Novel magnetic material high-frequency magnetization dynamic hysteresis model construction and multi-field coupling loss prediction method

Abstract

The invention relates to the field of high-frequency power equipment design and magnetic material application. A novel magnetic material high-frequency magnetization dynamic hysteresis model construction and multi-field coupling loss prediction method is characterized by comprising the following steps: 1) high-frequency magnetization data acquisition: taking an iron-based amorphous alloy as an object, and constructing an excitation generation-magnetic field application-data acquisition-environment control experiment platform; 2) dynamic hysteresis model construction: a dynamic hysteresis model construction module is based on a mesoscopic-macroscopic cross-scale modeling idea; 3) multi-field coupling loss quantification: a multi-field coupling loss quantification module realizes cross-physics field coupling of a temperature field and a frequency field; and 4) hybrid algorithm parameter optimization: a hybrid algorithm parameter optimization module adopts a genetic algorithm and particle swarm optimization collaborative strategy. The method can realize hysteresis characteristic accurate modeling and loss prediction of a novel magnetic material under high-frequency, wide-temperature-range and multi-excitation working conditions, and improves the efficiency and reliability of new energy power equipment.

Description

Technical Field

[0001] This invention relates to the field of high-frequency power equipment design and magnetic material application, specifically to a novel method for constructing a dynamic hysteresis model of high-frequency magnetization of magnetic materials and predicting multi-field coupling losses, applicable to the performance optimization and loss analysis of magnetic materials in high-frequency transformers and new energy power equipment. Background Technology

[0002] As new energy power systems develop towards higher frequencies and greater integration, novel magnetic materials (such as amorphous alloys and nanocrystalline alloys) are increasingly widely used in high-frequency transformers, energy storage devices, and other fields. Under high-frequency operating conditions (10kHz-100kHz), the magnetization characteristics and loss mechanisms of magnetic materials have become the core bottlenecks restricting equipment efficiency and reliability. Accurate modeling and prediction of these characteristics are urgently needed, but current technologies face multiple challenges.

[0003] The high-frequency performance of novel magnetic materials is determined by mesoscopic processes such as nanoscale grain structure, domain wall motion, and eddy current effects. However, traditional macroscopic models only fit magnetization curves with empirical parameters without coupling with mesoscopic physical laws, resulting in high-frequency hysteresis loop prediction errors exceeding 20%. Furthermore, the coupling effects of magnetic, thermal, and stress fields are significant in high-frequency environments. Existing models mostly analyze single physical fields independently, lacking cross-scale coupling modeling capabilities and failing to reveal the comprehensive impact of multi-field interactions on losses.

[0004] In multi-excitation scenarios, magnetic materials in new energy applications often experience combined excitation from non-sinusoidal waveforms and DC bias. Traditional models, based on the assumption of sinusoidal excitation, cannot decompose the contribution of harmonic components of non-sinusoidal waveforms to hysteresis loss, nor can they quantify the changes in core saturation caused by DC bias. The overall prediction error of multi-excitation is generally over 25%. Furthermore, the parameters of magnetic materials exhibit strong frequency dependence over a wide frequency range. Traditional models do not incorporate frequency adaptive mechanisms, resulting in insufficient generalization ability across the entire frequency band, with prediction accuracy differences of up to 30% between low and high frequencies.

[0005] In terms of parameter optimization and experimental verification, the optimization of magnetic materials needs to consider multiple objectives such as minimizing losses, reducing size, and controlling temperature rise. Traditional genetic algorithms have drawbacks such as being single-objective-oriented and having fixed search strategies, making them prone to getting trapped in local optima and having slow convergence speeds. High-frequency magnetic testing relies on specialized equipment, with a single full-condition test costing over a thousand yuan. After parameter adjustments, retesting is required, and the experimental cycle can last for several weeks. Existing methods lack an efficient "model prediction-verification and correction" closed-loop process, resulting in low R&D efficiency.

[0006] Current technologies have limitations in mesoscopic physical modeling, multi-excitation adaptability, parameter optimization efficiency, and experimental verification costs. There is an urgent need for more efficient theories and methods to meet the requirements of new energy power equipment for accurate modeling and loss control of high-frequency magnetic materials. Summary of the Invention

[0007] The purpose of this invention is to provide a novel method for constructing a dynamic hysteresis model of high-frequency magnetization of magnetic materials and predicting multi-field coupling losses. This method can achieve accurate modeling and loss prediction of the hysteresis characteristics of novel magnetic materials under high frequency, wide temperature range and multi-excitation conditions, thereby improving the efficiency and reliability of new energy power equipment.

