Motor electromagnetic-thermal coupling field inversion method based on missing field POD

Abstract

The invention provides a motor electromagnetic-thermal coupling field inversion method based on a missing field POD, and the method comprises the following steps: carrying out the electromagnetic-steady-state thermal field finite element simulation calculation of a plurality of operation conditions of a motor through employing a load current and a cogging angle as parameters, and respectively constructing a magnetic field POD reduced-order model and a temperature field POD reduced-order model; a temperature sensor is arranged on the outer surface of the stator, the temperature is measured, and motor temperature distribution is obtained through inversion with a missing field POD method; based on the temperature distribution, the load current of the motor is determined through SVR; and calling a magnetic field POD order reduction model based on the load current and the cogging angle parameter, calculating the magnetic field distribution of the motor, and realizing inversion of the electromagnetic-thermal coupling field of the motor. The method can consider the influence of the material parameter change caused by the temperature on the electromagnetic field distribution, and enables the motor electromagnetic characteristics obtained through inversion to be more accurate.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of online monitoring of electric machines, and particularly relates to a method for electromagnetic-thermal coupling field inversion of electric machines based on missing field POD. BACKGROUND

[0002] With the application of electric machines in new energy, aerospace and other fields, electric machine systems gradually develop towards high speed and high power density. During the operation of electric machines, electromagnetic fields and thermal fields interact with each other. Eddy current and magnetic hysteresis loss generated by electromagnetic fields act as heat sources to cause the temperature of electric machines to rise. The change of temperature affects the electrical conductivity or magnetic permeability of materials and further affects the distribution of electromagnetic fields. Therefore, in order to accurately characterize the operating characteristics of electric machines, the coupling effect of electromagnetic-thermal fields needs to be considered. Joint inversion of the electromagnetic and thermal field distribution of electric machines is of great significance for realizing online monitoring of the health status of electric machines. Through inversion, the coupling relationship between electromagnetic loss and temperature field can be determined, the hot spot area (such as winding or permanent magnet) can be accurately located, the risk of insulation aging or demagnetization can be warned in advance, and the operation safety can be ensured; inversion can quickly diagnose abnormalities (such as local short circuit, heat dissipation failure), support predictive maintenance, and reduce unplanned downtime; in addition, trend analysis based on historical inversion data can evaluate the health status of electric machines and prolong the service life. Electric machine physical field distribution inversion upgrades traditional offline detection to real-time regulation and control, which is the core support for intelligent and highly reliable operation of electric machines, and is particularly important for harsh working conditions such as new energy vehicles and wind power.

[0003] Currently, the inversion methods for physical fields mainly include the following methods: one is the numerical simulation method, which first establishes a multi-physical field coupling finite element model of electromagnetic field and temperature field according to the geometric structure and material properties of the electric machine; then the electromagnetic loss distribution (such as eddy current loss, iron loss) and the temperature rise field data caused by it are calculated through forward simulation; then the simulation results are compared with the actual measurement data (such as temperature sensor readings) to construct an objective function to quantify the difference between the two; next, an optimization algorithm (such as gradient descent method, genetic algorithm, etc.) is used to iteratively adjust the key parameters (such as material properties, heat dissipation coefficient, etc.) to gradually approach the simulation results to the measured data; finally, the optimal parameter solution is obtained through convergence verification, and the field distribution inversion result is obtained. Two is the analytical method, such as the equivalent thermal network method, etc., the method process of which is basically the same as that of the numerical simulation method, only the finite element model is replaced by a simplified model to improve the solving efficiency of the positive problem. Three is the data-driven method, such as constructing a machine learning model based on neural network, etc., and combining sensor measurement data to realize physical field inversion.

[0004] In addition to the above-mentioned methods, the missing field POD method proposed in recent years is a data reconstruction technology based on proper orthogonal decomposition (POD), which mainly restores missing data based on reduced order models. When data is missing (such as limited sensor measurement positions), the traditional POD model reduction method cannot be directly applied, while the missing field POD can mark the missing data position by defining a mask matrix, and reconstruct the complete field distribution using known data points. Missing field POD has been applied in the fields of flow field reconstruction, temperature field reconstruction, etc.

