Multi-physics field modeling method for curvature-driven dynamic compensation motor

Abstract

The invention discloses a curvature-driven dynamic compensation motor multi-physical field modeling method, and relates to the field of motor simulation modeling, and the method comprises the steps: constructing a bidirectional coupling basic model of an electromagnetic field, a temperature field and a structural mechanical field; thermally induced parameter drift and mechanical deformation errors in the electromagnetic field are corrected online through a dynamic parameter compensator; on the basis of local curvature characteristics of the parameter curve, smoothing the electromagnetic parameter curve by adopting a curvature-driven dual-mode adaptive switching mechanism, and generating a dynamic weight factor according to curvature distribution characteristics to fuse output results of the two modes; generating a parameter response surface based on a Kriging agent model, and replacing a finite element to calculate a nonlinear equation of a key physical field; and outputting a joint simulation result of the electromagnetic torque ripple frequency spectrum, the winding temperature distribution and the rotor vibration mode. According to the scheme, the ternary contradiction of precision, efficiency and stability in traditional motor modeling can be solved, and high application value is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of motor simulation modeling, in particular to a curvature-driven dynamic compensation motor multi-physical field modeling method. BACKGROUND

[0002] With the wide application of high-precision mechatronic systems in aerospace, high-end equipment, new energy and other fields, motor multi-physical field coupling modeling has become a core means to optimize mechatronic performance and predict thermal mechanical reliability. Traditional modeling methods mainly rely on single physical field finite element analysis, and realize parameter transmission through offline data table, but under high speed, variable working conditions and long-term operation conditions, there are significant electromagnetic, thermal and force multi-field dynamic coupling effects in the motor. For example, the centrifugal deformation of the high-speed motor rotor will cause the air gap magnetic field distortion, and the winding temperature rise will cause the material property drift, and the existing method cannot accurately capture the nonlinear behavior under transient working conditions due to the neglect of the real-time interaction mechanism of multi-fields. In addition, the development needs of industrial digital twin and predictive maintenance further require the model to have high fidelity and real-time performance, and it is urgent to break through the algorithm and precision bottleneck of traditional offline modeling.

[0003] The existing technology mainly has three systematic defects: 1. The mainstream method generates inductance and flux linkage parameter curves through finite element scanning, and makes offline data table after smoothing treatment by fast Fourier transform or least square method, which cannot correct the parameter time-varying error caused by thermal drift and mechanical deformation when calling the table, resulting in significant degradation of model accuracy under high-speed transient working conditions. 2. The existing modeling mainly adopts one-way parameter transmission, and lacks the real-time feedback loop of electromagnetic, thermal and force fields, resulting in the accumulation of errors in multi-physical domains. Especially at high speed, the positive feedback of rotor dynamics instability and local overheating aggravates the model divergence risk. 3. In order to improve the real-time performance, some studies introduce surrogate models to replace finite element calculation, but the sample sampling strategy relies on uniform distribution or single sampling, which does not focus on the actual working condition intensive area, resulting in sharp increase of prediction error of surrogate model in key working condition interval. At the same time, the traditional reduction method is difficult to meet the high-precision control demand due to the neglect of spatial harmonic saturation effect. Therefore, based on the above problems, the present application proposes a curvature-driven dynamic compensation motor multi-physical field modeling method. SUMMARY

[0004] PURPOSE OF THE INVENTION In order to solve the above problems, the purpose of the present application is to provide a curvature-driven dynamic compensation motor multi-physical field modeling method, which eliminates the numerical noise of inductance and flux linkage curves through an innovative data post-processing mechanism, constructs a high-fidelity lumped parameter model, thereby improving the simulation accuracy and calculation stability of electrical systems, and meeting the high reliability requirements of motor models in the fields of aviation and rail transportation.

