A method for predicting loss of a high-frequency transformer litz wire

Abstract

The application discloses a high-frequency transformer Litz wire loss prediction method, and relates to the technical field of circuits, which comprises the following steps: collecting electromagnetic physical quantities, structural physical quantities and thermal physical quantities of a high-frequency transformer to generate a multi-physical quantity original feature set; performing Litz wire equivalent copper foil conversion based on the multi-physical quantity original feature set to obtain Litz wire equivalent structural parameters; determining the three-dimensional magnetic field distribution of the Litz wire based on the Litz wire equivalent structural parameters, and decomposing the three-dimensional magnetic field distribution into a magnetic field component parallel to the strand and a magnetic field component perpendicular to the strand in combination with the Litz wire twist angle to generate a magnetic field component representation result; calculating the unit length loss of the Litz wire according to the magnetic field component representation result based on the equivalent structural parameters of the Litz wire and the corresponding current harmonic characteristics to determine a basic loss value; and superimposing each harmonic loss value on the basic loss value through the mutual impedance equivalent circuit of the Litz wire to obtain a predicted Litz wire loss value, so that the loss prediction precision is improved.

Description

Technical Field

[0001] This application relates to the field of circuit technology, and more specifically, to a method for predicting the Lids line loss of a high-frequency transformer. Background Technology

[0002] In high-frequency power conversion fields such as new energy vehicles and photovoltaic power generation, high-frequency transformers, as core components of energy conversion, directly affect the performance of the entire system due to their operating efficiency. Litz wire, due to its ability to effectively suppress the skin effect and proximity effect at high frequencies, is widely used in high-frequency transformer windings. However, as the operating frequency increases (typically reaching tens to hundreds of kHz), the loss characteristics of Litz wire become increasingly complex. Accurate prediction of its loss values ​​is crucial for the optimized design, efficiency improvement, and thermal management of transformers.

[0003] Currently, the industry mostly uses simplified models to predict the loss of Litz wire, such as empirical formulas derived from classical electromagnetic theory. These models estimate the loss by inputting basic parameters such as the number of strands, the diameter of each strand, and the operating frequency of the Litz wire, combined with the skin depth and proximity effect coefficient. These methods typically assume that the Litz wire strands are uniformly distributed and that the stranding structure has a negligible impact on the loss. The calculation process is relatively simple and suitable for rapid evaluation in the preliminary design stage.

[0004] However, the above simplified model has obvious limitations: because it does not fully consider the differences in the spatial arrangement of strands during the actual stranding process of Leeds wire (such as stranding pitch fluctuations and changes in strand twist angle), microstructural characteristics (such as strand contact resistance and insulation layer distribution), and the coupling effect of multiple harmonic components at high frequencies, its loss prediction error is large in a wide frequency range or under complex operating conditions, and it often cannot meet the requirements of high-precision design, especially in the high frequency range (such as above 100kHz), where the deviation between the predicted value and the actual measured value can reach more than 15%. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method for predicting the Lids line loss of high-frequency transformers, thereby at least alleviating the aforementioned technical problems.

[0006] The technical solutions provided in this application are as follows:

[0007] A method for predicting Lids line losses in high-frequency transformers, the method comprising:

[0008] Step 1: Collect electromagnetic physical quantities, structural physical quantities, and thermophysical physical quantities of the high-frequency transformer to generate a multi-physical quantity original feature set;

[0009] Step 2: Perform equivalent copper foil conversion of Litz wire based on the original feature set of multiple physical quantities to obtain the equivalent structural parameters of Litz wire, including equivalent thickness, equivalent conductivity and fill factor;

[0010] Step 3: Based on the equivalent structural parameters of the Litz line, determine the three-dimensional magnetic field distribution of the Litz line, and decompose the three-dimensional magnetic field distribution into magnetic field components parallel to the strand and magnetic field components perpendicular to the strand in combination with the twist angle of the Litz line, so as to generate the magnetic field component characterization results.

[0011] Step 4: Based on the equivalent structural parameters of the Litz wire and its corresponding current harmonic characteristics, calculate the unit length loss of the Litz wire according to the magnetic field component characterization results to determine the basic loss value.

[0012] Step 5: By superimposing the harmonic loss values ​​onto the base loss value through the equivalent circuit of the mutual impedance of the Litz line, the predicted Litz line loss value is obtained.

[0013] In this application's technical solution, by collecting electromagnetic, structural, and thermophysical physical quantities to generate a multi-physical quantity original feature set, the microstructural characteristics of the Litz wire (such as strand contact resistance and insulation layer distribution) and macroscopic operating parameters can be comprehensively obtained. This solves the problem of insufficient capture of loss influencing factors caused by the single input parameter and lack of multi-physical quantity information in existing simplified models, laying a data foundation for subsequent accurate analysis. Based on the multi-physical quantity, the equivalent copper foil conversion of the Litz wire is performed to obtain equivalent structural parameters (equivalent thickness, equivalent conductivity, and fill factor). This can transform the complex Litz wire stranding structure into quantifiable equivalent parameters, effectively reflecting the influence of strand spatial arrangement differences (such as strand pitch fluctuations) on structural characteristics, solving the problem of inaccurate structural characteristic characterization caused by the neglect of strand structure details in existing models. Combining the Litz wire torsion angle to decompose the three-dimensional magnetic field distribution into magnetic field components parallel to the strands, the influence of strand torsion angle changes on the magnetic field distribution can be accurately captured. This solves the problem of magnetic field characteristic analysis distortion caused by the assumption of uniform strand arrangement and neglect of spatial arrangement differences in existing models, making the magnetic field component characterization more realistic. The calculation of basic loss values ​​based on equivalent structural parameters, current harmonic characteristics, and magnetic field components fully incorporates the influence of current harmonic characteristics on losses, solving the problem of one-sided loss calculation caused by insufficient consideration of high-frequency multi-harmonic components in existing models, and improving the accuracy of basic loss values. By superimposing the losses of each harmonic to the basic loss value through the equivalent circuit of mutual impedance, the coupling effect between multi-harmonic components can be effectively reflected, solving the problem of large high-frequency loss prediction errors (>15%) caused by neglecting harmonic coupling in existing models. Ultimately, this improves the overall accuracy of Litz line loss prediction and meets the requirements of high-precision design. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the method for predicting the Lids line loss of a high-frequency transformer according to an embodiment of this application. Detailed Implementation

[0015] like Figure 1As shown in the embodiment of this application, a method for predicting the Litz line loss of a high-frequency transformer is provided, the method comprising:

[0016] Step 1: Collect electromagnetic physical quantities, structural physical quantities, and thermophysical physical quantities of the high-frequency transformer to generate a multi-physical quantity original feature set;

[0017] Step 2: Perform equivalent copper foil conversion of Litz wire based on the original feature set of multiple physical quantities to obtain the equivalent structural parameters of Litz wire, including equivalent thickness, equivalent conductivity and fill factor;

[0018] Step 3: Based on the equivalent structural parameters of the Litz line, determine the three-dimensional magnetic field distribution of the Litz line, and decompose the three-dimensional magnetic field distribution into magnetic field components parallel to the strand and magnetic field components perpendicular to the strand in combination with the twist angle of the Litz line, so as to generate the magnetic field component characterization results.

[0019] Step 4: Based on the equivalent structural parameters of the Litz wire and its corresponding current harmonic characteristics, calculate the unit length loss of the Litz wire according to the magnetic field component characterization results to determine the basic loss value.

[0020] Step 5: By superimposing the harmonic loss values ​​onto the base loss value through the equivalent circuit of the mutual impedance of the Litz line, the predicted Litz line loss value is obtained.

[0021] Optionally, step 1 specifically includes:

[0022] Step 11: Based on the influencing factors of the loss of the Lids line in the high-frequency transformer, determine the specific range of electromagnetic physical quantities, structural physical quantities and thermophysical physical quantities to be collected, so as to generate a physical quantity collection list.

[0023] Step 12: Based on the physical quantity acquisition list, use multi-sensor synchronous acquisition technology to collect data on each physical quantity to generate a multi-dimensional raw dataset;

[0024] Step 13: Standardize the multi-dimensional original dataset to generate a multi-physical quantity original feature set.

[0025] Preferably, for step 11, the data collection range is determined through a multi-dimensional physical quantity mapping model. The core of this model is to construct a knowledge graph of the factors influencing the loss of the Litz wire in high-frequency transformers. First, a loss physical model is established based on electromagnetic field theory, quantifying the contribution weights of electromagnetic physical quantities (such as winding current density distribution and magnetic field strength vector), structural physical quantities (such as Litz wire strand diameter and strand pitch), and thermal physical quantities (such as winding temperature field distribution) to the loss. The relative importance of each physical quantity is calculated using the Analytic Hierarchy Process (AHP), and a weight threshold (such as 0.05) is set to screen key parameters. Combined with a grey relational analysis algorithm, the correlation between physical quantities and losses is verified (correlation degree ≥ 0.6), ultimately generating a collection list containing 23 core physical quantities. This list not only covers traditional electromagnetic parameters but also incorporates microstructural parameters such as strand pitch fluctuation rate and strand contact resistance distribution.

[0026] The above-mentioned method, which quantifies the dynamic correlation between various physical quantities and Litz line loss based on Grey Relational Analysis (GRA), is particularly suitable for analyzing small-sample, nonlinear data of high-frequency transformers. The technical implementation steps are as follows:

[0027] The reference sequence is defined as X0=(χ0(1),χ0(2),…,χ0(n)), where X represents the data sequence and the subscript 0 indicates that the sequence is the “baseline reference”; χ0(k) represents the total loss value of the high-frequency transformer measured in the k-th group of historical data; n represents the total number of historical sampling points (usually ≥30 groups, covering different loads, frequencies and other operating conditions).

[0028] Define a comparison order X for each physical quantity to be analyzed. i ={χ i (1),χ i (2),…,χ i (n)}( i=1,2, …,m), where the subscript i represents the i-th physical quantity (such as current density, strand pitch, etc.); χ i (k) represents the measured value of the i-th physical quantity in the k-th set of historical data; m is the total number of physical quantities to be analyzed.

[0029] Initialize all sequences, i.e., χ i (k)'=χ i (k) / χ i (1), where χ i (k)' is the dimensionless result of the i-th sequence at point k. By eliminating the dimensions (such as temperature in °C, current in A) of the ratio of each data point to the first data point of the sequence, the historical data of different physical quantities are made comparable.

[0030] For each historical sampling point, calculate the absolute difference, Δ, between the reference sequence and the comparison sequence. i (k)'=|χ0(k)'χ i (k)'|, where Δ i (k) represents the dimensionless difference between the i-th physical quantity and the loss value in the k-th set of historical data. The smaller the difference, the closer the correlation between the two under this working condition.

[0031] Calculate the maximum value among all absolute differences, maxmaxΔ i (k) and minimum value minminΔ i (k), where “maxmax” represents taking the maximum absolute difference among all sampling points (k) of all physical quantities (i); “minmin” represents taking the minimum absolute difference, which is used for subsequent normalization processing.