[0008] To achieve the above objectives, the technical solution adopted by this invention is a novel method for constructing a dynamic hysteresis model of high-frequency magnetization in magnetic materials and predicting multi-field coupling losses. This method mainly includes high-frequency magnetization data acquisition, dynamic hysteresis model construction, multi-field coupling loss quantification, and hybrid algorithm parameter optimization. Its characteristics include the following steps: 1) High-frequency magnetization data acquisition: Using iron-based amorphous alloys as the object, an experimental platform was built consisting of "excitation generation - magnetic field application - data acquisition - environmental control". The excitation module uses a Keysight M9384A signal generator to output a 10kHz-200kHz sine wave, which is amplified by a power amplifier to drive a Helmholtz coil to generate a uniform magnetic field, and a 0-1.2T DC bias is superimposed. The acquisition module uses a Lake Shore flux sensor and an NI data acquisition card to synchronously record H(t) and B(t), where H(t) is the magnetic field strength and B(t) is the magnetic flux density, with a sampling rate of 1MS / s. Environmental control is achieved by maintaining the temperature at 25℃±0.5℃ using an ESPEC constant temperature chamber. The operating conditions are designed to cover frequencies of 10kHz, 50kHz, and 200kHz and a bias range of 0-1.2T. Data preprocessing uses moving average filtering and Hilbert transform to ensure an error of <0.5%. 2) Dynamic hysteresis model construction: The dynamic hysteresis model construction module is based on the mesoscopic-macroscopic cross-scale modeling approach. At the mesoscopic level, the Landau-Lifshitz-Gilbert (LLG) equations are used to describe the dynamic response of the magnetic moment, coupled with physical fields such as external field, demagnetizing field, and magnetocrystalline anisotropy field. The magnetization intensity is calculated iteratively using the fourth-order Runge-Kutta method. At the macroscopic level, the hysteresis loop is divided into a reversible magnetization region (|H|<50A / m), an irreversible magnetization region (50A / m≤|H|<500A / m), and a saturation region (|H|≥500A / m). Boundary interpolation is used to ensure the continuous smoothness of the loop, and the area error of the hysteresis loop is controlled within 5%. 3) Multi-field coupling loss quantization: The multi-field coupling loss quantization module realizes cross-physical field coupling between the temperature field and the frequency field; the temperature field affects the shape of the hysteresis loop through the saturation magnetization attenuation model. In the formula, The saturation magnetization at temperature T is... T is the room temperature saturation magnetization. c Here, T represents the Curie temperature; T represents the real-time temperature; the frequency field alters the dynamic loss characteristics through the relaxation time frequency sensitivity (\(\tau\propto f^{-1}\)) and the skin effect correction factor; hysteresis loss is calculated based on the area integral of the loop, eddy current loss is combined with the skin effect model, excess loss is quantified through the high-frequency domain wall resonance coefficient, and the total loss is superimposed and predicted over a wide temperature range using a temperature correction formula. Under multi-field coupling, the total loss error is less than 5%. In the formula, This represents the total loss at temperature T. The total loss is at the reference temperature T0; T represents the temperature coefficient of the loss; T is the real-time temperature; T0 is the reference temperature. 4) Hybrid Algorithm Parameter Optimization: The hybrid algorithm parameter optimization module adopts a collaborative strategy of genetic algorithm (GA) and particle swarm optimization (PSO); GA uses reversible permeability u rev Pinning strength k, JA model coupling coefficient α、 Excess loss coefficient ke, loss temperature coefficient Five parameters are the optimization targets. A global search is performed using tournament selection, arithmetic crossover, and Gaussian-uniform mixed mutation. The fitness function integrates hysteresis area error (weight 30%), loss error (60%), and stability constraints (10%). PSO extracts the top 20 best individuals in GA as initial particles, with initial velocities pointing towards the global optimum of GA. Local refinement is achieved through linear decay of inertia weights and frequency / temperature-sensitive perturbations, ultimately reducing the overall loss error E. P <5%, hysteresis loop error E A <3%.

[0009] The inventive point of this invention: 1. Innovative Dynamic Hysteresis Model Construction: For iron-based amorphous alloys, a dynamic hysteresis model from mesoscopic to macroscopic is constructed. Based on the LLG equation, a dynamic response model of the mesoscopic magnetic moment is established. The macroscopic hysteresis loop is divided into reversible magnetization region, irreversible magnetization region, and saturation region for segmented modeling. The magnetization intensity of each region is solved by the fourth-order Runge-Kutta method, and boundary interpolation is performed to ensure the continuity of the loop. This solves the problem that traditional macroscopic models do not reflect mesoscopic physical laws, thus reducing the prediction error of the hysteresis loop.

[0010] 2. Innovative Quantification of Multi-Field Coupling Losses: This innovation achieves coupling between the temperature field and the magnetic field, and between the frequency field and the magnetic field. The temperature field alters the shape of the hysteresis loop by affecting the saturation magnetization, while the frequency field influences dynamic loss characteristics through relaxation time and the skin effect. Hysteresis, eddy currents, and excess losses are quantified, calculated, and superimposed. A temperature correction formula is introduced to adapt to complex operating conditions, addressing the shortcomings of existing models such as lack of cross-scale coupling modeling capabilities and large prediction errors for multiple excitations.

[0011] 3. Innovative Hybrid Algorithm Parameter Optimization: A hybrid approach combining genetic algorithm and particle swarm optimization (PSO) is employed to optimize model parameters. The genetic algorithm performs a global search to generate initial individuals. Through selection, crossover, and mutation operations, it constructs a fitness function by integrating hysteresis loop area error, loss prediction error, and model stability constraints. The PSO algorithm initializes particles with the optimal solution from the genetic algorithm and performs local refinement through inertia weight decay and sensitive parameter perturbations, improving the model's prediction accuracy under wide-frequency domain and multi-field coupling conditions.

[0012] 4. Multi-dimensional verification system innovation: Design a three-dimensional verification matrix covering frequency, temperature, and excitation type, use a high-frequency magnetic field testing platform to collect data, and use various data processing and error analysis methods to verify the model under multiple working conditions, forming a closed-loop process of "model prediction - verification correction", which solves the problems of long experimental cycles and low R&D efficiency of existing methods.

[0013] The beneficial results of this invention are: overcoming the shortcomings of traditional macroscopic hysteresis models, such as lack of coupling with mesoscopic physical laws, insufficient analysis of multi-field interaction effects, large prediction errors in multi-excitation, and low efficiency of parameter optimization; achieving accurate modeling and loss prediction of the hysteresis characteristics of novel magnetic materials under high frequency, wide temperature range, and multi-excitation conditions; improving the efficiency and reliability of new energy power equipment; and promoting the engineering application of high-frequency magnetic materials in high-frequency transformers, energy storage devices, and other fields. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a novel method for constructing a dynamic hysteresis model for high-frequency magnetization of magnetic materials and predicting multi-field coupling losses, as described in this invention.