[0005] TECHNICAL SCHEME In order to achieve the above object, the application provides a curvature driving dynamic compensation motor multi-physical field modeling method, which constructs a finite element model based on motor structure parameters, extracts original inductance and flux linkage parameter curves by scanning current combinations in different working conditions through equivalent current vectors, performs fast Fourier transform analysis on the parameter curves by using a double-path error filtering mechanism, extracts direct current components and main harmonic components, reconstructs the permanent magnet flux linkage according to the harmonic number weighting and sorting, ensures the continuity of the first derivative of the curve by using the weighted least squares method, solves the phase distortion problem, stores the smoothed parameters into an offline data table, constructs circuit equations taking windings and bars as units in a circuit simulator, synchronously solves nonlinear equations of the electrical system based on the node voltage method, and realizes weak coupling iteration of mechanical and circuit models through electromagnetic torque and position parameters.

[0006] In the first aspect, the application provides a curvature driving dynamic compensation motor multi-physical field modeling method, comprising: Constructing a bidirectional coupling basic model of electromagnetic fields, temperature fields and structural mechanics fields, wherein the temperature field outputs material property changes to the electromagnetic field in real time, and the structural mechanics field outputs rotor deformation to the electromagnetic field in real time; Correcting thermal parameter drift and mechanical deformation error in the electromagnetic field online through a dynamic parameter compensator; Generating a parameter response surface based on a Kriging surrogate model to replace nonlinear equations for calculating key physical fields by finite element method; Outputting the joint simulation results of electromagnetic torque ripple spectrum, winding temperature distribution and rotor vibration mode.

[0007] Further, the bidirectional coupling of the multi-physical fields is realized through a weak coupling iteration paradigm. The electromagnetic field outputs iron loss power density and copper loss power density to the temperature field, and the temperature field feeds back a thermal conductivity matrix to the electromagnetic field. The structural mechanics field outputs rotor dynamic eccentricity to the electromagnetic field, and the electromagnetic field feeds back radial electromagnetic force density to the structural mechanics field. Through a cross-physical field real-time interaction mechanism, error accumulation caused by one-way coupling is suppressed, and the consistency of the joint simulation of the electromechanical system is improved.

[0008] Further, the dynamic parameter compensator dynamically adjusts resistivity parameters according to winding temperature rise, the mapping relationship is calibrated through material aging experiments, and the air gap permeance function is reconstructed according to rotor centrifugal deformation to suppress inductance nonlinear deviation caused by magnetic saturation, so as to eliminate electromagnetic parameter nonlinear distortion caused by thermal expansion and centrifugal deformation, and ensure the continuity of inductance and flux linkage prediction in the whole working condition range.

[0009] Further, the input variables of the Kriging surrogate model include stator current space vector amplitude, cooling medium flow rate and rotor angular velocity, and the output response includes permanent magnet eddy current loss density, end winding leakage inductance and rotor sheath equivalent stress; The model generates a response surface with a confidence greater than a certain threshold in the parameter space by using an adaptive sample density strategy, thereby realizing real-time solving of the multi-physical field equation while retaining the characteristics of spatial harmonics and saturation effects.

[0010] Further, the adaptive sample density strategy increases the sample point density in the region where the parameter gradient change rate is greater than a set threshold, and removes redundant sample points in the parameter space based on the Euclidean distance constraint, so as to generate a high-confidence response surface with a minimum sample amount, avoid the risk of overfitting of the surrogate model, and guarantee the prediction generalization ability.

[0011] Further, the method freezes the finite element mesh update of the stator region with stable boundary conditions through a computing resource dynamic allocation mechanism, and allocates the electromagnetic field eddy current loss calculation, temperature field heat conduction solving and structure field modal analysis tasks to independent computing units through a heterogeneous parallel architecture, so that the computing resource demand of the high-fidelity multi-physical field model can be significantly reduced, and the real-time constraint of online simulation of the embedded platform can be met.

[0012] Further, when the method is applied to a high-speed permanent magnet motor, the rotor dynamics stability constraint module is used to dynamically adjust the bearing support stiffness coefficient according to the critical speed offset, and an electromagnetic load reconstruction instruction is triggered when the eddy current loss density exceeds the material tolerance threshold, so as to simultaneously suppress the risk of high-speed rotor dynamics instability and local overheating, and expand the reliability boundary of the model under extreme working conditions.