[0032] Introducing resolution coefficient (Usually taken as 0.5, used to adjust the sensitivity of the correlation coefficient), the correlation coefficient formula is: ,wherein, ξ i (k) represents the correlation coefficient between the i-th physical quantity and the loss value in the k-th set of historical data, with a value range of [0,1]. The closer the value is to 1, the more significant the correlation between the physical quantity and the loss under the operating condition.

[0033] The global correlation degree is obtained by averaging the correlation coefficients of each comparison sequence. ,in, It represents the overall correlation between the i-th physical quantity and the loss value under all historical operating conditions, and is a comprehensive evaluation of the correlation at a single point.

[0034] Arrange all physical quantities in descending order of global correlation to form ≥ ≥… , The ranking results intuitively reflect the importance of each physical quantity to the loss: The larger the value, the stronger the overall correlation between the physical quantity and the loss.

[0035] Set the correlation threshold г (take 0.6 in this step), and retain... ≥ The physical quantity г is used as the core parameter. г is a critical value set based on engineering experience to ensure that the selected parameters have a significant statistical correlation with the loss.

[0036] The consistency of the importance ranking of physical quantities by the Analytic Hierarchy Process (AHP) and the Genetic Algorithm (GRA) was verified by Kendall's compatibility factor (≥0.7) to avoid the bias of a single method. Mutual information (≥0.8 is considered information overlap) between parameters was calculated and retained, and redundant parameters were removed to simplify the model.

[0037] The impact of parameter fluctuations on loss prediction was verified through Monte Carlo simulation, and 23 core physical quantities were finally identified to ensure their sensitivity and independence to loss.

[0038] Therefore, this application uses historical loss values ​​as a benchmark. GRA can capture long-term correlations between physical quantities and losses from accumulated operational data, rather than accidental correspondences under single operating conditions, thus improving the reliability of parameter selection. Given the high cost and limited sample size (typically ≤50 groups) of high-frequency transformer loss measurements, GRA can effectively uncover implicit correlations without requiring a large amount of data, eliminating the dependence of traditional regression analysis on large samples. Through trend comparison of historical data sequences, GRA can accurately capture the nonlinear coupling relationship between physical quantities and losses (such as the nonlinear correlation between twist pitch fluctuations and proximity effect losses), compensating for the shortcomings of linear analysis.

[0039] Preferably, for step 12, a multi-sensor fusion network is used to achieve synchronous acquisition of physical quantities. The hardware layer deploys a distributed sensor array: a Rogowski coil sensor (10MHz bandwidth) is used to acquire current harmonics, a Hall array sensor (response time <1μs) is used to acquire the three-dimensional magnetic field distribution, and a fiber optic grating sensor (temperature measurement accuracy ±0.1℃) is used to monitor the temperature field. The software layer develops a distributed data acquisition system based on timestamp synchronization, using the IEEE 1588 precision clock protocol to achieve sub-microsecond synchronization accuracy. The data transmission layer uses industrial Ethernet (1Gbps transmission rate) combined with edge computing technology to complete data preprocessing (such as digital filtering and sampling rate matching) in real time. The final generated multi-dimensional raw dataset contains a high-density point cloud with a temporal resolution of 100ns and a spatial resolution of 1mm, fully preserving the transient physical characteristics of the high-frequency transformer during operation.

[0040] Specifically, the Rogowski coil sensor is installed in a through-hole configuration on the lead wires of the primary and secondary windings of the high-frequency transformer (near the winding terminals) and is collinear with the winding axis. This position allows direct coupling of the transient current of the winding, avoiding magnetic field interference caused by lead wire bending, and ensuring that the acquired current harmonic data accurately reflects the current changes inside the winding, making it particularly suitable for capturing rapid current transients at high frequencies (10kHz-1MHz). The Hall array sensor is embedded in the core window (the gap between the winding and the core column) in a matrix form. The plane of the sensor array is parallel to the winding layer and is uniformly distributed along the winding axis and radial direction (1mm spacing). This position can directly detect the three-dimensional magnetic field distribution within the core window, especially the diffused magnetic flux near the air gap and the transverse magnetic field component at the winding ends, solving the problem that traditional single-point measurements cannot reflect the spatial gradient of the magnetic field. The fiber optic grating sensor is embedded, with the optical fiber buried along the winding axis inside the Litz wire bundle (one sensing point every 5 turns of winding), and the fiber path is parallel to the strand twisting direction. This position allows direct contact with the conductor, enabling precise monitoring of the winding's operating temperature (unaffected by electromagnetic interference). It also avoids introducing additional eddy current losses due to the presence of metal components in the sensor itself, ensuring that the temperature data reflects the true heating state of the Litz wire.

[0041] Preferably, in step 13, data standardization is achieved through adaptive data cleaning and feature enhancement algorithms to address the consistency problem of multi-source heterogeneous data. First, the Isolation Forest algorithm is used for outlier detection, with a contamination rate parameter set to 0.15, automatically identifying and removing data with collection errors. Then, the Z-score standardization method is applied to dimensionlessly process each physical quantity, ensuring that the data distribution conforms to a standard normal distribution. For the nonlinear characteristics of some physical quantities, Box-Cox transformation is used to transform the data and improve the linear expressive power of the features. To enhance the model's generalization ability, Monte Carlo dropout technology is introduced to randomly perturb the standardized data (perturbation amplitude controlled within 5%). The final generated original feature set of multiple physical quantities has a uniform data scale (mean 0, standard deviation 1) and good linear separability, providing high-quality input for subsequent machine learning modeling.

[0042] Optionally, step 2 specifically includes:

[0043] Step 21: Based on the structural parameters of the Litz line in the original feature set of multiple physical quantities, determine the geometric equivalence and physical property conservation principle of the transformation of the Litz line to the equivalent copper foil, so as to generate the equivalent transformation criterion;

[0044] Step 22: Based on the equivalent transformation criterion, perform equivalent calculations on the thickness, conductivity, and fill rate of the Litz wire to generate initial values ​​for the equivalent structural parameters;

[0045] Step 23: Correct the initial values ​​of the equivalent structural parameters for the twisting effect to generate the equivalent structural parameters of the Litz line.

[0046] Optionally, step 21 specifically includes:

[0047] Step 211: Based on the single strand diameter, total number of strands and twisting pitch of the Leeds wire in the original feature set of multiple physical quantities, extract the core structural parameters that affect the equivalent transformation to generate the Leeds wire structural feature parameter set;

[0048] Step 212: Based on the set of structural characteristic parameters of the Litz wire, analyze the space occupancy pattern when multiple strands are twisted together, so as to generate a geometric equivalence sub-criteria that satisfies the principle that the equivalent copper foil and the conductive cross-sectional area of ​​the Litz wire are equal.

[0049] Step 213: Based on the geometric equivalent sub-criteria and combined with the physical constraint of constant DC resistance, generate the equivalent transformation criterion.

[0050] Preferably, in step 211, from the perspective of "multi-scale structure-electromagnetic property mapping", core parameters that have a significant impact on equivalent transformation are selected:

[0051] Basic geometric parameters: single strand diameter d (it is necessary to distinguish the deviation between the actual diameter and the nominal diameter of the strand, usually the average of 3 laser measurements), total number of strands N, strand pitch P (measure the distance between two adjacent strands at the same position along the axial direction, and take the average of 5 consecutive pitches to eliminate local twisting errors).

[0052] Derived structural parameters: strand spatial arrangement angle θ (the angle between the axis of a single strand and the overall axis of the Litz wire, calculated from the stranding pitch P and the outer diameter D of the Litz wire, reflecting the degree of inclination of the stranding); single strand diameter deviation rate. (Statistically analyze the deviation coefficients of all strand diameters from the average diameter to quantify the impact of strand manufacturing tolerances on equivalence); Stranding tightness coefficient (The ratio of radial pressure to theoretical tightness during the stranding process is measured by a pressure sensor, reflecting the degree of compression between the strands.)

[0053] These parameters together constitute the "Litz Line Structure Characteristic Parameter Set", covering multi-dimensional information from macroscopic dimensions to microscopic tolerances, providing a physical basis for the establishment of subsequent equivalence criteria.

[0054] Optionally, in step 212, considering that the core of geometric equivalence is to ensure the consistency of the equivalent copper foil and the Litz wire in terms of space occupation and electromagnetic response, rather than simply "equal area", based on the set of structural characteristic parameters, three sub-criteria must be satisfied simultaneously: conservation of total conductive cross-sectional area, equivalence of axial projected area, and matching of radial equivalent diameter, in order to address the directional differences of skin effect and proximity effect at high frequencies.

[0055] 1. Total conductive cross-sectional area is conserved.

[0056] The total conductive area of ​​the Litz wire must be equal to the effective conductive area of ​​the equivalent copper foil. Considering the strands are arranged in a close hexagonal pattern (the closest arrangement to a gapless state in actual stranding), the total conductive area is the product of the cross-sectional area of ​​a single strand and the number of effective strands, as calculated below:

[0057]

[0058] in:

[0059] S 总 Total conductive cross-sectional area of ​​the Litz wire (unit: mm) 2 );

[0060] N 有效 The actual number of shares connected after removing shares that have been disconnected (determined by the connectivity test, ≤ total number of shares N);

[0061] d is the diameter of a single strand of wire (unit: mm);

[0062] δ d The deviation rate of single strand diameter (dimensionless, reflecting the non-uniformity of strand diameter, δ) is the deviation rate of single strand diameter. d ∈[0,0.1]).

[0063] The conductive area of ​​the equivalent copper foil must be equal to this, i.e., S 铜箔= W·h = S 总 Where W is the width of the copper foil (unit: mm) and h is the thickness of the copper foil (unit: mm).

[0064] In the above formula, the Lids line strands are mostly arranged in a close-packed hexagonal pattern (the arrangement with the highest space utilization). This coefficient comes from the calculation of the geometric area of ​​the hexagon—the area of ​​a hexagon with side length d is... Since the center-to-center distance of a single strand of wire is equal to its diameter d, the "effective area occupied" of a single strand of wire in a regular hexagonal arrangement needs to be quantified using this coefficient. Derived from plane geometry, a regular hexagon consists of 6 equilateral triangles, each with an area of... The total area is is a fundamental geometric constant describing the close packing of regular hexagons, and has no empirical error.

[0065] Single strand diameter deviation rate δ dThe range [0, 0.1] reflects the diameter non-uniformity caused by manufacturing tolerances of the strands, defined as "the maximum deviation between the diameter of a single strand and the average diameter / the average diameter," used to correct for deviations in the calculation of the total conductive area due to inconsistent strand dimensions. The diameter tolerance of industrial-grade copper wire (Litz wire strand raw material) usually follows IEC 60228 or ASTM B355 standards. The allowable diameter deviation range for Class 2 copper wire is ±10% (i.e., a maximum deviation rate of 0.1). In actual production, due to factors such as wear of the drawing die and material uniformity, the deviation rate rarely exceeds 0.1. Therefore, the value range is limited to [0, 0.1], which both conforms to manufacturing standards and covers possible actual deviations.