[0016] Figure 2 This is a schematic diagram of the high-frequency magnetization data acquisition platform of the present invention.

[0017] Figure 3This is a flowchart of the hybrid algorithm optimization process of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The complete steps of the method of the present invention are as follows: In one embodiment, such as Figure 1 As shown, a novel method for constructing a dynamic hysteresis model for high-frequency magnetization of magnetic materials and predicting multi-field coupling losses includes the following specific steps: Step S101, High-frequency magnetization data acquisition: In this step, iron-based amorphous alloys are used as a representative of novel magnetic materials. Experimental schemes are designed to study the high-frequency and low-loss characteristics of iron-based amorphous alloy magnetic materials, and magnetization data under typical working conditions are obtained to support the construction of dynamic hysteresis models.

[0020] In one embodiment, the experimental platform is built around the main steps of "excitation generation - magnetic field application - data acquisition - environmental control". The excitation module uses a Keysight M9384A signal generator to output a 10kHz-200kHz sine wave, which is amplified to a 10A current by an Amplifier Research power amplifier to drive a Helmholtz coil to generate a uniform magnetic field. Simultaneously, a 0-0.8T (Tesla) DC bias is superimposed through a Kikusui DC power supply. The acquisition module uses a Lake Shore 475 flux sensor to measure the magnetic flux density B, and a NI USB-6356 data acquisition card to simultaneously record the magnetic field strength H(t) and B(t), where H(t) is the magnetic field strength and B(t) is the magnetic flux density, with a sampling rate of 1MS / s. For environmental control, the sample is placed in an ESPEC LT-5 incubator, maintaining a temperature of 25℃±0.5℃ to suppress permeability fluctuations in the amorphous alloy caused by temperature drift.

[0021] The operating condition design combines sinusoidal excitation with DC bias, covering frequencies of 10kHz, 50kHz, and 200kHz, and a bias range of 0-1.2T. Taking 50kHz sinusoidal excitation as an example, with a voltage amplitude of 3V (corresponding to a coil current of 6A and a magnetic field strength of approximately 300A / m), hysteresis loop data is collected over 30 seconds, generating approximately 3×10⁻⁶ ohmmeters per operating condition. 4The data consists of 10 data points. In the data preprocessing stage, a 50-point moving average filter is used for noise reduction. The complex permeability is calculated using Hilbert transform and corrected for the permeability based on the temperature coefficient to ensure an error of <0.5%.

[0022] The experimental platform ensures data accuracy through Helmholtz coil calibration, signal link debugging, and synchronous acquisition verification, achieving a magnetic field uniformity >99% and a phase difference error <5°. Simultaneously, limitations are mitigated through Fourier decomposition extrapolation of non-sinusoidal conditions and multi-point temperature calibration, efficiently meeting the high-frequency data acquisition requirements of iron-based amorphous alloys.

[0023] Step S102, Construction of dynamic hysteresis model: In this step, a dynamic hysteresis model is constructed, which involves analyzing the mesoscopic mechanism and dividing the model into macroscopic components, based on the nanocrystalline structure and high-frequency magnetization characteristics of iron-based amorphous alloys.

[0024] Mesoscopic physical mechanism modeling: The unique properties of iron-based amorphous alloys originate from their composite structure of nanoscale grains and amorphous matrix. At the microscale, the distribution and motion of magnetic moments within the material are governed by various physical fields. Based on the Landau-Lifshitz-Gilbert (LLG) equations, a fundamental model describing the dynamic response of magnetic moments can be established: In the formula: / The overall value represents the time-varying rate of magnetization; M is the magnetization vector; γ is the gyromagnetic ratio; H eff For effective fields; M is the damping coefficient; s F is the saturation magnetization. th This is due to thermal expansion.

[0025] Where H eff It is the superposition of multiple physical fields: In the formula: H is the magnetic field strength of the applied magnetic field; H d For demagnetizing the field; H a For magnetocrystalline anisotropic field; H ex For the exchange field.

[0026] The LLG equations are solved using the fourth-order Runge-Kutta method, with a set time step based on the initial magnetization M0 and the effective field H. eff The magnetization intensity M(t) at different times is calculated iteratively.

[0027] After obtaining the basic framework for the dynamic response of magnetic moments in iron-based amorphous alloys through mesoscopic physical mechanism modeling, the macroscopic hysteresis loop is divided into reversible magnetization, irreversible magnetization, and saturation regions based on the mesoscopic modeling results. Modeling is performed separately for each region, with the magnetic field strength H as the dividing criterion: the boundary of the reversible magnetization region is set as (|H| < 50 A / m), where domain wall motion is mainly reversible displacement; the range of the irreversible magnetization region is (50 A / m < |H| < 500 A / m), corresponding to irreversible jumps where domain walls overcome pinning; the range of the saturation region is (|H| ≥ 500 A / m), where the magnetic moments are nearly perfectly oriented.

[0028] Specifically, in the reversible magnetization region, based on the reversible small displacements of mesoscopic domain walls and the calculation results of anisotropic and exchange fields, a linear equation is used to describe the relationship between magnetization M and magnetic field strength H: In the formula: / The overall magnetic susceptibility is reversible, and µ0 is the free permeability; µ rev H is the inverse permeability; rev This is the reversible threshold.

[0029] In the irreversible magnetization region, considering the irreversible jumping characteristic of mesoscopic domain walls overcoming the pinning barrier, the Jiles-Atherton model is introduced. Combining the damping coefficient and effective field to characterize the nonlinear magnetization behavior, parameters such as the pinning strength k = 200 A / m and the coupling coefficient α = 0.01 are first set. The nonhysteresis magnetization intensity is then calculated using formulas. : In the formula: α is the coupling coefficient; k is the pinning strength; Where M is the saturation magnetization, H is the magnetization, and H is the magnetic field strength. The fourth-order Runge-Kutta method is used to update the magnetic field strength H with a step size of 0.1 A / m, and then the obtained non-hysteresis magnetization is... Substitute into the following formula: In the formula: / The overall value represents the rate of change of magnetic flux density, δ is the direction factor, k is the pinning strength, and η is the dynamic loss coefficient. B is the magnetic flux density.