[0013] Further, the method further includes a curvature-driven dual-mode switching mechanism, which is used to switch to the improved fast Fourier transform method to retain key harmonic components in a high-curvature region based on the local curvature characteristics of the parameter curve, and to ensure the continuity of the first derivative in a low-curvature region by using the least square method, and the formula is as follows:

[0014] In the formula, is the least square method weight, when the full weight is used, the least square method is used; is a sensitive coefficient, which is set through a calibration experiment, and a typical value is ; is the second derivative of the parameter curve, i.e., the curvature change rate.

[0015] Further, the method further includes a dynamic relaxation factor node voltage iteration method, which is used to dynamically adjust the iteration relaxation factor by evaluating the degree of ill-conditioning of the Jacobian matrix in real time, and the formula is as follows:

[0016] wherein, is a dynamic relaxation factor; is a minimum relaxation boundary; is a maximum relaxation boundary; is a damping coefficient; is a Jacobian matrix condition number.

[0017] In a second aspect, the present application also provides a curvature-driven dynamic compensation motor multi-physical field modeling system, which is based on the method of the aforementioned first aspect and comprises: a multi-field coupling solving engine integrating an electromagnetic finite element solver, a thermal network model solver and a rotor dynamics equation solver; a dynamic parameter database for storing material attribute matrices and geometric topology data updated in real time with temperature and deformation; a surrogate model acceleration interface for calling a Kriging response surface to perform online prediction of electromagnetic parameters.

[0018] Further, the dynamic parameter database is associated with a material degradation evaluation unit for recording a thermal aging index of insulating materials and predicting residual life and a deformation accumulation monitoring unit for counting rotor fatigue load cycles and outputting deformation accumulation, so as to improve the prediction accuracy of full-life-cycle parameter compensation by fusing long-term service degradation factors.

[0019] Further, when the system is deployed on an embedded platform, the system compresses surrogate model data volume through a topology structure pruning algorithm and adopts a time-sensitive network protocol to synchronize motor controller running state data packets in real time, so as to realize high-concurrency real-time simulation of vehicle-mounted edge devices and provide online decision support for motor predictive maintenance.

[0020] The present application constructs a finite element model according to motor stator and rotor structures and winding arrangement parameters, generates original inductance and flux linkage parameter curves by scanning current combinations in multiple working conditions through equivalent current vectors, performs fast Fourier transform analysis on the parameter curves, extracts direct current components and main harmonic components for inductance parameters, introduces a harmonic number weighted sorting mechanism for permanent magnet flux linkage parameters, retains key harmonic characteristics, adopts a weighted least square method to force the first derivative of the parameter curves to be continuous, and ensures phase fidelity; the smoothed parameters are also stored in an offline data table, winding and bar are taken as units to construct circuit equations in a circuit simulator, nonlinear equations of an electrical system are solved synchronously based on a node voltage method, and mechanical and circuit models are weakly coupled and iterated through electromagnetic torque and position parameters.

[0021] The method effectively filters out numerical noise introduced by grid discretization in finite element scanning through a smoothing mechanism, suppresses the harmonic error transmission of the parameter curve in the derivation operation, and significantly reduces the torque pulsation simulation deviation; the smoothed parameter curve guarantees the convergence of the node voltage method iteration, reduces the divergence risk in the strong nonlinear working condition, and the weak coupling architecture separates the circuit and mechanical equation solving, reducing the computational complexity; both harmonic feature preservation and dynamic process fidelity are considered, making the model have significant advantages in aerospace motor fault analysis, rail transit system simulation and other high reliability scenarios.