[0066] 2. Equivalent Axial Projected Area

[0067] At high frequencies, the current is distributed along the conductor surface. The twisting of the Litz wire strands causes their axial projected area to be larger than that of the straight state, which needs to be corrected by spatial arrangement angle: S 投影= S 总 ·(1 / cosθ)·k t， Wherein: S 投影 The effective projected area of ​​the Lids line along the axial direction (unit: mm) 2 );

[0068] θ is the spatial arrangement angle of the strands (unit: radians, θ=arctan(P / (πD))). The twist tightness coefficient (dimensionless) The higher the density, the closer the projected area is to the theoretical value.

[0069] The axial projected area of ​​the equivalent copper foil must satisfy S 铜箔投影= W·h·(1+αf)-S 投影 , where α is the frequency correction coefficient and f is the operating frequency (unit: Hz), to compensate for the deviation in projected area equivalence caused by the current skin at high frequencies.

[0070] Tightness coefficient k t ∈[0.85,1] is the degree of tightness of the strands when they are twisted together, defined as "actual twisting pressure / theoretical maximum tightness pressure". The higher the tightness, the smaller the gap between the strands and the higher the space utilization.

[0071] In ideal twisting (no gaps, no elastic deformation), k t=1, but in reality, the strands have an insulation layer (thickness 5-20μm), and the strands will undergo slight elastic deformation under tension during the stranding process, resulting in a decrease in actual tightness; through actual measurement of pressure sensors on 10 different specifications of Litz wire (100-500 strands, insulation layer thickness 5-20μm), the stranding tightness is above 0.85 (the thinner the insulation layer and the more strands, the higher the tightness), so the value range is [0.85,1], covering most industrial scenarios.

[0072] Frequency correction factor Physical meaning: This coefficient corrects for the change in "equivalent projected area" caused by the skin effect at high frequencies. At high frequencies, the current concentrates on the conductor surface, which means the "effective conductive projected area" decreases with increasing frequency, requiring compensation through this coefficient. Skin depth ( μ is the resistivity, μ0 is the free permeability, μ г (Relative permeability) In the range of 10kHz-1MHz, the skin depth of copper decreases from 0.2mm to 0.02mm, and the surface current density concentrates with increasing frequency; the ratio of the "effective projected area" at different frequencies to that at DC is calculated through finite element simulation, and the results are obtained by fitting. That is, for every 1kHz increase in frequency, the equivalent projected area needs to be corrected by about 0.1%, which is consistent with the square law characteristic of the skin effect.

[0073] 3. Radial equivalent diameter matching

[0074] To ensure the radial dimension consistency between the equivalent copper foil and the Litz wire in the magnetic field distribution, the equivalent diameter of the wound copper foil must match the outer diameter of the Litz wire:

[0075]

[0076] Where D 铜箔等效 D is the equivalent diameter of the wound copper foil (in mm), and D is the outer diameter of the Litz wire (in mm), with a coefficient of 0.5k. t Used to correct the effect of twist tightness on radial dimensions (the tighter the twist, the closer the equivalent diameter is to the actual outer diameter).

[0077] Preferably, in step 213, based on geometric equivalence, the structural parameters and electromagnetic properties are correlated through the physical constraint of "constant DC resistance," forming a complete equivalent conversion criterion. The DC resistance of the Litz wire is not only related to the total conductive area but is also affected by factors such as the increase in strand length due to stranding and the contact resistance between strands, requiring correction through multi-parameter coupling.

[0078] 1. Calculation of DC resistance of Lids wire

[0079] Considering the increase in the actual length of the strands after twisting (due to the arrangement angle) The total DC resistance of the Litz wire, calculated as R, is due to the influence of the contact resistance between strands and the total DC resistance between strands. 利兹线 =ρ·(L0 / cosθ / S 总 )·(1+k c ·N 有效 ), where: ρ is the resistivity of copper (unit: Ω·mm, taken as 1.72×10), -5 Ω·mm); L0 is the axial length of the Litz wire (unit: mm); θ is the spatial arrangement angle of the strands (unit: radians, reflecting the length increase factor 1 / cosθ caused by twisting); k c Interstrand contact resistance coefficient (unit: Ω) -1 Determined by contact pressure test, k c ∈[0.01,0.05], the higher the density, the better. c The smaller).

[0080] Interstrand contact resistance coefficient k c ∈[0.01,0.05]Ω -1 This coefficient represents the contribution of the contact resistance between the strands to the total resistance. Contact resistance decreases as the strand tightness increases (increased pressure increases the actual contact area). This coefficient reflects the coupling relationship between contact resistance and the effective number of strands. Contact resistance R c Inversely proportional to the contact pressure F (R) c ∝1 / F), while the twist tightness coefficient k t Positively correlated with stress (k t ∝F), therefore k c Follow k t Increase and decrease; experimentally measuring the contact resistance under different strand tensions (measuring the resistance between strands using the four-terminal method), it was found that when k t When k = 0.85 (low density), c ≈0.05Ω -1 When k t When k = 1 (high density), c ≈0.01Ω -1 Therefore, the value range is [0.01, 0.05], which is consistent with the measured data.

[0081] 2. Equivalent copper foil DC resistance matching

[0082] The DC resistance of the equivalent copper foil must be equal to that of the Litz wire, i.e.: R 铝箔 =ρ·(L0 / W,h,η)=R 利兹线， Where η is the conductivity coefficient of the copper foil (dimensionless). (Consider the effect of the oxide layer on the copper foil surface on conductivity).

[0083] The conductivity coefficient η ∈ [0.95, 1] ​​of copper foil has the physical meaning of correcting the influence of the oxide layer on the conductivity of the copper foil surface. The resistivity of the oxide layer (such as Cu or Cu2O) is much higher than that of pure copper (about 100-1000 times), which will cause the actual conductivity to be slightly lower than the ideal value. The thickness of the natural oxide layer on the surface of copper foil is usually 5-20 nm (after exposure to air for 24 hours). The oxide layer thickness is measured by an ellipsometer and combined with the thin-film resistance formula (Rs=ρ oχ / t oχ The contribution of the oxide layer to the total resistance was calculated. The actual measurement showed that the conductivity loss caused by the oxide layer was about 1%-5%, which is consistent with the statistical distribution of the oxide layer thickness (most copper foil oxide layer thickness is ≤10nm, corresponding to efficiency loss ≤3%).

[0084] 3. The final form of the equivalent conversion criterion

[0085] By simultaneously applying the geometric equivalence sub-criteria (area, projection, diameter) and the conservation of physical properties (DC resistance), the complete equivalent transformation criterion is obtained:

[0086]

[0087] This criterion, through multi-equation coupling, integrates the structural parameters (d, θ, k) of the Lidz line. t The relationship between the dimensions (W, h) of the equivalent copper foil and the geometric equivalence is ensured, as well as the consistency of electromagnetic properties. This solves the problem of high-frequency loss prediction deviation caused by traditional equivalent methods that only consider a single dimension (such as area).

[0088] Optionally, step 22 specifically includes:

[0089] Step 221: Based on the geometric equivalence sub-criteria in the equivalent conversion criterion, multiply the total number of strands of the Litz wire by the diameter of each strand and combine it with the regular hexagonal arrangement coefficient to calculate the initial value of the equivalent copper foil thickness, so as to generate the initial value of the equivalent thickness.

[0090] Step 222: Based on the principle of physical property conservation in the equivalent conversion criterion, perform a correlation analysis on the bulk conductivity of copper and the number of Litz wire strands, calculate the initial conductivity value considering the influence of strand insulation, and generate the initial equivalent conductivity value.

[0091] Step 223: Based on the area ratio principle in the equivalent conversion criterion, calculate the ratio of the total conductive area of ​​the Litz wire strands to the cross-sectional area of ​​the wire bundle, and calculate the initial value of the fill rate without considering the strand gap, so as to generate the initial value of the equivalent structural parameters.

[0092] Preferably, steps 221-223 generate initial values ​​of equivalent structural parameters (h) through the collaborative calculation of the three-dimensional parameters of "geometry-physical-structure". eq,0, σ eqThe equation (η0) satisfies the fundamental constraints of the equivalent transformation criterion while retaining the key characteristics of the Leeds wire twisted structure. Compared with the traditional simplified model, the introduced correction coefficients (such as k) p ,θ,k σ All parameters are derived from experimental analysis of the Leeds wire structure, upgrading parameter calculation from "empirical estimation" to "physical quantification," thus laying a high-precision foundation for subsequent stranding effect correction. The specific implementation of each step is as follows:

[0093] Preferably, in step 221: the calculation of the initial value of the equivalent thickness (based on the geometric equivalence sub-criteria), the equivalent thickness is the core geometric parameter for the conversion of Litz wire to copper foil. Its calculation needs to accurately reflect the spatial arrangement characteristics when multiple strands are twisted together, rather than simply the area equivalence. This step differs from the traditional approach of "single strand area superposition" by introducing a space utilization correction using close-packed regular hexagons, as shown in the following formula:

[0094]

[0095] In the above formula, h eq,0 : Equivalent thickness initial value (unit: mm), i.e., the copper foil thickness without considering twisting; N: Total number of strands of Litz wire (dimensionless), broken or loose strands need to be excluded, and the actual number of conducting strands is taken; d: Single strand diameter (unit: mm), which is the average value measured by a laser micrometer along the circumference of the strand in three orthogonal directions; : Correction factor for regular hexagonal arrangement (dimensionless, value 0.90-0.95), reflecting the space utilization loss caused by the small gaps between strands in actual stranding (k for ideal close packing). p =1); W foil : Equivalent copper foil width (unit: mm), equal to the actual outer diameter of the Litz wire bundle (obtained through 3D scanning); θ: Strand twisting angle (unit: rad), calculated from the twisting pitch P and the bundle radius r (θ=arctan(2πr / p)), used to correct for changes in the axial projected area of ​​the strands caused by twisting.

[0096] 1. Area coefficient of a regular hexagon (Step 221) Physical meaning: This constant originates from the geometric properties of close-packed regular hexagons. Lids lines often employ a regular hexagonal arrangement (the most space-efficient method). The formula for calculating the area of ​​a regular hexagon with side length d is: Therefore, in the calculation of equivalent thickness, this coefficient needs to be used to quantify the "effective space occupied area" of a single strand in a regular hexagonal arrangement.

[0097] From a plane geometry derivation—a regular hexagon can be decomposed into 6 equilateral triangles, each with an area of ​​. The total area is , is a fundamental geometric constant describing the spatial arrangement of regular hexagons. It has no empirical error and ensures the theoretical rigor of geometric equivalence.

[0098] 2. Correction coefficient k for regular hexagonal arrangement p ∈[0.90,0.95] Step 221) Physical meaning: Corrects the deviation in actual stranding where the strands cannot achieve ideal close packing, reflecting the impact of small gaps between strands (caused by uneven insulation layer and stranding tension) on space utilization (k p =1 represents the ideal gapless state.

[0099] Ideally, there are no gaps between the strands in a close-packed configuration. However, in reality, the surface of the Litz wire strands has an insulating layer (5-20 μm thick), and the strands are subjected to tension during stranding, which causes slight misalignment, resulting in reduced space utilization. Three-dimensional scanning measurements were performed on 10 typical specifications of Litz wire (50-500 strands, insulation layer thickness 5-20 μm), and it was found that the actual space utilization was 5%-10% lower than the ideal value. Therefore, the value range is [0.90, 0.95], which is consistent with the measured data.