[0030] In the saturation region, based on the characteristic that mesoscopic magnetic moments tend to be perfectly aligned, a paramagnetic correction term is added to improve the macroscopic model. Based on the analysis of the magnetic moment saturation characteristics using the mesoscopic model, the magnetic field strength H is calculated with increments of 1 A / m starting from 500 A / m, and the magnetization M is calculated using the following formula: Where: H sat It is a saturated magnetic field; It represents the paramagnetic susceptibility.

[0031] After the models for each region are built, the difference in the calculated M values ​​at the boundaries of adjacent regions (H = 50 A / m and H = 500 A / m) is checked. If the difference exceeds the threshold, the boundary point data is adjusted by linear interpolation to ensure the hysteresis loop is continuous and smooth. Finally, the hysteresis loops calculated by the piecewise models are compared with the experimental data to calculate the errors in characteristic parameters such as loop area and coercivity. If the error exceeds 5%, the model parameters for each interval are adjusted until the accuracy requirements are met.

[0032] Step S103, loss quantification and prediction calculation for multi-field coupling extension: After completing the segmented construction of the macroscopic model, multi-field coupling extension is performed to adapt to the performance prediction requirements of iron-based amorphous alloys under actual complex working conditions.

[0033] The coupling between the temperature field and the magnetic field is mainly manifested in the thermal activation effect and the structural relaxation mechanism. Specifically, an increase in temperature will reduce the saturation magnetization Ms, and its calculation formula is as follows: In the formula: M is the saturation magnetization at temperature T. s (0) represents the room temperature saturation magnetization; T c T represents the Curie temperature; T represents the real-time temperature.

[0034] The coupling between the frequency field and the magnetic field is achieved through the frequency sensitivity of the relaxation time and the eddy current skin effect. Specifically, this manifests as the magnetic moment relaxation time under high-frequency excitation. The shortening calculation formula is: In the formula: f is the external field frequency; The time is the magnetic moment relaxation time. Eddy current loss is affected by the skin effect, and the calculation formula is as follows: In the formula: Where d is the eddy current loss and d is the strip thickness; B is the resistivity; m The peak value of the magnetic flux density; µ r λ is the high-frequency relative permeability; λ is the skin depth; f is the external field frequency.

[0035] In one embodiment, temperature T and external field frequency f are first acquired in real time. The magnetic parameters Ms(T) and dynamic response parameters τd are updated using the above coupling formula. Then, these parameters are substituted into the mesoscopic LLG equation (Landau-Rifshitz-Gilbert equation) and the macroscopic piecewise model to correct the effective field calculation and magnetization distribution. Finally, the coupling coefficient is optimized by combining multi-field experimental data and the algorithm to ensure that the prediction error of the model in multi-field environments is less than 5%.

[0036] After completing the multi-field coupling extension, the effects of the temperature and frequency fields are quantified into the loss calculation. Temperature affects the shape of the hysteresis loop by changing the saturation magnetization Ms(T), while frequency is affected by the relaxation time. τ The skin depth λ alters the dynamic loss characteristics.

[0037] Specifically, hysteresis loss P hyst The formula, calculated by integrating the area of ​​the hysteresis loop and combining it with the frequency, is as follows: In the formula, This is hysteresis loss; d is the magnetic field frequency; d is the strip thickness; M is the magnetization; eddy current loss is calculated based on the skin effect model, and the formula is: In the formula, The relative permeability is given by the formula for calculating excess loss, which originates from high-frequency domain wall resonance. In the formula: For excess loss, k e B is the excess loss factor. m denoted as peak magnetic flux density; f is the frequency of the external field.

[0038] Total loss is obtained by superimposing hysteresis, eddy current, and excess loss, as shown in the formula: In the formula: Total loss, To account for eddy current losses, a temperature correction formula is introduced to predict losses at different temperatures: In the formula: This represents the predicted total loss at temperature T. This represents the total loss at time T0, where T0 is the reference temperature and T is the actual temperature. This is the temperature coefficient.

[0039] Step S104, Hybrid Algorithm Parameter Optimization: In this step, the global optimization and high-precision calibration of the dynamic hysteresis model parameters are achieved through the collaborative optimization of the genetic algorithm (GA) and the particle swarm optimization (PSO).

[0040] Specifically, five core parameters to be optimized were extracted from the model established in S102, namely: reversible permeability u. rev Pinning strength k, JA model coupling coefficient α、 Excess loss coefficient ke, loss temperature coefficient Five parameters, with their value ranges set based on material physical properties: reversible permeability µ rev ∈[1000,10000]; pinning strength k∈[50,500]A / m; JA model coupling coefficient α∈[0.001,0.1]; excess loss coefficient k e ∈[10 -8 10 -4 Temperature coefficient αT∈[-5×10] -3 5×10 -1 ]K -1 .

[0041] A global search is performed using a genetic algorithm (GA), generating 100 initial individuals with real-number encoding. Each individual is a parameter vector [µ]. rev [, k, α, ke, αT], where the parameter values ​​follow a log-normal distribution. Then, a three-objective function is defined, incorporating hysteresis loop area error, loss prediction error, and model stability constraints: a three-objective fitness function that comprehensively considers model accuracy and physical consistency. In the formula: E represents the model fit. A E represents the area error of the hysteresis loop. P The average error of losses under multiple operating conditions; C is the stability constraint term; weighting coefficients. w A , w B , w C The values ​​are 0.3, 0.6, and 0.1, respectively.