[0005] The main motor multi-physics inversion methods currently have the following shortcomings which need to be improved. For the inversion method based on numerical simulation, the forward calculation usually accompanies high calculation cost, especially for complex multi-physics or large-scale problems, the iterative inversion process needs to call the simulation model multiple times, and the calculation efficiency becomes a bottleneck, and the inversion accuracy depends heavily on the quality and coverage of the observation data, and sparse or incomplete observation data may lead to insufficient constraints on the solution space, further amplifying the error. These shortcomings limit the application of this method in scenarios with high real-time requirements or data scarcity. For the inversion method of the analytical model, the model often depends on highly simplified assumptions (such as uniform medium, regular boundary or linear relationship), which is difficult to accurately depict the actual complex scene, leading to deviation of the inversion result from the true situation. In addition, the analytical solution is often only applicable to specific equations or boundary conditions, with narrow application scope and difficulty in generalization to multi-physics coupling or nonlinear problems. For the data-driven model based on neural networks, it often needs to construct a data set containing a large number of samples, which leads to unacceptable long calculation time for pre-data preparation, and it is difficult to capture the characteristic rules of multi-physical coupling and highly nonlinear systems with small samples. SUMMARY

[0006] To solve the above technical problems, the present application proposes a motor electromagnetic-thermal coupling field inversion method based on missing field POD. According to the temperature data measured by a small number of sensors on the outer surface of the stator, the temperature distribution of the motor is reconstructed using the missing field POD method, and the working condition of the motor is determined based on the temperature distribution, and finally the electromagnetic distribution characteristics of the motor under the working condition are obtained. Compared with directly applying numerical simulation method, the present application avoids the problem of too long inversion time caused by repeatedly calling high calculation cost finite element simulation full model in iterative inversion process; compared with analytical model method, the present application can consider the influence of motor nonlinearity based on the construction of reduced order basis by finite element calculation results, and can be applied to complex multi-physical coupling problems and has higher inversion accuracy; compared with the data-driven method based on neural network, the present application can construct a high-precision reduced order model without a large number of training samples, and the reduced order model has high calculation efficiency and can ensure the real-time performance of inversion.

[0007] The specific technical scheme of the present application is:

[0008] A motor electromagnetic-thermal coupling field inversion method based on missing field POD, comprising the following steps:

[0009] Step 1, taking motor load current and rotating slot angle as parameters, carrying out finite element calculation on motor electromagnetic-steady thermal coupling problem, and constructing motor electromagnetic snapshot sample set and temperature snapshot sample set under different working conditions and , , , ,

[0010] Step 2, singular value decomposition is carried out on the electromagnetic snapshot sample set and the temperature snapshot sample set respectively, to obtain electromagnetic POD basis and temperature POD basis; the electromagnetic POD basis is , and the corresponding modal coefficient is , wherein k=1, 2, …, n, is the modal vector in the left singular vector obtained by singular value decomposition of the electromagnetic snapshot set , and the temperature POD basis is , and the corresponding modal coefficient is , wherein k=1, 2, …, n; is the modal vector in the left singular vector obtained by singular value decomposition of the temperature snapshot sample set .

[0011] Step 3, according to the given error , the electromagnetic mode and the temperature mode are truncated respectively, and the truncation criterion is determined by the following formula:

[0012] (1)

[0013] , wherein p is the number of truncated modes, respectively take the electromagnetic POD modal coefficient and the temperature POD modal coefficient;

[0014] Step 4, define a mask matrix for identifying the position of missing data, if a position i can obtain data, the corresponding element of the mask matrix is 1; otherwise, the corresponding element of the mask matrix is 0;

[0015] Step 5, according to the sensor measurement result, determine the incomplete temperature field distribution , the incomplete temperature field distribution can be represented by the complete temperature field distribution and the mask matrix :

[0016] (2)

[0017] Step 6, according to the truncated temperature POD basis and the undetermined modal coefficient to be determined Determine the guessed temperature field distribution , which is in the form of:

[0018] (3)

[0019] Step 7, solve the minimum value problem of the error between the residual temperature field and the guessed temperature field by using the least square method Determine the undetermined modal coefficient in the guessed temperature field ;

[0020] Step 8, restore the complete temperature field distribution according to the undetermined modal coefficient And the temperature POD basis :

[0021] (4)

[0022] Step 9, identify the temperature field image based on the support vector regression SVR algorithm, and determine the motor load current size corresponding to the temperature field;

[0023] Step 10, substitute the motor load current size corresponding to the temperature field into the motor load current parameter size to the motor electromagnetic POD reduced order model, and calculate the motor electromagnetic field distribution to realize the inversion of the motor electromagnetic-thermal coupling field.