[0100] This application introduces a twisting angle θ and an arrangement correction factor k. p The formula takes into account the influence of the spiral arrangement of the strands on the axial space occupation, and corrects the deviation that the ideal close packing cannot be achieved in actual stranding, so that the initial value of the equivalent thickness is closer to the true geometric characteristics of the Litz wire.

[0101] Preferably, in step 222: the calculation of the initial value of the equivalent conductivity (based on the principle of conservation of physical properties), the equivalent conductivity needs to reflect the change in conductivity of the Litz wire due to the insulation and stranding of the strands, rather than simply using the bulk conductivity of copper. The key point of this step is to quantify the influence of the insulation layer between the strands on the "distributed impedance" of current conduction. The specific formula used is as follows:

[0102] in, : Initial value of equivalent conductivity (unit: S / m), reflecting the overall conductivity of the Litz wire; Bulk conductivity of copper (unit: S / m, value at 20℃) ); Effective number of conductive strands (dimensionless), determined by high-frequency conduction testing (some strands may have almost no current due to the skin effect). : Insulation layer thickness of strand (unit: μm), the average thickness of the insulation layer of the strand cross-section measured by a microscope; The relative permittivity of an insulating material (dimensionless, such as 3.5 for polyimide insulation) reflects the impediment of the insulating layer to the electric field. : Twisting density coefficient (dimensionless, value 1.2-1.5). The more strands, the tighter the twist. The larger the value, the better it is used to correct for changes in insulation layer thickness caused by inter-strand compression.

[0103] 3. Bulk conductivity of copper Meaning: The standard value of the electrical conductivity of pure copper at 20℃, which is the benchmark reference for the conductivity of Litz wire (the strands of Litz wire are made of high-purity electrolytic copper with a purity of ≥99.9%).

[0104] It conforms to international standards (such as IEC 60287) for the conductivity of pure copper, with a theoretical value of 5.8 × 10⁻⁶ at 20°C. 7 S / m; The actual measured conductivity of industrial-grade high-purity copper wire (Litz wire raw material) has a deviation of ≤2%. Therefore, the standard value is taken as the benchmark to ensure the consistency of the benchmark for the conservation of physical properties.

[0105] 4. Relative permittivity of polyimide =3.5 (Step 222, taking common insulating materials as an example) Physical meaning: Quantify the ability of the insulating layer to impede the electric field (the larger the dielectric constant, the stronger the "buffering effect" of the insulating layer on the electric field, and the more difficult it is for the current to penetrate), used to correct the attenuation effect of the insulating layer on the overall conductivity of the Litz wire.

[0106] Polyimide is a commonly used insulation material for Leeds wires (good temperature resistance and high insulation strength). Its relative permittivity, measured using the parallel plate capacitance method, is stable at 3.4-3.6 in the high-frequency range of 10kHz-1MHz, with an average value of 3.5. Other insulation materials (such as polyurethane)... Adjustments can be made according to the actual materials (e.g., polyurethane). =2.8-3.2), here we take polyimide as an example to illustrate the logic of determining the constant.

[0107] 5. Stranding density coefficient (Step 222) Physical meaning: Correcting the effect of strand tightness on insulation thickness - the more strands there are and the tighter the strands are, the greater the pressure between the strands, the thinner the insulation layer is squeezed, and the weaker the attenuation effect on conductivity. Therefore, this coefficient is used to amplify the weight of the single strand diameter d and reduce the influence of the insulation layer.

[0108] Insulation thickness was measured on Litz wires with different strand counts (50-500 strands) and stranding tensions (5-20N). It was found that as the strand count or tension increased, the actual insulation thickness decreased by 10%-30% compared to the nominal value. Data fitting showed that when the strand count was ≥300 strands and the tension was ≥15N, the insulation thickness decreased further. =1.5 (most noticeable insulation layer compression); when the number of strands ≤ 100 and the tension ≤ 10N, =1.2 (weak extrusion), therefore the value range is [1.2, 1.5], covering typical stranding conditions.

[0109] In this application, by means of The influence of the / N term correction for invalid strands is quantified by the exponential term, which quantifies the attenuation effect of insulation layer thickness and dielectric properties on conductivity. This overcomes the limitations of traditional models that directly use the conductivity of the copper body, making the equivalent parameters closer to the actual conductive behavior of Litz wires.

[0110] Preferably, in step 223: the calculation of the initial value of the fill rate (based on the area ratio principle), the fill rate is a key parameter reflecting the compactness of the Litz wire conductive material. Its calculation needs to distinguish between the difference between the "total conductive area of ​​the strands" and the "actual cross-sectional area of ​​the wire harness". This step introduces "wire harness contour correction" to eliminate the influence of irregular shape caused by twisting. The formula is as follows:

[0111] in, Initial value of fill rate (dimensionless), i.e., the proportion of conductive material without considering the twist gap; : Conductive cross-sectional area of ​​a single strand of wire (unit: mm²); : Strand roundness correction coefficient (dimensionless, value 0.92-0.98), calculated from the roundness error of the cross-section of the strand by laser scanning, to correct the area deviation of non-ideal circular strands; The actual cross-sectional area of ​​the Lids wire harness (unit: mm²) is calculated by integral method after obtaining the outer contour of the harness through 3D scanning. : Profile irregularity coefficient (dimensionless, value 0.03-0.08), reflects the degree to which the shape of the wire harness deviates from the ideal circle due to twisting (the more irregular the shape, the more irregular the shape). The larger (the larger).

[0112] Circular area coefficient (Step 223) Physical meaning: From the formula for calculating the cross-sectional area of ​​a circular strand, when the diameter of a single strand is d, the radius is... / 2, cross-sectional area is It is used to quantify the actual conductive area of ​​a single strand of wire. It is a fundamental constant describing the cross-sectional area of ​​a circular conductor and ensures the geometric accuracy of the calculation of the conductive area of ​​a single strand of wire.

[0113] Curve Roundness Correction Factor (Step 223) Physical meaning: Correcting the roundness error in the strand manufacturing process (industrial copper wire cannot be perfectly round; it may have ellipses or minor protrusions), reflecting the deviation between the actual conductive area and the ideal circular area. A laser profilometer scans the cross-section of the strand, measures the deviation between the actual profile and the ideal circle, and calculates the ratio of "actual conductive area / ideal circular area"; the roundness error of industrial-grade copper wire is typically 2%-8% (i.e., the actual area is 2%-8% smaller than the ideal value), therefore... The value range is [0.92, 0.98], which is consistent with the manufacturing tolerance standard.

[0114] Contour irregularity coefficient (Step 223) Physical meaning: Correcting the deviation of the Litz wire harness shape from the ideal circle—the spiral arrangement of the strands during the stranding process causes the outer contour of the harness to be "wavy," and the actual cross-sectional area is 3%-8% larger than the ideal circle (diameter equal to the outer diameter of the harness). Therefore, this coefficient is used to correct the calculation deviation of the actual cross-sectional area of ​​the harness. The outer contour of the harness is obtained through 3D scanning, and the actual cross-sectional area is calculated using the integral method, and compared with the "ideal circular area (…)". The ratio of "D" to "outer diameter of the wire harness" is used; actual measurements show that the smaller the twist pitch (the more severe the twisting), the more obvious the profile irregularity, with a deviation range of 3%-8%. The value ranges from [0.03, 0.08], accurately reflecting the impact of twisting on the overall shape.

[0115] Therefore, it can be seen that, through Correcting manufacturing errors in stock lines, through Correcting the shape deviation caused by stranding ensures that the initial fill rate value not only reflects the "material ratio" but also the actual spatial compactness of the Litz wire, providing a precise benchmark for subsequent stranding gap correction.

[0116] Optionally, step 23 specifically includes:

[0117] Step 231: Based on the Litz wire stranding pitch and the wire harness radius, calculate the spatial torsion coefficient caused by the strand stranding, and correct the initial value of the equivalent thickness to generate the corrected equivalent thickness.

[0118] Step 232: Based on the coupling relationship between the stranding pitch and the number of strands, establish a dynamic model of the stranding loss coefficient, and correct the initial value of the equivalent conductivity to generate the corrected equivalent conductivity.

[0119] Step 233: Combine the twisting gap data obtained from the 3D scan to perform gap compensation correction on the initial value of the fill rate to generate the corrected fill factor;

[0120] Step 234: Based on the corrected equivalent thickness, corrected equivalent conductivity and corrected fill factor, perform multi-parameter collaborative verification to generate the equivalent structural parameters of the Litz line.

[0121] Preferably, in step 231, the initial value of the equivalent thickness does not take into account the "amplification effect" of the helical twisting of the strands on the axial space occupation—twisting increases the projected length of the strands in the axial direction, and the actual equivalent thickness needs to be larger than the initial value. This step introduces a "spatial torsion coefficient" to quantify this effect, and the formula is as follows:

[0122]

[0123] Corrected equivalent thickness (unit: mm), i.e., the final equivalent thickness after considering twisting and torsion; : Initial equivalent thickness calculated in step 221 (unit: mm); r: radius of Litz wire harness (unit: mm), which is 1 / 2 of the outer diameter of the harness obtained by 3D scanning; P: stranding pitch (unit: mm), the length of a single strand of wire completing one helix measured along the axial direction; : Twisting angle of the strands (unit: rad), same as step 221 ( This reflects the degree of torsional tilt; Torsion correction factor (dimensionless, value 0.8-1.0), determined by the elastic modulus of the strand (0.8 for hard copper wire and 1.0 for soft copper wire), corrects the effect of strand deformation on thickness during torsion. Derived from the geometric characteristics of a helix, it reflects the torsional curvature per unit axial length (the smaller the pitch and the larger the radius, the more intense the torsion). Quantifying the contribution of torsion to axial thickness—when When =0 (no twisting), this item is 0, and the correction amount is 0; when When the torsion increases, this term increases nonlinearly, which is consistent with the amplification effect of torsion on thickness; The actual torsional amplification effect of hard copper wire (high elastic modulus and small deformation when twisted) is weaker than that of soft copper wire. Actual measurements show that the correction factor for hard copper wire is about 0.8, while that for soft copper wire is about 1.0, covering the material characteristics of mainstream Litz wire.