[0042] Hysteresis loop area error E A The formula is: In the formula: A calc The area of ​​the hysteresis loop calculated by integrating the piecewise model for the reversible magnetization region, the irreversible magnetization region, and the saturation region; A exp The area of ​​the hysteresis loop as measured experimentally; is the loop integral symbol; d is the strip thickness; M is the magnetization intensity.

[0043] Average error E of multi-condition loss P The formula is: In the formula: This represents the total loss error; For each experimental condition, there are frequency / magnetic flux density sampling points; N represents the i-th loss value predicted by the model. case For different frequencies, temperatures, and excitation types; P calc,i Calculation of losses under different operating conditions; P exp,i This represents the loss measured in the experiment.

[0044] The formula for the stability constraint term C is: In the formula, α is the coupling coefficient; k is the pinning strength; Let be the saturation magnetization intensity. Based on this, genetic operations are performed, selecting operators by randomly selecting three individuals each time, retaining the one with the lowest fitness, with a 40% elimination rate, to ensure the population evolves towards the optimal solution. Arithmetic crossover is used, where offspring parameters are generated by a linear combination of parent parameters, with the formula: In the formula: X child λ is the parameter vector of the offspring individuals generated by the crossover operation; λ is a random number ∈ (0, 1). and These represent the parameter vectors of the two parent individuals participating in the crossover operation.

[0045] For high-frequency sensitive parameters ke and α T Gaussian mutation is used, with the step size proportional to the current value: In the formula: k e ′、α T The values ​​are divided into the excess loss coefficient after variation and the temperature coefficient after variation. , The standard deviation of the variation; α is a normally distributed random number; ke is a high-frequency sensitive parameter (ke is 1.2 × 10−4); T For high-frequency sensitive parameters (α) T 0.0025∘C -1 ).

[0046] For low-frequency parameters μrev and k adopt uniform variation with a perturbation amplitude of ±10%, and the formula is as follows: In the formula: k is the low-frequency parameter (k is 1500A / m); , It is divided into the value of reversible permeability after variation and the value of pinning strength after variation; The amplitude of the disturbance.

[0047] Particle Swarm Optimization (PSO) is used for local refinement. The top 20 best individuals after 50 generations of Genetic Algorithm (GA) iterations are extracted as the initial particles for PSO, with their initial velocities pointing towards the global optimum of GA. In the formula: Let be the initial velocity vector of the i-th particle; The parameter vector of the optimal individual obtained during the global search phase of the genetic algorithm (GA). Let be the initial position vector of the i-th particle.

[0048] Inertia weight w The performance decreases linearly with the number of iterations, balancing global search and local exploitation capabilities. In the formula: The inertia weight is the weight at iteration number t. The total number of iterations is set for the particle swarm optimization algorithm; t is the number of iterations.

[0049] The particles are updated, and the velocity update formula is: In the formula: Let be the velocity of particle i in the (t+1)th iteration; Let c1 and c2 be the learning factors (c1=2.0, c2=2.1) for the velocities of i particles at the t-th iteration; r1 and r2 be random numbers ∈ (0, 1); gbest represents the optimal positions experienced by the entire particle swarm; pbest represents the optimal positions experienced by the entire particle swarm. i Let i be the best historical position that each particle has experienced. Let be the position vector of the i particles at the t-th iteration; The inertia weights are used in the particle swarm optimization algorithm.

[0050] After each update, the parameters need to be projected back to the initially set value range to avoid exceeding the limit. The position update formula is: In the formula: Let i be the position of the i particles in the (t+1)th iteration; Let be the velocity of particle i in the (t+1)th iteration; Let be the position vector of the i particles at the t-th iteration.

[0051] For high-frequency loss coefficient k e Applying a frequency-dependent sinusoidal perturbation enhances the capture of resonance effects, thus adjusting the high-frequency loss coefficient. The calculation formula is: In the formula, f is the external field frequency; for the temperature coefficient α T The step size is adjusted based on the difference between the current temperature and the reference temperature, resulting in an adjusted temperature coefficient. The calculation formula is: Where T is the actual temperature; The temperature coefficient is used; the termination condition for particle swarm optimization is: the fitness change of the global optimum is less than 0.1% for 15 consecutive generations; under all test conditions, the loss error E P <5%, hysteresis loop area error E A <3%.

[0052] Specifically, the working condition matrix design is as follows: The obtained optimized parameters are substituted into the hysteresis model and loss prediction model in S102 and S103 to verify the error rate of the calculated and experimental parameters. Based on the verification results, the parameters are adjusted to realize the accurate construction of the hysteresis model and loss prediction of the new magnetic material.