[0022] Advantages By implementing the above-mentioned curvature-driven dynamic compensation motor multi-physical field modeling method, the following technical effects are achieved: (1) By performing fast Fourier transform analysis on the inductance and flux linkage curves and introducing a harmonic number weighting sorting mechanism, the intrinsic harmonic characteristics and numerical noise are effectively separated. This method solves the non-smoothness problem of the curve caused by grid discretization in finite element scanning, eliminates the false high-frequency components introduced by numerical calculation while preserving the spatial harmonic and saturation effect characteristics of the reconstructed parameter curve, and significantly improves the accuracy consistency of the model in steady and transient working conditions.

[0023] (2) By mathematical optimization constraints, the first derivative of the inductance and flux linkage parameter curves is forced to be continuous and smooth, and the time domain continuity characteristics of the parameter curve are reconstructed. Compared with the frequency domain reconstruction method, this technology significantly improves the phase fidelity of the parameter curve, effectively suppresses the torque pulsation calculation deviation caused by phase distortion in motor acceleration and locked-rotor dynamic process simulation, enhances the prediction accuracy of the lumped parameter model for electromechanical transient processes, and provides convergence guarantee for the node voltage method solver.

[0024] (3) The curvature feature-driven smoothing processing mode is adaptively switched, the least square method is used in the low curvature area to ensure the continuity of the derivative, and the improved fast Fourier transform is used in the high curvature area to preserve the key harmonic components, realizing the collaborative optimization of harmonic fidelity and derivative continuity. This method eliminates the inherent limitations of traditional single methods in inductance curve processing, significantly reduces the electromagnetic torque harmonic distortion rate in transient condition simulation, greatly attenuates the oscillation amplitude of the node voltage iteration convergence trajectory, and fundamentally avoids the risk of harmonic amplification of numerical errors in the derivation operation, providing a high-fidelity parameter basis for electromechanical coupling simulation.

[0025] (4) Real-time feedback adjustment of relaxation factor by Jacobian matrix condition number, to establish adaptive balance mechanism of node voltage iteration stability and efficiency. In strong nonlinear working condition, this method automatically shrinks step length to inhibit divergence according to the degree of ill-conditioned equation system; in good condition, it expands step length to accelerate convergence. This method synchronously improves the stability of solution in extreme transient state and the calculation efficiency in normal condition, significantly reduces the sensitivity of electromechanical system joint simulation to initial value, and improves the robustness of modeling method to industrial application level. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to make the above-mentioned curvature driving dynamic compensation motor multi-physical field modeling method of the present application more obvious and easy to understand, the drawings needed in the specific embodiment of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained from these drawings without creative labor for those skilled in the art.

[0027] Figure 1 The method flowchart of the present application is shown; Figure 2 The schematic diagram of bidirectional coupling is shown; Figure 3 The schematic diagram of parameter curve smoothing is shown. DETAILED DESCRIPTION

[0028] Example 1: A curvature driving dynamic compensation motor multi-physical field modeling method is provided, and the method flow is shown as Figure 1 including: Constructing a bidirectional coupling basic model of electromagnetic field, temperature field and structural mechanics field, wherein the temperature field outputs material property change amount to the electromagnetic field in real time, and the structural mechanics field outputs rotor deformation amount to the electromagnetic field in real time; Correcting the thermal parameter drift and mechanical deformation error in the electromagnetic field by a dynamic parameter compensator; Generating a parameter response surface based on a Kriging surrogate model to replace the nonlinear equation of finite element calculation of key physical field; Outputting the joint simulation results of electromagnetic torque ripple spectrum, winding temperature distribution and rotor vibration mode.

[0029] The multi-physical field bidirectional coupling is realized by a weak coupling iteration paradigm, as shown in Figure 2 ; The electromagnetic field outputs iron loss power density and copper loss power density to the temperature field, and the temperature field feeds back thermal conductivity matrix to the electromagnetic field; The structural mechanics field outputs rotor dynamic eccentricity to the electromagnetic field, and the electromagnetic field feeds back radial electromagnetic force density to the structural mechanics field.