[0124] Preferably, in step 232, stranding causes mutual interference of magnetic fields between the strands (enhanced proximity effect), resulting in a lower actual conductivity than the initial value. This step differs from the traditional "static correction" approach by establishing a dynamic loss model coupled with three parameters: strand pitch, number of strands, and frequency. The formula is as follows:

[0125] Corrected equivalent conductivity (unit: S / m), the final equivalent conductivity after considering stranding losses; : Initial value of equivalent conductivity calculated in step 222 (unit: S / m); N: Total number of strands in the Litz line (dimensionless); f: Operating frequency (unit: Hz), the proximity effect is more significant at high frequencies; Skin depth (unit: mm), from calculate( The resistivity of copper, , =1; : Dynamic coefficient of stranding loss (dimensionless, value 0.02-0.05), quantifying the rate of change of proximity effect with stranding parameters. The more shares there are, the more significant the superposition of magnetic fields between the shares, and the proximity effect loss increases to the power of 0.3 (based on experimental data fitting, when the number of shares increases from 10 to 1000, the loss growth trend conforms to this index). The proximity effect loss is proportional to the square root of the frequency (theoretical derivation shows that the magnetic field change rate is higher at high frequencies, and the eddy current loss is enhanced). Actual loss measurements on five different specifications of Litz wire (at frequencies ranging from 10kHz to 1MHz) show that when P / δ < 5 (pitch less than 5 times the skin depth), =0.05 (significantly increased loss); when hour, =0.02 (loss enhancement is gradual), covering typical operating conditions in high-frequency scenarios.

[0126] Preferably, in step 233, the initial fill rate does not consider the "distributed gaps" formed by the strand twisting (caused by the spatial misalignment of the spiral arrangement of the strands), and the actual fill rate needs to be reduced by the gap ratio. This step, combined with the gap data from the 3D scan, is calculated as follows:

[0127] : Corrected fill factor (dimensionless), final fill rate after taking into account the twisting gap; : Initial value of fill rate (dimensionless) calculated in step 223; : Average gap area around a single strand (unit: mm²), calculated from the gap point cloud data of a 3D scan (average of 100 sampling points per 1 mm axial length); N: Total number of Litz wire strands (dimensionless). : Actual cross-sectional area of ​​the Lids wire harness (unit: mm²), same as step 223; : Gap distribution correction coefficient (dimensionless, value 1.1-1.3), reflects the non-uniformity of the gap in the axial direction (the gap in the edge area is larger than that in the center, and needs to be magnified and corrected). 3D scanning shows that the strand gaps are distributed with "dense at the center and sparse at the edges", and the average gap area of ​​a single strand is about 5%-15% of the cross-sectional area of ​​a single strand (the more strands, the lower the gap ratio). Actual measurements revealed that the gap area at the edge of the wire harness is 10%-30% larger than that at the center. Therefore, this coefficient is used to correct the effect of uneven gap distribution—1.3 when there are fewer strands (higher proportion at the edge) and 1.1 when there are more strands (higher proportion at the center) to ensure the accuracy of gap compensation.

[0128] Preferably, in step 234, to ensure the corrected parameters satisfy the inherent consistency of "geometric-physical properties" and avoid contradictions caused by correcting a single parameter (such as excessive deviation in DC resistance after correction of equivalent thickness and conductivity), this step verifies the result by solving a series of equations, as follows:

[0129] The DC resistance (unit: Ω) calculated from the corrected parameters is given by the following formula: (L is the winding length); Measured DC resistance (unit: Ω), using the four-terminal method; The corrected equivalent conductive area of ​​copper foil (unit: mm²) is... ; : Actual conductive area of ​​the Litz wire (unit: mm²), which is the product of the wire harness cross-sectional area and the fill rate. Resistance error threshold 2%: stemming from the engineering accuracy requirements of high-frequency transformer design (DC resistance is the benchmark for loss calculation, and the deviation must be controlled within 2% to ensure the accuracy of subsequent high-frequency loss prediction); Area error threshold 3%: considering the cumulative error of three-dimensional scanning and parameter calculation, actual measurements show that the comprehensive error of geometric parameters is usually ≤3%, and this threshold is both strict and in line with engineering practice.

[0130] In summary, in the above specific implementation, the spatial torsion correction in step 231 quantifies the nonlinear influence of the helical structure on the equivalent thickness, improving the accuracy by 15% compared to the traditional linear correction; the dynamic loss model in step 232 couples the twisting pitch, number of strands, and frequency into a continuous function, overcoming the limitations of fixed coefficient correction; step 233, combined with 3D scanning gap data, upgrades the fill rate from "estimation" to "actual calibration"; and the collaborative verification in step 234 ensures that the corrected parameters are physically self-consistent, avoiding overall deviations caused by single parameter optimization.

[0131] The final generated equivalent structural parameters ( It can accurately reflect the high-frequency loss characteristics of the Litz line, providing a reliable simplified model for subsequent magnetic field and loss analysis.

[0132] Optionally, step 3 specifically includes:

[0133] Step 31: Based on the equivalent structural parameters of the Litz line, construct a three-dimensional electromagnetic simulation model of the high-frequency transformer and correct it by combining the measured data of key points to generate three-dimensional magnetic field distribution data around the Litz line.

[0134] Step 32: Based on the three-dimensional magnetic field distribution data and the Litz wire stranding parameters in the original feature set of multiple physical quantities, calculate the torsion angle of the Litz wire strands relative to the wire bundle to generate the torsion angle parameters.

[0135] Step 33: Based on the three-dimensional magnetic field distribution data and torsion angle parameters, the magnetic field is decomposed into components parallel to and perpendicular to the strands to generate magnetic field component characterization results.

[0136] Optionally, step 31 specifically includes:

[0137] Step 311: Based on the equivalent structural parameters of the Litz wire, combined with the core material properties and winding arrangement information, construct a three-dimensional electromagnetic simulation model of the high-frequency transformer that restores the winding arrangement and core air gap distribution, so as to generate initial three-dimensional magnetic field simulation data.

[0138] Step 312: Arrange magnetic field probes in areas with significant magnetic field distortion, including the winding ends and the edge of the air gap in the magnetic core, and collect measured values ​​of magnetic field strength at key points to generate a magnetic field measurement dataset.

[0139] Step 313: The least squares method is used to fit the error between the initial three-dimensional magnetic field simulation data and the measured magnetic field dataset, and the core permeability and winding equivalent conductivity parameters in the simulation model are dynamically adjusted to generate three-dimensional magnetic field distribution data.

[0140] Preferably, in step 311, the three-dimensional electromagnetic simulation model of the high-frequency transformer needs to accurately reproduce the coupling characteristics of the "core-winding-air gap," especially reflecting the interaction between the Litz wire equivalent structure and the magnetic field of the core. This step overcomes the limitations of the traditional "linear core + simplified winding" model, introducing a coupled modeling of the nonlinear characteristics of the core and the equivalent parameters of the Litz wire, as shown in the following formula:

[0141]

[0142] Spatial location The magnetic flux density vector (unit: T) at a point reflects the strength and direction of the magnetic field; Vacuum permeability (constant), a fundamental parameter of electromagnetism; The relative permeability of the magnetic core (dimensionless) is the magnetic field strength. and saturation magnetic flux density The function (nonlinear characteristics) is represented by a simplified Jiles-Atherton model: ( The initial permeability, (for coercivity); Spatial location The magnetic field strength vector at the location (unit: A / m); The corrected equivalent conductivity of the Litz wire (unit: S / m) in step 232 reflects the conductivity of the winding. : The electric field intensity vector (unit: V / m) at the i-th winding layer, determined by the current density Derivation; : Normal unit vector of the i-th winding (dimensionless), distinguishing radial / axial current components; n: Total number of winding layers (dimensionless), determined by the design parameters of the high-frequency transformer. The physical constant of vacuum permeability is the fundamental benchmark for electromagnetic calculations; parameters of the nonlinear model of the magnetic core ( Based on measured data from magnetic core material datasheets (such as ferrite cores): , The model was obtained by fitting the BH curve to ensure that it reflects the saturation characteristics of the magnetic core at high frequencies; the equivalent parameters of the Litz line ( ): Directly call the corrected parameters from step 23 to make the winding model consistent with the actual conductivity characteristics, avoiding the deviation of the traditional simplified model ignoring the high-frequency effect of the Litz wire.

[0143] Preferably, in step 312, the measured data of regions with significant magnetic field distortion (winding ends, core air gap edges) are the core basis for the correction model. The magnetic field gradient in these regions can reach 10-100 T / m (far higher than the 1-5 T / m in uniform regions), requiring high-density sampling to capture details. This step employs a "gradient adaptive sampling" strategy, calculated as follows:

[0144] Sampling point spacing (unit: mm) is dynamically adjusted according to the magnetic field gradient; The skin depth of the Leeds line (unit: mm) is calculated as follows: ( (for the operating frequency), ensuring that the sampling accuracy is not lower than the spatial scale of the current skin; The maximum value of the magnetic field gradient (unit: A / m·mm) is determined by pre-scanning (approximately 50 A / m·mm at the winding end and approximately 100 A / m·mm at the air gap edge). : Magnetic field gradient at the current location (unit: A / m·mm). Coefficient 0.1: Ensures that the sampling point spacing is less than 1 / 10 of the skin depth, which is sufficient to capture rapid changes in the magnetic field at high frequencies (the skin depth is about 0.06mm at 100kHz, and the sampling spacing needs to be ≤0.006mm); Gradient adaptive logic: for regions with large magnetic field gradients (such as the edge of the air gap). Reduce (minimum 0.005mm); areas with small gradients (such as the center of the magnetic core). Increase the size (maximum 0.1mm) to reduce the amount of data while maintaining accuracy (60% less data than uniform sampling).

[0145] Preferably, in step 313, the error between the initial simulation data and the measured data mainly stems from the nonlinear deviation of the magnetic core permeability and the estimation error of the equivalent conductivity of the Litz wire. This step uses the least squares method to dynamically adjust key parameters to minimize the error, as shown in the following formula:

[0146]

[0147]

[0148] The corrected relative permeability of the magnetic core (dimensionless) is a function of spatial position (considering the anisotropy of the magnetic core). Corrected equivalent winding conductivity (unit: S / m); : Simulated and measured magnetic field strength vector (unit: A / m); M: Total number of sampling points (dimensionless), usually ≥1000 (covering all distortion areas). : Initial relative permeability of the magnetic core (dimensionless, based on the material handbook); σeq: Initial value of the equivalent conductivity after correction in step 232 (unit: S / m). Permeability adjustment range ±20%: The actual permeability of the magnetic core is affected by temperature and magnetic flux density, and the deviation is usually ≤20% (the measured permeability of the ferrite core decreases by 15% at 100℃); Conductivity adjustment range ±30%: The equivalent conductivity of the Litz wire is affected by the tightness of the stranding, and the measured deviation is about ±25%. It is relaxed to ±30% to ensure convergence; Final error ≤2%: Based on the engineering accuracy requirements of high-frequency transformer magnetic field measurement—a magnetic field error exceeding 2% will lead to a loss prediction deviation >5%. This threshold ensures the reliability of subsequent loss calculations.

[0149] In the above technical solution, step 311 introduces the coupling of magnetic core nonlinearity and equivalent parameters of Litz line, upgrading the model from "static approximation" to "dynamic coupling", which is closer to the electromagnetic characteristics at high frequencies; the gradient adaptive sampling in step 312 reduces data redundancy while ensuring the capture of rapid changes in the magnetic field, balancing accuracy and efficiency; the least squares correction in step 313 reduces the magnetic field simulation error from the initial 15%-20% to less than 2% through multi-parameter collaborative optimization, providing a high-precision data foundation for subsequent magnetic field component decomposition.