[0053] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0054] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A novel method for constructing a dynamic hysteresis model of high-frequency magnetization in magnetic materials and predicting multi-field coupling losses, characterized in that... Includes the following steps: 1) High-frequency magnetization data acquisition: Using iron-based amorphous alloys as the object, an experimental platform was built consisting of "excitation generation - magnetic field application - data acquisition - environmental control". The excitation module uses a Keysight M9384A signal generator to output a 10kHz-200kHz sine wave, which is amplified by a power amplifier to drive a Helmholtz coil to generate a uniform magnetic field, and a 0-1.2T DC bias is superimposed. The acquisition module uses a Lake Shore flux sensor and an NI data acquisition card to synchronously record H(t) and B(t), where H(t) is the magnetic field strength and B(t) is the magnetic flux density, with a sampling rate of 1MS / s. The environmental control uses an ESPEC constant temperature chamber to maintain the temperature at 25℃±0.5℃. The operating conditions are designed to cover frequencies of 10kHz, 50kHz, and 200kHz and a bias range of 0-1.2T. Data preprocessing uses moving average filtering and Hilbert transform to ensure an error of <0.5%. 2) Construction of Dynamic Hysteresis Model: The dynamic hysteresis model construction module is based on the mesoscopic-macroscopic cross-scale modeling approach. At the mesoscopic level, the Landau-Lifshitz-Gilbert equation is used to describe the dynamic response of the magnetic moment, coupled with physical fields such as external field, demagnetizing field, and magnetocrystalline anisotropic field. The magnetization intensity is calculated iteratively using the fourth-order Runge-Kutta method. At the macroscopic level, the hysteresis loop is divided into reversible magnetization region, irreversible magnetization region, and saturation region. Boundary interpolation is used to ensure the continuous and smoothness of the loop, and the area error of the hysteresis loop is controlled within 5%. 3) Multi-field coupling loss quantization: The multi-field coupling loss quantization module realizes cross-physical field coupling between the temperature field and the frequency field; the temperature field affects the shape of the hysteresis loop through the saturation magnetization attenuation model. In the formula, The saturation magnetization at temperature T is... T is the room temperature saturation magnetization. c T represents the Curie temperature; T represents the real-time temperature; the frequency field alters the dynamic loss characteristics through relaxation time frequency sensitivity and skin effect correction factors; hysteresis loss is calculated based on the area integral of the loop, eddy current loss is combined with the skin effect model, excess loss is quantified through the high-frequency domain wall resonance coefficient, and the total loss is superimposed and predicted over a wide temperature range using a temperature correction formula. The total loss error under multi-field coupling is less than 5%. In the formula, This represents the total loss at temperature T. The total loss is at the reference temperature T0; T represents the temperature coefficient of the loss; T is the real-time temperature; T0 is the reference temperature. 4) Hybrid Algorithm Parameter Optimization: The hybrid algorithm parameter optimization module adopts a collaborative strategy of genetic algorithm and particle swarm algorithm; GA with reversible permeability u rev Pinning strength k, JA model coupling coefficient α、 Excess loss coefficient ke, loss temperature coefficient Five parameters are the optimization targets. A global search is performed using tournament selection, arithmetic crossover, and Gaussian-uniform mixed mutation. The fitness function integrates hysteresis area error, loss error, and stability constraints. PSO extracts the top 20 best individuals in GA as initial particles, with initial velocities pointing towards the global optimum of GA. Local refinement is achieved through linear decay of inertial weights and frequency / temperature-sensitive perturbations, ultimately reducing the overall loss error E under all operating conditions. P <5%, hysteresis loop error E A <3%.