[0024] The present application has the following beneficial effects:

[0025] The present application provides a new method for motor electromagnetic-thermal coupling field inversion. The temperature distribution of the motor is reconstructed based on the temperature results measured by the sensors on the outer surface of the stator by the missing field POD method, and the operating condition of the motor is determined based on the temperature distribution to obtain the motor electromagnetic field distribution. The method uses finite element results to construct the POD reduced order basis and carries out electromagnetic-thermal coupling reconstruction, which has higher calculation efficiency compared to directly using finite element model calculation, can consider the influence of nonlinearity compared to the simplified model of the motor, and has lower requirement on the number of samples compared to the data-driven model based on neural network. The method establishes the connection between the thermal field and the electromagnetic field through the motor load current parameter, thereby realizing the joint inversion of the electromagnetic-thermal field, breaking the technical bottleneck of the difficulty of multi-physical coupling field inversion, and promoting the further development of the motor health state online monitoring technology. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 Flowchart of motor electromagnetic-thermal coupling field inversion based on missing field POD; DETAILED DESCRIPTION

[0027] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application. In addition, the technical features involved in the various embodiments of the present application described below can be combined with each other as long as they do not conflict with each other. In order to achieve the above-mentioned purpose, the technical solutions adopted by the present application are as follows.

[0028] The present application provides a motor electromagnetic-thermal coupling field inversion method based on missing field POD, and a flow chart thereof is shown as Figure 1 The method comprises the following steps:

[0029] Step 1, taking the motor load current and the rotating slot angle as parameters, performing finite element calculation on the motor electromagnetic-steady thermal coupling problem, and constructing the motor electromagnetic snapshot sample set and the temperature snapshot sample set under different working conditions and , wherein , , is a real number space, m is the snapshot sample dimension, and n is the snapshot sample number.

[0030] Step 2, performing singular value decomposition on the electromagnetic snapshot sample set and the temperature snapshot sample set . The singular value decomposition algorithm is realized by solving the formula , wherein is the matrix that needs to be singularly decomposed, is the left singular vector obtained by decomposition, is the right singular vector obtained by decomposition, is the eigenvalue vector. After singular value decomposition, the electromagnetic POD basis and the temperature POD basis are obtained. The electromagnetic POD basis is , and the corresponding modal coefficient is , wherein k=1, 2, …, n, is the electromagnetic snapshot set , the modal vector in the left singular vector obtained by singular value decomposition, is the eigenvalue obtained by singular value decomposition; the temperature POD basis is , and the corresponding modal coefficient is , wherein k=1, 2, …, n, is the temperature snapshot sample set , the modal vector in the left singular vector obtained by singular value decomposition, is the eigenvalue obtained by singular value decomposition.

[0031] Step 3, according to the given error The truncation criterion is determined by the following formula:

[0032] (1)

[0033] where p is the number of truncated modes, respectively, are the electromagnetic POD basis mode coefficients and the temperature POD mode coefficients.

[0034] Step 4, define the mask matrix is used to identify the location of missing data. If a certain position i can obtain data (such as the position of the sensor), the corresponding element of the mask matrix is 1; otherwise the corresponding element of the mask matrix is 0.

[0035] Step 5, determine the incomplete temperature field distribution according to the sensor measurement results , the incomplete temperature field distribution can be represented by the complete temperature field distribution and the mask matrix :

[0036] (2)

[0037] Step 6, determine the guessed temperature field distribution according to the truncated temperature POD basis and the undetermined mode coefficient , which is in the form of:

[0038] (3)

[0039] Step 7, use the least squares method to solve the minimum value problem of the error between the incomplete temperature field and the guessed temperature field, to determine the in formula (3).

[0040] Step 8, restore the complete temperature field distribution according to the undetermined mode coefficient and the temperature POD basis:

[0041] (4)

[0042] Step 9, identify the temperature field image based on the support vector regression (SVR) algorithm to determine the motor load current size corresponding to the temperature field. The SVR algorithm realizes regression fitting by constructing a hyperplane function with an interval band , and the function corresponding to the hyperplane is obtained by solving the following optimization problem:

[0043] (5)

[0044] Constraints:

[0045] ;

[0046] ;

[0047] ;

[0048] wherein, is a regularization parameter for controlling the model error and training complexity; and are undetermined coefficients for determining the hyperplane expression; is a margin bandwidth for describing the allowable error range; the slack variable and are respectively used to measure the deviation of the i-th data point above or below the margin band.

[0049] Step 10, substitute the motor load current size corresponding to the temperature field into the motor load current parameter size to the motor electromagnetic POD reduced order model, and the reduced order model is calculated by formula to obtain the motor electromagnetic field distribution, wherein is the i-th row vector in the electromagnetic POD base, is the load current parameter size, is the magnetic vector potential corresponding to the load current p, and the inversion of the motor electromagnetic-thermal coupled field is realized.

[0050] The application has the following alternatives:

[0051] In step 7, other methods can be used to solve the minimum value problem, such as gradient descent algorithm, evolution algorithm.