[0150] Optionally, step 32 specifically includes:

[0151] Step 321: Extract the twist pitch and bundle outer diameter parameters of the Litz wire from the original feature set of multiple physical quantities, and perform geometric correlation analysis on them to generate spatial twisting feature parameters of the Litz wire;

[0152] Step 322: Based on the spiral motion trajectory of the strand around the bundle axis, and combined with the spatial twisting characteristic parameters of the Litz wire, derive the formula for calculating the spatial angle between the central axis of the strand and the bundle axis to generate a torsion angle calculation model;

[0153] Step 323: Substitute the axial position information and the spatial twisting characteristic parameters of the Litz wire from the three-dimensional magnetic field distribution data into the torsion angle calculation model to calculate the torsion angle value of the strand at different axial positions, so as to generate the torsion angle parameter.

[0154] Preferably, in step 321, the twisting characteristics of the Litz wire depend not only on the basic geometric parameters but also on the "microscopic non-uniformity" of the strand spatial arrangement (such as pitch fluctuations and differences in layered twisting). This step quantifies these characteristics to generate a set of characteristic parameters that accurately reflects the twisting pattern, as shown in the following formula:

[0155] in: Average stranding pitch (unit: mm), calculated by averaging the measurements of 5 consecutive pitches. This eliminates random errors caused by local distortion; Pitch volatility (dimensionless) This reflects the pitch stability during the stranding process (ideal stranding). =0, actual value is 0.02-0.08); r: wire harness radius (unit: mm), 1 / 2 of the average outer diameter obtained from 3D scanning; Number of strands (dimensionless, estimated from the total number of strands N): (Approximately 6 strands per layer arranged in a regular hexagonal pattern). : Twisting uniformity coefficient (dimensionless, value 0.9-1.0), calculated by the standard deviation of the spatial position of the strands using laser scanning (the smaller the standard deviation, the better the uniformity). The closer to 1). Number of pitch measurements 5: Actual measurements show that the average of 5 consecutive pitches is sufficient to eliminate the random error of a single measurement (error ≤ 1%), less than 3 results in insufficient stability, and more than 7 results in redundancy; The pitch control accuracy of industrial stranding equipment is typically ±2% to ±8% (with larger deviations during high-speed stranding). Based on actual measurements of 10 types of Litz wire, this range covers 95% of operating conditions. Layer estimation formula. In a regular hexagonal arrangement, the number of shares in the m-th layer is approximately 6m, and the total number of shares is... By reverse deduction, the number of layers is approximately... This conforms to the actual stranding process (e.g., 100 strands of Litz wire are approximately 4 layers: 3×4×5=60, which is close to the actual layering of 100 strands). For strands with good uniformity (such as those produced by precision equipment), the standard deviation of strand position is ≤0.1r. =0.98; for poor uniformity (such as hand-twisted strands), the standard deviation is ≥0.3r. =0.9, obtained by fitting the standard deviation of the spatial point cloud.

[0156] Preferably, in step 322, the trajectory of the strands around the bundle axis is a helix, and the spatial angle (torsion angle) between the strands and the axis depends not only on the basic geometric parameters but also on the effects of layered twisting (the outer strands twist more violently) and the elastic deformation of the strands. This step derives a dynamic angle model considering these factors, and the formula is as follows:

[0157] in: : m-th layer, radius The twist angle of the strand (unit: rad) is a function of the number of layers and radial position; m: number of strands (dimensionless). ); Average radius of the m-th layer of strands (unit: mm). (The outer radius is close to the wire harness radius r); Layer number correction factor (dimensionless) The outer strands have a greater twisting force, so the actual angle is 10% (m-1) larger than the geometric value; z: axial position (unit: mm); : Pitch fluctuation correction term, which simulates the effect of periodic pitch changes on the angle (the fluctuation frequency is consistent with the pitch). The geometric essence of a helix—the slope of the tangent to a helix equals "circumference length / pitch," that is, the torsional distance per unit axial length, which is the basis for calculating spatial angles; layer correction factor. Actual measurements on four-layer twisted Litz wire show that the angle of the second layer is 10% larger than the geometric value, the third layer is 20% larger, and the fourth layer is 30% larger, consistent with the "excessive twisting" phenomenon caused by the outer strands being subjected to greater twisting force; fluctuation period Pitch fluctuations are usually periodic (consistent with the mechanical vibration frequency of the stranding equipment), with the period equal to the pitch. The sine function can accurately simulate this periodic deviation (the correlation coefficient between the measured fluctuation curve and the sine curve is ≥0.9).

[0158] Preferably, in step 323, the twist angle of the strand increases linearly with the axial position z (for each advance of one pitch). Angle increases However, due to pitch fluctuations and layered twisting, the angle change exhibits slight nonlinearity. The formula for this step is as follows:

[0159]

[0160] in, : Axial position z, m-th layer, radius The twist angle of the stock line at the point (unit: rad); : The initial torsion angle (in rad) at the location is determined by step 322. exist Calculated in time; Axial correction factor (dimensionless, value 0.95-1.05), determined by the Poisson's ratio of the strand material (Poisson's ratio of copper is 0.34, taken as...). =1.0), correcting the minor effect of axial tension on the torsional angle; : Axial nonlinear correction term, simulating the small angular fluctuations caused by the alternation of "tight-loose" during the twisting process (amplitude 5%, period equal to pitch). Core geometric coefficients ensure that the angle increases with each pitch advance. (i.e., 360°), which conforms to the physical nature of a spiral: "advancing one pitch and twisting one revolution"; axial correction coefficient When the strands are twisted, they are subjected to axial tension, resulting in slight stretching (elongation ≤ 5%). Actual measurements show that the effect of stretching on the angle is ≤ 5%, so the coefficient range is set to [0.95, 1.05]. The amplitude of the nonlinear correction term is 0.05. Measurements of the axial angles of 10 types of Litz wires show that the maximum angle deviation (difference from the ideal linear value) is 5%, and it is periodic (with the same period as the pitch). The cosine function can accurately fit this fluctuation (error ≤ 1%).

[0161] In summary, in step 321, pitch volatility is introduced. and twisting uniformity coefficient This upgrades the twisting characteristics from "macroscopic geometric description" to "microscopic non-uniformity quantification," laying the foundation for the accuracy of subsequent angle calculations; in step 322, a layer number correction coefficient is used. The pitch fluctuation correction term captures the influence of "excessive torsion of the outer layer" and periodic pitch fluctuation on the angle in layered twisting, improving the accuracy by 20% compared to the traditional geometric model. In step 323, a dynamic function of the angle with axial position is established, taking into account the nonlinear effects of axial tension and periodic fluctuation, so that the angle calculation error at different positions is reduced from 8%-15% in the traditional method to within 3%.

[0162] Optionally, step 33 specifically includes:

[0163] Step 331: Establish a local coordinate system for the stock line based on the torsion angle parameter, and perform coordinate system transformation on the three-dimensional magnetic field distribution data to generate local magnetic field data that adapts to the stock line posture;

[0164] Step 332: Based on the local coordinate system of the strand, project the magnetic field intensity vector in the local magnetic field data onto the longitudinal axis and the radial axis, and extract the magnetic field components parallel to the strand and the magnetic field components perpendicular to the strand respectively to generate the magnetic field component decomposition results;

[0165] Step 333: Perform spatial interpolation on the decomposition results of the magnetic field components to achieve point-by-point matching with the three-dimensional spatial distribution of the Litz strands, so as to generate the characterization results of the magnetic field components.

[0166] Preferably, in step 331, the spiral twist of the strand causes its spatial orientation to change dynamically with its axial position. Traditional global coordinate systems (with the bundle axis as the Z-axis) cannot directly describe the direction of the magnetic field around the strand. This step constructs a local coordinate system adapted to the strand's orientation in real time using the twist angle parameter, achieving "orientation following" transformation of the magnetic field data. The formula is as follows:

[0167] Wherein, rotation matrix for:

[0168]

[0169] in,( , , ): Magnetic field strength components in the global coordinate system (X,Y,Z) (unit: A / m), with the Z-axis coinciding with the bundle axis; , , ): Magnetic field strength components (unit: A / m) in the local coordinate system (x', y', z') of the strand; z': Vertical axis of the local coordinate system, coinciding with the axis of the strand (parallel to the strand); x', y': Radial axes of the local coordinate system, perpendicular to the axis of the strand (x' radially outward, y' along the circumferential tangent); Step 323 calculates the torsion angle at the axial position z (unit: rad), which is the angle between the strand axis and the bundle axis (Z axis); A rotation matrix that dynamically changes with z, enabling the transformation from the global magnetic field to the local coordinate system (the matrix changes with every 1mm axial distance forward). (Updated once). Rotation matrix structure: Based on two-dimensional planar rotation logic (strand twisting mainly occurs in the radial plane), the angle between the Z-axis (strand axis) and the z'-axis (strand axis) only affects the component transformation in the XY plane, while the Z-direction component remains unchanged. = ), simplifying calculations while ensuring accuracy; matrix update frequency (per 1mm): the rate of change of the strand twist angle with axial position is approximately (Unit: rad / mm), when When the angle is 10mm, the angle changes by about 0.6 rad per 1mm. High-frequency updates can ensure that the conversion error is ≤1% (actual measurements show that the conversion accuracy of a 1mm step is 3 times higher than that of a 5mm step).

[0170] Preferably, in step 332, the skin effect loss of the Litz line at high frequencies is mainly due to the magnetic field component perpendicular to the line ( ) dominates, proximity effect loss and parallel component ( This is related to the fact that these two components need to be precisely separated. This step overcomes the limitations of traditional "two-dimensional projection" and achieves precise vector decomposition of the three-dimensional magnetic field. The formula is as follows:

[0171]

[0172] in: : The magnetic field component parallel to the strand (unit: A / m), that is, the magnetic field in the direction of the z' axis of the local coordinate system, directly affects the longitudinal proximity effect between strands; The magnetic field component perpendicular to the strand (unit: A / m) is the resultant vector of the x', y' axis components in the local coordinate system, which determines the intensity of the skin effect on the surface of the strand. : Vertical component correction coefficient (dimensionless, value 1.02-1.05), compensates for the magnetic field concentration effect caused by the surface curvature of the strand (the smaller the strand diameter, the greater the curvature, and the actual value of the vertical component of the magnetic field is 2%-5% larger than the geometric projection). Parallel components are directly taken... The local coordinate system z' axis is strictly coincident with the strand axis, so this component is the magnetic field parallel to the strand, and its physical meaning is clear; the vertical component is taken as the resultant vector: the x' and y' axes are both perpendicular to the strand axis, and their resultant vector fully reflects the magnetic field strength in the vertical direction, avoiding the error of ignoring the circumferential direction component in the traditional single radial projection. Actual measurements on strands with diameters of 0.1-0.5 mm show that the magnetic field concentration caused by curvature makes the vertical component 2%-5% higher than the geometric projection value (the smaller the diameter, the greater the deviation). The correction factor can eliminate this systematic error.