[0030] The dynamic parameter compensator adjusts resistivity parameters according to winding temperature rise dynamics, and reconstructs air gap permeance function according to rotor centrifugal deformation variables to suppress inductance nonlinear deviation caused by magnetic saturation. Through testing, under the transient working condition of motor rated load and sudden increase of rotating speed from 3000 rpm to 15000 rpm, the prediction error of air gap magnetic flux density fundamental amplitude is reduced to within ± 3% from ± 12% of the traditional offline table lookup method, and the nonlinear deviation of inductance parameters is reduced by more than 80%.

[0031] The parameter curve smoothing processing is as shown in Figure 3 The input variables of the Kriging surrogate model include stator current space vector amplitude, cooling medium flow rate and rotor angular velocity, and the output response includes permanent magnet eddy current loss density, end winding leakage inductance and rotor sheath equivalent stress; The model generates a response surface with a confidence greater than a certain threshold in the parameter space by using an adaptive sample density strategy.

[0032] The adaptive sample density strategy increases the sample point density in the region where the parameter gradient change rate is greater than a set threshold, and removes redundant sample points in the parameter space based on the Euclidean distance constraint.

[0033] The method updates the stator region frozen finite element grid through a computing resource dynamic allocation mechanism with stable boundary conditions, and distributes electromagnetic field eddy current loss calculation, temperature field heat conduction solving and structure field modal analysis tasks to independent computing units through a heterogeneous parallel architecture.

[0034] The method is applied to a high-speed permanent magnet motor model with a rated power of 200 kW for verification. As shown in Table 1, under various typical working conditions, the calculation accuracy and efficiency of the method are significantly better than those of the traditional method.

[0035] Table 1, performance comparison between the method and the traditional method

[0036] When the method is applied to a high-speed permanent magnet motor, the rotor dynamics stability constraint module is used to dynamically adjust the bearing support stiffness coefficient according to the critical speed deviation, and an electromagnetic load reconstruction instruction is triggered when the eddy current loss density exceeds the material tolerance threshold.

[0037] A curvature-driven dynamic compensation motor multi-physical field modeling system is also provided, and the system comprises: A multi-field coupled solving engine integrating an electromagnetic finite element solver, a thermal network model solver and a rotor dynamics equation solver; A dynamic parameter database for storing material attribute matrices and geometric topology data updated in real time with temperature and deformation; A proxy model acceleration interface is used to invoke a Kriging response surface to perform online prediction of electromagnetic parameters.

[0038] The dynamic parameter database is associated with a material degradation evaluation unit for recording a thermal aging index of the insulating material and predicting a remaining life, and a deformation accumulation monitoring unit for counting a number of rotor fatigue load cycles and outputting a deformation accumulation amount.

[0039] When the system is deployed on an embedded platform, the proxy model data volume is compressed by a topology pruning algorithm, and a time-sensitive network protocol is used to synchronize motor controller running state data packets in real time.

[0040] Embodiment 2: To solve the contradiction between harmonic preservation and phase continuity in traditional smoothing methods, a curvature-driven dual-mode switching mechanism is proposed. Based on the local curvature characteristics of the parameter curve, the least square method is used in the low curvature area to ensure the continuity of the first derivative, and the improved fast Fourier transform method is switched to in the high curvature area to preserve the key harmonic components. By introducing an adaptive weight function, the contribution weights of the two methods are dynamically allocated, as follows:

[0041] In the formula, is the weight of the least square method, is the full weight of the least square method; is the sensitivity coefficient, which is set through calibration experiments, and the typical value is ; is the second derivative of the parameter curve, that is, the rate of change of curvature.

[0042] The second derivative of the original inductance and flux linkage curve is calculated ; When , the weighted least square method is used for smoothing; When , the improved fast Fourier transform method is enabled: Harmonic component selection criteria: , in which is the th harmonic component; is the fast Fourier transform coefficient; is the harmonic number; only the components of are retained to reconstruct the curve; The final curve , in which is the final fused curve; is the output curve of the weighted least square method; is the output curve of the fast Fourier transform reconstruction.