[0052] In step 9, other image recognition methods can be used for motor load current size prediction, such as K-nearest neighbor algorithm, deep learning method, convolutional neural network method.

[0053] The application object of the application is not only limited to motors, but also can be popularized to other power equipment involving electromagnetic-thermal field coupled inversion calculation.

Claims

1. A method for inverting the electromagnetic-thermal coupled field of a motor based on a missing field POD, characterized in that, Includes the following steps: Step 1: Using the motor load current and the cogging angle of the rotating teeth as parameters, perform finite element analysis on the electromagnetic-steady-state thermal coupling problem of the motor, and construct an electromagnetic snapshot sample set for different operating conditions of the motor. and temperature snapshot sample set ,in , , Let m be the dimension of the snapshot sample and n be the number of snapshot samples; Step 2: Perform electromagnetic snapshot sample set analysis respectively. and temperature snapshot sample set Singular value decomposition was performed to obtain electromagnetic POD substrates and temperature POD substrates; Electromagnetic POD substrate is The corresponding modal coefficients are , where k=1, 2, …, n, Electromagnetic snapshot set The mode vectors in the left singular vectors obtained from singular value decomposition, with the temperature POD basis as follows: The corresponding modal coefficients are , where k = 1, 2, …, n; Temperature snapshot sample set The modal vectors in the left singular vectors obtained from singular value decomposition; Step 3: Based on the given error The electromagnetic mode and the temperature mode are truncated separately, and the truncation criterion is determined by the following formula: （1） Where p is the number of truncated modes. The values ​​are taken as the modal coefficients of the electromagnetic POD substrate and the modal coefficients of the temperature POD, respectively; Step 4: Define the mask matrix Used to identify the location of missing data. If data can be obtained at a certain location i, then the corresponding element of the mask matrix... =1; otherwise, the corresponding element of the mask matrix is ​​1. =0; Step 5: Determine the incomplete temperature field distribution based on the sensor measurement results. An incomplete temperature field distribution can be derived from a complete temperature field distribution. and mask matrix Represented as: （2） Step 6: Based on the truncated temperature POD substrate and the undetermined modal coefficients Determine the predicted temperature field distribution Its form is: （3） Step 7: Use the least squares method to solve for the error between the incomplete temperature field and the guessed temperature field. The minimum value problem is to determine the undetermined modal coefficients in the guessed temperature field. ; Step 8: Based on the undetermined modal coefficients Restoring the complete temperature field distribution on the temperature-controlled POD substrate : （4） Step 9: Identify the temperature field image based on the Support Vector Regression (SVR) algorithm to determine the magnitude of the motor load current corresponding to the temperature field; Step 10: Substitute the magnitude of the motor load current corresponding to the temperature field into the magnitude of the motor load current parameter in the reduced-order model of the motor electromagnetic POD, calculate the distribution of the motor electromagnetic field, and realize the inversion of the electromagnetic-thermal coupling field of the motor.

2. The method according to claim 1, characterized in that, The SVR algorithm constructs a spaced-band structure. hyperplane function To achieve regression fitting, the function corresponding to the hyperplane is obtained by solving the following optimization problem: （5） Constraints: ； ； ； in, For regularization parameters; and These are coefficients to be determined; For interval band width, slack variable and These are used to measure the deviation of the i-th data point from being above or below the interval band.

3. The method according to claim 1, characterized in that, The reduced-order model of the electromagnetic POD of the motor is obtained through the formula. The electromagnetic field distribution of the motor was calculated, where Let i be the i-th row vector in the electromagnetic POD basis. For the load current parameter, This is the magnetic vector potential corresponding to the load current p.

4. The method according to claim 1, characterized in that, Singular value decomposition (SVD) algorithms solve the formula Implementation, in which For the matrix that needs to be singular value decomposition, The left singular vector obtained from the decomposition, The right singular vector obtained from the decomposition. This is the eigenvalue vector.

5. The method according to claim 1, characterized in that, The location where data can be obtained is the location of the sensor.

6. The method according to claim 1, characterized in that, Step 7 uses other methods to solve the minimum problem, including gradient descent algorithm and evolutionary algorithm.

7. The method according to claim 1, characterized in that, In step 9, other image recognition methods are used to predict the magnitude of the motor load current, including the K-nearest neighbor algorithm, deep learning methods, and convolutional neural network methods.

8. The method according to claim 1, characterized in that, The application of the method can be extended to other power equipment involving electromagnetic-thermal field coupling inversion calculations.

9. An electronic device, characterized in that, include: One or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, cause the processor to perform the method described in any one of claims 1 to 7.

Images