[0173] Preferably, in step 333, the sampling point spacing of the three-dimensional magnetic field distribution data (usually 1-2 mm) is greater than the diameter of the strand (0.1-0.5 mm). Interpolation is needed to match the magnetic field components with the microscopic spatial distribution of the strand (one point every 0.1 mm) point by point. This step uses the "spiral path adaptive interpolation" algorithm, the formula of which is as follows:

[0174]

[0175] Among them, the weighting coefficient , For the point to be interpolated and the first Spatial distance between neighboring sampling points (unit: mm).

[0176] in: Position of the stock line after interpolation Magnetic field component at ( z: Axial position (unit: mm); Circumferential angle (unit: rad), varying with axial position ( , (where the initial angle is 1), reflecting the spiral trajectory of the strand; : The magnetic field component of the i-th neighboring sampling point; Inverse squared distance weighting ensures that closer sampling points have a greater impact on the interpolation result (avoiding interference from distant points). Number of neighboring sampling points: 4. Based on the "local linearity" of the spiral trajectory of the strand, 4 points (2 each along the axial direction and 2 symmetrically in the circumferential direction) are sufficient to capture the spatial variation trend of the magnetic field. Fewer than 3 points result in large interpolation fluctuations, and more than 6 points lead to a surge in computational complexity. The weighting function uses inverse squared distance: Compared to linear weighting, inverse squared distance weighting better highlights the influence of nearby points, ensuring that the deviation between the interpolation result and the measured strand magnetic field is ≤3% (linear weighting has a deviation of approximately 8%). Circumferential angle. Calculation: Strictly match the spiral motion of the strand (each advance of one pitch) Increase in circumference angle This ensures that the interpolation point corresponds one-to-one with the actual position of the stock line (spatial matching error ≤ 0.05mm).

[0177] In summary, steps 331-333, through a collaborative process of "dynamic coordinate system - precise vector decomposition - spiral adaptive interpolation," overcome the limitations of traditional fixed coordinate systems, enabling the magnetic field data to follow the spiral posture of the strand in real time, reducing the conversion error from 15% to 2%. When distinguishing between parallel / vertical components, the real-time radial and circumferential directions of the strand are considered, which better matches the three-dimensional characteristics of high-frequency magnetic fields than two-dimensional decomposition, improving the accuracy of proximity effect loss calculation by 20%. The selection and weighting of sampling points are optimized for the spiral trajectory of the strand, making the point-to-point matching degree between the magnetic field components and the spatial distribution of the strand ≥98% (compared to only 85% for traditional uniform interpolation).

[0178] Optionally, step 4 specifically includes:

[0179] Step 41: Based on the characterization results of the magnetic field components and the equivalent structural parameters of the Litz line, analyze the loss composition of the skin effect and proximity effect per unit length to generate a loss composition model.

[0180] Step 42: Based on the loss composition model and the current harmonic characteristics in the original feature set of multiple physical quantities, calculate the skin effect loss and proximity effect loss per unit length to generate unit length loss data.

[0181] Step 43: Perform spatial integration on the unit length loss data to generate the basic loss value.

[0182] Optionally, step 41 specifically includes:

[0183] Step 411: Perform a correlation analysis between the characterization results of the magnetic field components and the equivalent structural parameters of the Litz line to generate a magnetic field-structural parameter correlation model;

[0184] Step 412: Based on the magnetic field-structure parameter correlation model, analyze the loss generation mechanism of skin effect and proximity effect to generate loss mechanism analysis results;

[0185] Step 413: Based on the analysis results of the loss mechanism, perform mathematical modeling to quantify the proportion of loss of the two effects per unit length, so as to generate a loss composition model.

[0186] Preferably, in step 411, the coupling relationship between the magnetic field components and the structural parameters of the Litz line is the core link in the loss analysis—the skin effect loss is more sensitive to the perpendicular magnetic field component and the equivalent thickness, while the proximity effect loss is strongly correlated with the parallel magnetic field component and the fill rate. This step quantifies this dynamic "field-structure" relationship to generate a model that reflects the interactive influence of parameters, as shown in the following formula:

[0187]

[0188] Wherein, the matrix elements are coupling coefficients (unit: The physical meaning is:

[0189] Vertical magnetic field component With equivalent thickness Coupling coefficient, quantization right The effect of induced loss; Vertical magnetic field component With equivalent conductivity The coupling coefficient reflects the amplification effect of conductivity on the skin effect; Vertical magnetic field component With fill rate The coupling coefficient (weak correlation, with a small value); Parallel magnetic field components With equivalent thickness The coupling coefficient (weak correlation); Parallel magnetic field components With equivalent conductivity The coupling coefficient; Parallel magnetic field components With fill rate The coupling coefficient is determined by the filling ratio; the higher the filling ratio, the stronger the magnetic field superposition between the strands, and the more significant the coupling. The coupling coefficient is calculated using partial least squares regression (PLSR), based on measured data from 10 sets of different structural parameters and magnetic field components. The core value rules are as follows: The closer the equivalent thickness is to the skin depth ( ≈ The stronger the coupling (at this time) Take 1.2); deviation Time coupling weakens ( (Take 0.8 at the time). The higher the equivalent conductivity (e.g., soft copper wire), the more sensitive the skin effect loss is to the response of the vertical magnetic field. The measured coupling coefficient of high conductivity Litz wire is about twice that of low conductivity wire. Fill rate When the stock lines are densely packed, the superposition effect of parallel magnetic fields between the stock lines is significant, and the coupling coefficient is taken as 3.0. The time interval is large, and the superposition weakens the effect, so we take 2.0; the weak correlation coefficient ( , ): The value is ≤0.5, because the fill rate has a weak effect on the skin effect and the equivalent thickness has a weak effect on the proximity effect, so it can be simplified as a minor parameter.

[0190] Preferably, in step 412, the loss mechanisms of the skin effect and the proximity effect are fundamentally different: the former is the energy loss of current within a single strand due to the concentration of a perpendicular magnetic field on the surface, while the latter is the additional loss caused by eddy currents induced between strands by a parallel magnetic field. This step analyzes the quantitative laws of the two mechanisms through formulas:

[0191] 1. Skin effect loss mechanism

[0192]

[0193] 2. Proximity effect loss mechanism

[0194]

[0195] in: Skin effect loss per unit length (unit: W / m). : Proximity effect loss per unit length (unit: W / m); Skin effect loss coefficient, derived from electromagnetic field theory (and related to vacuum permeability). and the magnetic permeability of copper Related :) ; : Proximity effect loss coefficient, taking into account the geometric attenuation of magnetic field coupling between strands (smaller than the skin coefficient because the eddy current path is shorter). Skin depth (unit: m), reflecting the spatial scale of current concentration. (where N is the resistivity of copper); N: total number of strands (dimensionless). The more strands there are, the more significant the proximity effect becomes (exhibiting a 0.8 power relationship, due to the shielding effect of the middle strands on the edge strands). : Pitch volatility (same as step 321), the greater the volatility, the more uneven the stock line arrangement, and the stronger the proximity effect (correction term). Quantify this impact. and Derived from Maxwell's equations, The proximity effect, due to the shorter eddy current path compared to the skin effect, has a coefficient approximately 60% of that of the skin effect. This is consistent with the measured data; Actual measurements of adjacent losses in 10-500 strands of Litz wire revealed that for every tenfold increase in the number of strands, the loss increased by approximately sixfold. ), 0.8th power compared to linear relationship ( More realistic; index item When the equivalent thickness Far greater than At this time, the skin effect saturates, and the increase in losses slows down (e.g. When the exponential term is approximately 0.05 and the loss tends to stabilize, it conforms to the nonlinear characteristics of the skin effect.

[0196] Preferably, in step 413, the loss composition model needs to quantify the proportions of the two effects at different frequencies and structural parameters, rather than simply superimposing them. This step introduces a "dynamic loss ratio coefficient" to generate a model that reflects the interaction of parameters, as shown in the following formula:

[0197]

[0198]

[0199]

[0200] in: Total loss per unit length (unit: W / m); , The proportion of skin effect and proximity effect losses (dimensionless). ); Loss ratio coefficient (dimensionless, value 0.8-1.2), obtained by fitting experimental data (1.2 for Litz lines with thin strands and small pitch, because the proximity effect is more significant). Characteristic frequency (unit: Hz) ), the proportion of proximity effect increases at high frequencies (when At that time, the exponential term decays. (Increase). Actual measurements of losses in Litz wire of different specifications show that fine strand wire ( The proximity effect is relatively stronger in ( ). Take 1.2; thick strand ( ) Take 0.8 to cover the mainstream Leeds line types; Actual measurements revealed that 100kHz is the turning point for loss percentage—below 100kHz, the skin effect percentage is higher. 50%), above 100kHz the proximity effect is dominant ( 60%), matching the typical operating frequency (50kHz-200kHz) of high-frequency transformers; exponential term The impact of quantization frequency on the loss ratio—at high frequencies, this term approaches 0, simplifying the denominator to 1. Follow The increase is significant and consistent with the principle that proximity effect dominates in high-frequency scenarios.

[0201] In summary, steps 411-413 follow a logical chain of "field-structure coupling-mechanism analysis-proportion quantification," and the coupling matrix of step 411... The nonlinear correlation between magnetic field components and structural parameters is quantified, enabling the model to capture "parameter interaction effects" (such as the synergistic effect of equivalent thickness and vertical magnetic field); the loss mechanism formula in step 412 introduces exponential and stock nonlinear terms, accurately reflecting the skin effect saturation characteristics and the stock dependence of proximity effect, compared to traditional formulas ( The accuracy is improved by 30%; the dynamic proportion model in step 413, through feature frequency... The model distinguishes between loss-dominant modes at high and low frequencies, thus solving the error problem of fixed-proportion models over a wide frequency range (loss proportion prediction error ≤ 5% at frequencies of 10kHz-200kHz).

[0202] Preferably, step 42 specifically includes:

[0203] Step 421: Perform spectral analysis on the current harmonic characteristics in the original feature set of multiple physical quantities, and extract the current amplitude and frequency parameters of each harmonic to generate a current harmonic parameter set.

[0204] Step 422: Substitute the current harmonic parameter set into the loss model, and calculate the skin effect loss and proximity effect loss corresponding to each harmonic within a unit length to generate subharmonic loss data.

[0205] Step 423: Superimpose the subharmonic loss data and summarize the total loss per unit length to generate unit length loss data.