[0043] Verification shows that, while achieving a similar average error to the above embodiments, compared to the traditional single FFT reconstruction method, the root mean square relative error of harmonic amplitude is reduced from 5.2% to 2.1%, a reduction of approximately 59.6% in harmonic amplitude error, while simultaneously increasing the iteration convergence speed by 28%. The results indicate that the curvature-driven dual-mode switching mechanism achieves synergistic optimization of harmonic fidelity and derivative continuity. In the simulation experiment, an improved Fast Fourier Transform (FFT) method is used for the high curvature region of the A-phase winding inductance curve of the permanent magnet motor to accurately capture spatial harmonic components; in the low curvature region, the weighted least squares method is switched to ensure continuous and smooth first-order derivatives. This method significantly suppresses the inherent contradictions of the traditional single smoothing method, avoiding phase distortion caused by FFT reconstruction and overcoming the smoothing effect of the least squares method on higher harmonics. In the simulation of transient conditions such as motor start-up and stall, the processed parameter curves reduce the harmonic distortion rate of the electromagnetic torque ripple spectrum to below an acceptable threshold, and the oscillation amplitude of the node voltage iterative convergence trajectory is reduced year-on-year. This confirms that it eliminates the risk of harmonic amplification of numerical errors in the derivative calculation from the root, providing a fundamental guarantee for high-precision electromechanical coupling simulation.

[0044] Example 3: To address the slow convergence of the nodal voltage method in strongly nonlinear systems, a dynamic relaxation factor based on the condition number of the Jacobian matrix is ​​proposed. The iterative relaxation factor is dynamically adjusted by evaluating the ill-conditioning of the Jacobian matrix in real time, as shown in the following formula:

[0045] In the formula, It is a dynamic relaxation factor; It is the minimum relaxation boundary; This represents the maximum relaxation boundary. The attenuation coefficient; Let be the condition number of the Jacobian matrix.

[0046] Calculate in each iteration condition number; like The condition is serious. Small step sizes prevent divergence; like Good state Larger step sizes accelerate convergence; Voltage update: In the formula, For the first The node voltage vector of the next iteration; For the first The node voltage vector of the next iteration; This represents the residual of the circuit equation.

[0047] The effect of the dynamic relaxation factor node voltage iteration method is shown in Table 2.

[0048] Table 2, effect summary of the dynamic relaxation factor node voltage iteration method Operating condition Conventional nodal voltage method Dynamic relaxation factor nodal voltage iterative method Rated steady state 35 steps to converge 22 steps to converge Short circuit fault transient Divergence 58 steps to converge Time consuming to calculate 100% 62% The verification shows that in the case of obtaining an average error similar to the above-mentioned embodiment, the convergence rate of the strong nonlinear working condition is improved by 100%, and the overall simulation speed is improved by 38%. Experimental data show that the dynamic relaxation mechanism based on the condition number of the Jacobian matrix breaks through the convergence bottleneck of the node voltage method in the strong nonlinear working condition. The simulation experiment constructs a motor system model containing short-circuit faults and load mutations. The dynamic relaxation factor is automatically adjusted according to the ill-conditioning degree of the equation set. When the ill-conditioning degree is high, the step length is contracted to suppress divergence, and when the ill-conditioning degree is low, the step length is expanded to accelerate convergence. This method simultaneously realizes the synchronous leap of stability and efficiency in the rated steady state and extreme transient state working conditions. Especially in the comparison of the traditional fixed relaxation factor benchmark test, the short-circuit fault simulation is changed from iteration divergence to stable convergence, and the iteration step sequence in the rated working condition presents an exponential decay characteristic. The smoothness of the electromagnetic torque and speed dynamic response curve verifies that this method significantly reduces the sensitivity of the solving process to the initial value, making the robustness of the multi-physical field joint simulation cross the industrial application level.