2. The novel method for constructing a dynamic hysteresis model for high-frequency magnetization of magnetic materials and predicting multi-field coupling losses according to claim 1, characterized in that... The specific steps are as follows: Step S101, High-frequency magnetization data acquisition: Taking iron-based amorphous alloys as a representative of novel magnetic materials, we designed an experimental scheme to study the high-frequency and low-loss characteristics of iron-based amorphous alloy magnetic materials, obtained their magnetization data under typical working conditions, and supported the construction of a dynamic hysteresis model. The experimental platform was built around the main steps of "excitation generation - magnetic field application - data acquisition - environmental control". The excitation module used a Keysight M9384A signal generator to output a 10kHz-200kHz sine wave, which was amplified to a 10A current by an Amplifier Research power amplifier to drive a Helmholtz coil to generate a uniform magnetic field. At the same time, a 0-0.8T DC bias was superimposed through a Kikusui DC power supply. The acquisition module used a Lake Shore 475 flux sensor to measure the magnetic flux density B, and used an NI USB-6356 data acquisition card to synchronously record the magnetic field strength H(t) and B(t), where H(t) is the magnetic field strength and B(t) is the magnetic flux density, with a sampling rate of 1MS / s. For environmental control, the sample was placed in an ESPEC LT-5 constant temperature chamber and the temperature was maintained at 25℃±0.5℃ to suppress the permeability fluctuation of the amorphous alloy caused by temperature drift. The operating condition design combines sinusoidal excitation with DC bias, covering frequencies of 10kHz, 50kHz, and 200kHz, and a bias range of 0-1.2T. Taking 50kHz sinusoidal excitation as an example, the voltage amplitude is set to 3V, corresponding to a coil current of 6A and a magnetic field strength of approximately 300A / m. Hysteresis loop data is collected over 30 seconds, generating approximately 3×10⁻⁶ ohmmeters per operating condition. 4 One data point; In the data preprocessing stage, a 50-point moving average filter is used for noise reduction. The complex permeability is calculated using Hilbert transform and corrected for the permeability based on the temperature coefficient to ensure that the error is <0.5%. The experimental platform ensures data accuracy through Helmholtz coil calibration, signal link debugging, and synchronous acquisition verification, with magnetic field uniformity >99% and phase difference error <5°. At the same time, it compensates for limitations by Fourier decomposition extrapolation of non-sinusoidal conditions and multi-point temperature calibration, efficiently meeting the high-frequency data acquisition requirements of iron-based amorphous alloys. Step S102, Construction of dynamic hysteresis model: Based on the nanocrystalline structure and high-frequency magnetization characteristics of iron-based amorphous alloys, a dynamic hysteresis model is constructed, which involves analysis of the mesoscopic mechanism and partitioning of the macroscopic model. Mesoscopic physical mechanism modeling: The unique properties of iron-based amorphous alloys originate from their composite structure of nanoscale grains and amorphous matrix; at the microscale, the distribution and motion of magnetic moments within the material are governed by multiple physical fields; based on the Landau-Lifshitz-Gilbert equations, a fundamental model describing the dynamic response of magnetic moments can be established: In the formula: / The overall value represents the rate of change of magnetization over time; M is the magnetization vector; γ is the gyromagnetic ratio; H eff For effective fields; M is the damping coefficient; s F is the saturation magnetization. th For thermal expansion term; Where H eff It is the superposition of multiple physical fields: In the formula: H is the magnetic field strength of the applied magnetic field; H d For demagnetizing the field; H a For magnetocrystalline anisotropic field; H ex For exchange; The LLG equations are solved using the fourth-order Runge-Kutta method, with a set time step based on the initial magnetization M0 and the effective field H. eff The magnetization intensity M(t) at different times is calculated iteratively. After obtaining the basic framework for the dynamic response of magnetic moments in iron-based amorphous alloys through mesoscopic physical mechanism modeling, the macroscopic hysteresis loop is divided into reversible magnetization, irreversible magnetization, and saturation regions based on the mesoscopic modeling results. Modeling is performed separately for each region, using magnetic field strength H as the dividing line: the boundary of the reversible magnetization region is set as |H| < 50 A / m, where domain wall motion is mainly reversible displacement; the range of the irreversible magnetization region is 50 A / m < |H| < 500 A / m, corresponding to irreversible jumps of domain walls overcoming pinning; the range of the saturation region is |H| ≥ 500 A / m, where magnetic moments are nearly perfectly oriented. Specifically, in the reversible magnetization region, based on the reversible small displacements of mesoscopic domain walls and the calculation results of anisotropic and exchange fields, a linear equation is used to describe the relationship between magnetization M and magnetic field strength H: In the formula: / The overall magnetic susceptibility is reversible, and µ0 is the free permeability; µ rev H is the inverse permeability; rev The reversible threshold; In the irreversible magnetization region, considering the irreversible jumping characteristic of mesoscopic domain walls overcoming the pinning barrier, the Jiles-Atherton model is introduced. Combining the damping coefficient and effective field to characterize the nonlinear magnetization behavior, parameters such as pinning strength k = 200 A / m and coupling coefficient α = 0.01 are first set. The nonhysteresis magnetization intensity is then calculated using formulas. : In the formula: α is the coupling coefficient; k is the pinning strength; Where M is the saturation magnetization, H is the magnetization, and H is the magnetic field strength. The fourth-order Runge-Kutta method is used to update the magnetic field strength H with a step size of 0.1 A / m, and then the obtained non-hysteresis magnetization is... Substitute into the following formula: In the formula: / The overall value represents the rate of change of magnetic flux density, δ is the direction factor, and k is the pinning strength. η is the dynamic loss coefficient; B It represents the magnetic flux density; In the saturation region, based on the characteristic that mesoscopic magnetic moments tend to be perfectly aligned, a paramagnetic correction term is added to improve the macroscopic model. Based on the analysis of the magnetic moment saturation characteristics of the mesoscopic model, the magnetic field strength H is calculated in increments of 1 A / m starting from 500 A / m, and the magnetization M is calculated using the following formula: In the formula: H sat It is a saturated magnetic field; It is the paramagnetic susceptibility; After the model of each region is built, check the difference of the calculated M value at the boundary of adjacent regions. If the difference exceeds the threshold, adjust the boundary point data by linear interpolation to ensure that the hysteresis loop is continuous and smooth. Finally, compare the hysteresis loop calculated by the segmented model with the experimental data to calculate the error of characteristic parameters such as loop area and coercivity. If the error exceeds 5%, return to adjust the model parameters of each interval until the accuracy requirements are met. Step S103, loss quantification and prediction calculation for multi-field coupling extension: After completing the segmented construction of the macroscopic model, multi-field coupling extension is performed to adapt to the performance prediction requirements of iron-based amorphous alloys under actual complex working conditions. The coupling between the temperature field and the magnetic field is mainly manifested in the thermal activation effect and the structural relaxation mechanism. Specifically, an increase in temperature will reduce the saturation magnetization Ms, and its calculation formula is as follows: In the formula: M is the saturation magnetization at temperature T. s (0) represents the room temperature saturation magnetization; T c T represents the Curie temperature; T represents the real-time temperature. The coupling between the frequency field and the magnetic field is achieved through the frequency sensitivity of the relaxation time and the eddy current skin effect; specifically, it is manifested as the magnetic moment relaxation time under high-frequency excitation. The shortening calculation formula is: In the formula: f is the external field frequency; The time is the magnetic moment relaxation time. Eddy current loss is affected by the skin effect, and