width [mm] 0.9 Yoke [mm] 10.54 Bs1 [mm] 5.16 Hs0 [mm] 1.41
[0073] Table 5 Parameter combination 2
[0074] (9) Using these two sets of parameter combinations, the finite element analysis software was returned for verification. Under the condition that the maximum equivalent stress of the V-type built-in permanent magnet synchronous motor is not greater than 300MPa and the average torque remains unchanged, the torque fluctuation obtained by parameter combination 1 is 39.8288NM, which is about 13.3% lower than the original 45.9442NM. Under the condition that the maximum equivalent stress of the V-type built-in permanent magnet synchronous motor is not greater than 300MPa and the torque fluctuation remains unchanged, the average torque obtained by parameter combination 2 is 258.0065NM, which is about 5.4% higher than the original 244.8074NM. The expected structural and topological optimization of the V-type built-in permanent magnet synchronous motor was achieved for different needs.
[0075] Example 2
[0076] This embodiment relates to a motor topology optimization device based on prediction coefficient analysis, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the motor topology optimization method based on prediction coefficient analysis of Example 1.
[0077] Example 3
[0078] This embodiment relates to a computer-readable storage medium, characterized in that a program is stored thereon, and when the program is executed by a processor, the motor topology structure optimization method based on prediction coefficient analysis of Example 1 is implemented.
[0079] The above implementation methods are only exemplary descriptions of this patent and do not limit its scope of protection. Those skilled in the art may also make partial changes thereto. As long as they do not exceed the spirit of this patent, they are within the scope of protection of this patent.
Claims
1. A motor topology optimization method based on prediction coefficient analysis, characterized in that: The optimal meta-model established based on the prediction coefficients of the motor is used to simulate the optimization results of the motor under different structural parameters and provide some optimal variable combination parameters, thereby improving the size design of the motor to take into account both electromagnetic performance and structural strength; The specific steps include: S1. Determine the optimization target according to the design requirements, and then scan the space of the motor's topology-related dimensions, i.e., input variables, by the Latin hypercube sampling method, and establish a polynomial regression. The polynomial regression includes: generating a variable set x i That is, the sample set X i The model output set y i , which is an approximate value and the error term ε i The sum is expressed by the following formula: where p(x) is a second-order coupled polynomial basis: is a vector containing the unknown regression coefficients; these coefficients are usually estimated from a given set of sampled support points by assuming independent errors with equal variance at each point; using matrix notation, the resulting least squares solution is given by: Where P is the matrix of the basis polynomials containing the support point samples, and y is the vector of support point values; S2. Evaluate these variables using a finite element solver to determine the model response of each variable and calculate the prediction coefficient; the model response of the variable includes: deriving the final approximate function according to Formula 1 and Formula 3: The diagonal matrix W(x) contains the weighting function value corresponding to each support point; the distance-dependent weighting function w=w(||xx) has been introduced. i ||); using a Gaussian weighting function: Among them, the influence radius D directly affects the approximation error, and α is a numerical constant; the samples closer to the center point, that is, the samples with smaller influence radius D, have greater weight and greater influence on the results; The prediction coefficient is defined by the following formula: Among them SS T is the total variation of the output: is the sum of squared prediction errors: S3. The meta-model with the largest prediction coefficient is selected as the optimal meta-model, and the corresponding variable space is established for each approximate response model. The approximate model is then used to estimate the first-order and total effect sensitivity indices based on variance to quantify the importance of individual input variables, i.e., the prediction coefficients of individual variables. S4. After analyzing the prediction coefficients of individual variables, the sampling is optimized again to obtain some optimal variable combination parameters. The finite element method is used to calculate the motor under the parameter combination structure to determine whether the performance meets the optimization requirements. If so, the optimization is completed; otherwise, return to step S1.
2. The motor topology optimization method based on prediction coefficient analysis according to claim 1, characterized in that: The Latin hypercube sampling method described in step S1 includes: firstly determining the sampling dimension n and the sampling size m according to the variable dimension, where m=(n+1) 2 ; Secondly, for the ath dimension, divide it into m randomly distributed sampling point intervals with equal spacing, where a = 1, 2, 3...n. Randomly select a number from an interval as a sample value, and then delete the interval from the ath dimension list. Repeat this operation in other dimensions to obtain a sample set X with n sample values. i , i=1,2,3...m, each sample set X i That is, a parameter combination, and finally m parameter combinations are obtained; this makes each sample value of each dimension appear only once, to ensure the uniformity and efficiency of sampling; finally, the obtained sample set X i Arrange them by columns to obtain a Latin hypercube sample matrix with n rows and m columns.
3. The motor topology optimization method based on prediction coefficient analysis according to claim 1, characterized in that: The prediction coefficient of a single variable in step S3 is defined by the following formula: in are the first-order and total effect sensitivity indices: where V(Y|X ~i ) are all model inputs without X i The variance of Y caused by .
4. The motor topology optimization method based on prediction coefficient analysis according to claim 1, characterized in that: The optimized sampling after the prediction coefficient analysis of a single variable described in step S4 can improve the efficiency and accuracy of sampling, making the samples more reasonable and minimizing the correlation between sample points.
5. A motor topology optimization device based on prediction coefficient analysis, characterized in that: The method comprises a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, the method is used to implement the motor topology optimization method based on prediction coefficient analysis according to any one of claims 1 to 4.
6. A computer-readable storage medium, characterized in that A program is stored thereon, and when the program is executed by a processor, the motor topology optimization method based on prediction coefficient analysis according to any one of claims 1 to 4 is implemented.
Citation Information
Patent Citations
A method for predicting concrete durability based on data mining and artificial intelligence algorithm
AU2020101854A4
Engine combustion noise optimization prediction method and device and storage medium
CN113806991A