Claims

1. A method for multi-physical field modeling of a curvature-driven dynamic compensation motor, characterized in that, Comprise: A bidirectional coupling basic model of electromagnetic field, temperature field and structural mechanics field, wherein the temperature field outputs material property variation to the electromagnetic field in real time, and the structural mechanics field outputs rotor deformation to the electromagnetic field in real time; A dynamic parameter compensator is used to correct thermal-induced parameter drift and mechanical deformation error in the electromagnetic field in real time; Based on the local curvature characteristics of the parameter curve, a curvature-driven bimodal adaptive switching mechanism is used to smooth the electromagnetic parameter curve, wherein a time-domain continuous optimization mode is used to reconstruct the curve in a low curvature area, a frequency-domain harmonic reservation mode is used to reconstruct the curve in a high curvature area, and a dynamic weight factor is generated according to the curvature distribution characteristics to fuse the output results of the two types of modes; A parameter response surface is generated based on a Kriging surrogate model to replace the nonlinear equation of the finite element calculation of the key physical field; The joint simulation results of the electromagnetic torque ripple spectrum, winding temperature distribution and rotor vibration mode are output.

2. The method of claim 1, wherein: The multi-physical field bidirectional coupling is realized through a weak coupling iteration paradigm; The electromagnetic field outputs iron loss power density and copper loss power density to the temperature field, and the temperature field feeds back a thermal conductivity matrix to the electromagnetic field; The structural mechanics field outputs rotor dynamic eccentricity to the electromagnetic field, and the electromagnetic field feeds back radial electromagnetic force density to the structural mechanics field.

3. The method of claim 1, wherein: The dynamic parameter compensator adjusts the resistivity parameter according to the winding temperature rise and reconstructs the air gap permeance function according to the rotor centrifugal deformation to suppress the inductance nonlinear deviation caused by magnetic saturation.

4. The method of claim 1, wherein: The input variables of the Kriging surrogate model include stator current space vector amplitude, cooling medium flow rate and rotor angular velocity, and the output responses include permanent magnet eddy current loss density, end winding leakage inductance and rotor sheath equivalent stress; The model uses an adaptive sample density strategy to generate a response surface with a confidence greater than a certain threshold in the parameter space.

5. The method of claim 4, wherein: The adaptive sample density strategy increases the sample point density in the area where the parameter gradient change rate is greater than a set threshold, and removes redundant sample points in the parameter space based on the Euclidean distance constraint.

6. The method of claim 1, wherein: The method freezes the finite element mesh update of the stator region with stable boundary conditions through a computing resource dynamic allocation mechanism, and distributes the electromagnetic field eddy current loss calculation, temperature field heat conduction solving and structural field modal analysis tasks to independent computing units through a heterogeneous parallel architecture.

7. The method of claim 1, wherein: When the method is applied to a high-speed permanent magnet motor, a rotor dynamics stability constraint module is used to dynamically adjust the bearing support stiffness coefficient according to the critical speed offset, and an electromagnetic load reconstruction instruction is triggered when the eddy current loss density exceeds the material tolerance threshold.

8. A curvature drive dynamic compensation motor multi-physical field modeling system, characterized in that: The system is implemented based on the method of any one of claims 1-7, comprising: A multi-field coupling solver engine integrating an electromagnetic finite element solver, a thermal network model solver and a rotor dynamics equation solver; A dynamic parameter database is configured to store material property matrix and geometric topology data which are updated in real time with temperature and deformation; A proxy model acceleration interface is configured to invoke the Kriging response surface to perform online prediction of electromagnetic parameters.

9. The system of claim 8, wherein: The dynamic parameter database is associated with a material degradation evaluation unit configured to record thermal aging index of the insulation material and predict remaining life, and a deformation accumulation monitoring unit configured to count fatigue load cycles of the rotor and output deformation accumulation.

10. The system of claim 8, wherein: When the system is deployed on an embedded platform, a topology pruning algorithm is used to compress the data volume of the proxy model, and a time-sensitive network protocol is used to synchronize the motor controller operating state data packet in real time.

Images