the calculation formula is as follows: In the formula: Where d is the eddy current loss and d is the strip thickness; B is the resistivity; m The peak value of the magnetic flux density; µ r λ is the high-frequency relative permeability; λ is the skin depth; f is the external field frequency; First, real-time temperature T and external field frequency f data are acquired. Then, magnetic parameters Ms(T) and dynamic response parameters τd are updated using the above coupling formula. These are then substituted into the mesoscopic LLG equation and the macroscopic piecewise model to correct the effective field calculation and magnetization distribution. Finally, combined with multi-field experimental data, the coupling coefficient is optimized through algorithms to ensure that the prediction error of the model in multi-field environments is less than 5%. After completing the multi-field coupling extension, the effects of the temperature and frequency fields are quantified into the loss calculation. Temperature affects the shape of the hysteresis loop by changing the saturation magnetization Ms(T), while frequency is affected by the relaxation time. τ The skin depth λ alters the dynamic loss characteristics; Specifically, hysteresis loss P hyst The formula, calculated by integrating the area of ​​the hysteresis loop and combining it with the frequency, is as follows: In the formula, This is hysteresis loss; d is the magnetic field frequency; d is the strip thickness; M is the magnetization; eddy current loss is calculated based on the skin effect model, and the formula is: In the formula, The relative permeability is given by the formula for calculating excess loss, which originates from high-frequency domain wall resonance. In the formula: For excess loss, k e B is the excess loss factor. m f is the peak value of the magnetic flux density; f is the frequency of the external field. Total loss is obtained by superimposing hysteresis, eddy current, and excess loss, as shown in the formula: In the formula: Total loss, To account for eddy current losses, a temperature correction formula is introduced to predict losses at different temperatures: In the formula: This represents the predicted total loss at temperature T. This represents the total loss at time T0, where T0 is the reference temperature and T is the actual temperature. Temperature coefficient; Step S104, Hybrid Algorithm Parameter Optimization: In this step, the global optimization and high-precision calibration of the dynamic hysteresis model parameters are achieved through the collaborative optimization of genetic algorithm and particle swarm algorithm. Specifically, five core parameters to be optimized were extracted from the model established in S102, namely: reversible permeability u. rev Pinning strength k, JA model coupling coefficient α、 Excess loss coefficient ke, loss temperature coefficient Five parameters, with their value ranges set based on material physical properties: reversible permeability µ rev ∈[1000,10000]; pinning strength k∈[50,500]A / m; JA model coupling coefficient α∈[0.001,0.1]; excess loss coefficient k e ∈[10 -8 10 -4 Temperature coefficient αT∈[-5×10] -3 5×10 -1 ]K -1 ; A genetic algorithm is used for global search, generating 100 initial individuals using real-number encoding. Each individual is a parameter vector [µ]. rev [, k, α, ke, αT], where the parameter values ​​follow a log-normal distribution; then, a three-objective function is defined, incorporating hysteresis loop area error, loss prediction error, and model stability constraints: a three-objective fitness function that comprehensively considers model accuracy and physical consistency. In the formula: E represents the model fit. A E represents the area error of the hysteresis loop. P C represents the average error of losses under multiple operating conditions; C is a stability constraint term. Weighting coefficient w A , w B , w C The values ​​are 0.3, 0.6, and 0.1 respectively. Hysteresis loop area error E A The formula is: In the formula: A calc The area of ​​the hysteresis loop calculated by integrating the piecewise model for the reversible magnetization region, the irreversible magnetization region, and the saturation region; A exp The area of ​​the hysteresis loop as measured experimentally; The loop integral symbol is d; the strip thickness is d; and the magnetization is M. Average error E of multi-condition loss P The formula is: In the formula: This represents the total loss error; For each experimental condition, there are frequency / magnetic flux density sampling points; N represents the i-th loss value predicted by the model. case For different frequencies, temperatures, and excitation types; P calc,i Calculation of losses under different operating conditions; P exp,i The loss was measured experimentally. The formula for the stability constraint term C is: In the formula, α is the coupling coefficient; k is the pinning strength; The saturation magnetization is used as the basis for genetic operations. A selection operator is used, randomly selecting three individuals each time, retaining the one with the lowest fitness at a 40% elimination rate to ensure the population evolves towards the optimal solution. Arithmetic crossover is performed, where offspring parameters are generated by a linear combination of parent parameters, using the following formula: In the formula: X child λ is the parameter vector of the offspring individuals generated by the crossover operation; λ is a random number ∈ (0, 1). and These represent the parameter vectors of the two parent individuals participating in the crossover operation; For high-frequency sensitive parameters ke and α T Gaussian mutation is used, with the step size proportional to the current value: In the formula: k e ′、α T The values ​​are divided into the excess loss coefficient after variation and the temperature coefficient after variation. , The standard deviation of the variation; α is a normally distributed random number; ke is a high-frequency sensitive parameter (ke is 1.2 × 10−4); T For high-frequency sensitive parameters (α) T 0.0025∘C -1 ); For low-frequency parameters μ rev and k adopt uniform variation with a perturbation amplitude of ±10%, and the formula is as follows: In the formula: k is the low-frequency parameter, and k is 1500 A / m; , It is divided into the value of reversible permeability after variation and the value of pinning strength after variation; The amplitude of the disturbance; Particle swarm optimization (PSO) is used for local refinement. The top 20 best individuals after 50 generations of genetic algorithm iterations are extracted as the initial particles for PSO, with initial velocities pointing towards the global optimum of GA. In the formula: Let be the initial velocity vector of the i-th particle; The parameter vector of the optimal individual obtained during the global search phase of the genetic algorithm. Let be the initial position vector of the i-th particle; Inertia weight w The performance decreases linearly with the number of iterations, balancing global search and local exploitation capabilities. In the formula: The inertia weight is the weight at iteration number t. The total number of iterations set for the particle swarm optimization algorithm; t is the number of iterations; The particles are updated, and the velocity update formula is: In the formula: Let be the velocity of particle i in the (t+1)th iteration; Let c1 and c2 be the velocities of the i particles at the t-th iteration, respectively, and be learning factors, where c1 = 2.0 and c2 = 2.1; r1 and r2 are random numbers ∈ (0, 1); gbest represents the optimal positions experienced by the entire particle swarm; pbest represents the optimal positions experienced by the entire particle swarm. i Let i be the best historical position that each particle has experienced. Let be the position vector of the i particles at the t-th iteration; The inertia weights are used in the particle swarm optimization algorithm. After each update, the parameters need to be projected back to the initially set value range to avoid exceeding the limit. The position update formula is: In the formula: Let i be the position of the i particles in the (t+1)th iteration; Let be the velocity of particle i in the (t+1)th iteration; Let be the position vector of the i particles at the t-th iteration; For high-frequency loss coefficient k e Applying a frequency-dependent sinusoidal perturbation enhances the capture of resonance effects, thus adjusting the high-frequency loss coefficient. The calculation formula is: In the formula, f is the external field frequency; for the temperature coefficient α T The step size is adjusted based on the difference between the current temperature and the reference temperature, resulting in an adjusted temperature coefficient. The calculation formula is: Where T is the actual temperature; The temperature coefficient is used; the termination condition for particle swarm optimization is: the fitness change of the global optimum is less than 0.1% for 15 consecutive generations; under all test conditions, the loss error E P <5%, hysteresis loop area error E A <3%.

Images