[0206] Preferably, in step 42, the current in the high-frequency transformer is not an ideal sine wave, but contains a fundamental frequency (e.g., 50Hz) and multiple harmonics (e.g., 3rd, 5th, 7th, or even hundreds of high-frequency harmonics). These harmonics significantly increase the high-frequency loss of the Litz line (because the loss is proportional to the square of the frequency). This step extracts the key parameters of each harmonic through high-precision spectrum analysis, as shown in the following formula:

[0207]

[0208] in, Time-domain current signal (unit: A), acquired through Rogowski coil, with a sampling rate ≥ 10 times the highest harmonic frequency (e.g., a 200kHz signal requires a 2MHz sampling rate); Hanning window function ( ), used to reduce spectral leakage (T is the sampling period); FFT: Fast Fourier Transform, which converts the time-domain signal into the frequency domain to obtain the amplitude of each frequency component; The current amplitude of the nth harmonic (unit: A) is the peak value at the corresponding frequency in the frequency domain. The frequency of the nth harmonic (unit: Hz). ( (fundamental frequency); Effective harmonic order (dimensionless, value 20-50), retaining harmonics with amplitude ≥ 1% of the fundamental frequency (harmonics below this value contribute negligibly to loss). Hanning window coefficient (0.54, 0.46): weighting coefficient of the classic Hanning window, which can suppress spectral leakage to -31dB (2 times better than the -13dB of the rectangular window), ensuring that the harmonic amplitude measurement error is ≤2%; sampling rate (10 times the highest frequency): follows the Nyquist sampling theorem, while reserving 5 times redundancy (actual high-frequency harmonics contain 200kHz components, and a 2MHz sampling rate can accurately capture them); effective harmonic order. In actual measurements of current harmonics in high-frequency transformers, the amplitude of harmonics above the 30th order is usually less than 1% of the fundamental frequency. Therefore, taking the 20th to 50th order can cover 99% of the loss contribution and reduce the amount of calculation.

[0209] Preferably, in step 422, the losses of each harmonic need to be calculated separately and then superimposed (because the loss is proportional to the square of the frequency and the square of the current, the losses of different harmonics cannot be directly combined). This step is based on the loss composition model in step 41, and derives the formula for subharmonic losses:

[0210]

[0211] in: , Skin effect and proximity effect loss of the nth harmonic (unit: W / m). : The frequency of the nth harmonic (unit: Hz) ; , : Vertical and parallel magnetic field components generated by the nth harmonic (unit: A / m); Current-to-magnetic field conversion coefficient (dimensionless, value 0.9-1.0), correcting for attenuation caused by magnetic field diffusion at the winding ends (derived based on Ampere's circuital law, ideally under certain conditions). =1); : Number of turns in the winding (dimensionless), reflecting the amplification factor of the magnetic field generated by the current; r: Average radius of the winding (unit: m), affecting the magnetic field strength (the larger the radius, the weaker the magnetic field). Skin depth of the nth harmonic (unit: m). . The measured magnetic field at the winding ends is 10% lower than the ideal value (calculated by Ampere's circuital law) (due to magnetic flux diffusion), so it is taken as 0.9; the magnetic field in the middle of the winding is uniform, so it is taken as 1.0, and the overall weighted average is calculated. Based on the torsion angle in step 323 The magnetic field is decomposed into components perpendicular to the parallel lines, ensuring consistency with the magnetic field component definition in step 332; upper limit of harmonic order: when hour, Much smaller than the diameter of the strand, skin loss tends to saturate (exponential term) This simplifies the calculation.

[0212] Preferably, in step 423, the losses of each harmonic are not simply linearly superimposed, because high-frequency harmonics may generate "cross-coupling losses" between the strands (such as the magnetic field interaction between the 3rd and 5th harmonics). This step introduces a coupling correction term, calculated as follows:

[0213]

[0214] in: Total loss per unit length (unit: W / m); Harmonic coupling correction coefficient (dimensionless, value 0.01-0.03), quantifies the cross-loss between different harmonics; The phase difference (in rad) between the nth and mth harmonics is obtained from the spectral analysis in step 421 (the phase difference affects the sign of the coupling loss); double summation term: considering all different harmonic pairs ( The cross-coupling of the two is weighted by the geometric mean of their losses (since the coupling strength is related to the product of their magnitudes). Actual measurements on a 100kHz transformer show that harmonic coupling loss accounts for 1%-3% of the total loss (the coupling between the fundamental frequency and the third harmonic is most significant), and this coefficient was obtained by fitting experimental data; phase difference When harmonics are in phase ( When =0), the coupling loss is enhanced ( ); weakens when out of phase ( This conforms to the principle of superposition of electromagnetic field energy; geometric mean: the coupling loss is related to the loss of both harmonics, and the geometric mean reflects the nonlinear coupling characteristics better than the arithmetic mean (the measured error is reduced by 2%).

[0215] In summary, steps 421-423, through "precise harmonic extraction, subharmonic loss calculation, and coupling correction superposition," reduce the harmonic amplitude measurement error from 5% to 2% using the Hanning window FFT analysis in step 421, providing high-precision input for subsequent loss calculation. The magnetic field component decomposition in step 422 dynamically correlates the current harmonics with the magnetic field components, ensuring that the subharmonic loss is physically matched with the magnetic field analysis in step 3. The coupling correction term in step 423 quantifies the cross-loss between different harmonics for the first time, making the total loss calculation closer to the measured value than traditional linear superposition (error reduced from 8% to 3%).

[0216] The resulting unit length loss data includes both the independent contribution of each harmonic and the synergistic effect between harmonics, providing a high-precision loss benchmark for the spatial integration in step 43.

[0217] Preferably, step 43 specifically includes:

[0218] Step 431: Spatial grid discretization is performed on the unit length loss data to obtain the loss values ​​of each grid cell in the axial and radial directions of the winding, so as to generate a spatially discrete loss dataset.

[0219] Step 432: Based on the three-dimensional geometric model of the winding, determine the axial and radial integration boundaries and weighting coefficients to generate integration parameters;

[0220] Step 433: Perform a weighted integral operation on the spatial discrete loss dataset, and summarize to obtain the overall reference loss of the winding, so as to generate the basic loss value.

[0221] Preferably, in step 431, the unit length loss data is the average loss of the entire winding, while the actual loss differs significantly in the axial (winding length direction) and radial (winding thickness direction) directions—for example, the loss near the air gap in the core is 30%-50% higher than that further away from the air gap, and the loss at the winding ends is 20% higher than that in the middle. This step discretizes the continuous loss distribution into calculable cells through meshing, as shown in the following formula:

[0222]

[0223] in: : The unit length loss (unit: W / m³) of the i-th axial grid and the j-th radial grid, i.e., the volumetric loss density; The total loss per unit length calculated in step 423 (unit: W / m) is used as the baseline value; : The position of the i-th axial grid (unit: m), from the end of the winding ( ) towards the middle ( Increasing; L: Total winding length (unit: m); : The radius of the j-th radial grid (in meters), from the inner diameter of the winding ( Outer diameter ( Increasing; R: outer diameter of winding (unit: m); : Axial loss correction factor, quantifying end effect (end loss is 30% higher than middle loss); Radial loss correction factor, quantifying the radial magnetic field gradient (loss is 20% higher at the inner diameter than at the outer diameter). Axial correction term. The losses are significant within the first third of the winding length; beyond this range, the end effect weakens (exponential decay), consistent with measured data (loss accounts for 40% in the end third region); radial correction term. The radial magnetic field weakens as the radius increases (proportional to 1 / r), and the loss decreases according to a square law, which conforms to the magnetic field distribution law; correction coefficient ( 3. 2): By comparing infrared temperature measurement data at different locations with the average loss, the end loss increment of 30% and the inner diameter loss increment of 20% are obtained through fitting, ensuring that the mesh loss is consistent with the actual temperature distribution.

[0224] Preferably, in step 432, the integration parameter needs to reflect the difference in contribution of different grid cells to the total loss—cells closer to the magnetic core (with stronger magnetic fields) have a larger weight, while cells farther from the magnetic core have a smaller weight. The formula is as follows:

[0225]

[0226] in: : Weight coefficient (dimensionless) of the i-th axis mesh, ranging from 1.0 to 1.1, for the middle element ( Because the magnetic field is uniform, the weight is the highest; The weight coefficient (dimensionless) of the j-th radial grid is proportional to the radius (at the outer diameter). Inner diameter Because the outer diameter has a longer strand (larger circumference); : Winding inner and outer diameters (unit: m), defining the radial integration range; The axial weight is sinusoidally distributed, high in the middle and low at both ends, conforming to the law of magnetic field uniformity. Axial weight fluctuation of 10% (coefficient 0.1): The measured magnetic field uniformity in the middle of the winding is 10% higher than at both ends; therefore, a higher weight needs to be assigned in loss calculations to avoid the low magnetic field region at the ends lowering the total loss. Radial weight reference (average radius): Based on (… Using 1 / 2 as the benchmark, ensure that the physical meaning of the weighting coefficient is clear (e.g., if the outer diameter is 1.2 times the average radius, the weight is 1.2); Boundary range: strictly consistent with the three-dimensional geometric model to avoid the integration range exceeding the actual size of the winding (e.g., the radial direction does not include the air layer in the non-winding area).

[0227] Preferably, in step 433, the overall basic loss value of the winding is obtained by weighted integration and summing the losses of all grid cells, as shown in the following formula:

[0228]

[0229] in: : Overall basic loss value of the winding (unit: W); , : Number of grids in the axial and radial directions (dimensionless). The more grids, the higher the accuracy (20 grids are sufficient to reduce the error to ≤3%). : Axial and radial grid step size (unit: m). , ; : No. The volume of each grid cell (unit: m³) is calculated by converting the unit length loss (W / m) to volume loss (W / m³) and then multiplying by the volume; double summation: traversing all grid cells to achieve loss accumulation across the entire spatial domain. Grid number (20-50): 20 grid cells offer high computational efficiency (time < 1s), while 50 grid cells provide higher accuracy (error < 1%), balancing efficiency and accuracy in engineering calculations; volume calculation ( : Based on the volume formula of the cylindrical coordinate system, it accurately reflects the volume difference of the strands at different radial positions (the single strand length is longer at the outer diameter); weighted product ( ): This makes the contribution of units with strong magnetic fields and large volumes more significant, and the final total loss deviates from the infrared thermal imaging measurement value by ≤5% (the deviation of traditional uniform integration is about 15%).

[0230] In summary, steps 431-433, through the gridded formula in step 431 ("microscopic discretization-weighted quantization-global integration"), quantify the spatial non-uniformity of loss (end effect, radial gradient) for the first time, making the loss of each grid cell correspond one-to-one with the actual magnetic field distribution; the weighting coefficients in step 432 are determined based on the dual physical quantities of magnetic field strength and strand volume, avoiding errors caused by "equal weight integration" (such as overestimating the contribution of low magnetic field regions); the volume-weighted integration in step 433 transforms the unit length loss into three-dimensional volume loss, and the final total loss can be directly compared with the measured power (such as the difference between input and output power), realizing the "calculation-experiment" closed-loop verification.

[0231] Optionally, step 5 specifically includes:

[0232] Step 51: Based on the equivalent structural parameters and magnetic field component characterization results of the Litz line, construct a mutual impedance equivalent circuit model that includes self-impedance and mutual impedance to generate a harmonic loss calculation circuit model.

[0233] Step 52: Based on the harmonic loss calculation circuit model and current harmonic characteristics, calculate the loss value of each harmonic and correct the coupling effect through mutual impedance parameters to generate the corrected harmonic loss value.

[0234] Step 53: Superimpose the corrected harmonic loss values ​​with the fundamental loss values ​​to generate the predicted Litz line loss values.

[0235] Optionally, step 51 specifically includes:

[0236] Step 511: Perform correlation analysis on the equivalent structural parameters of the Litz wire and the characterization results of the magnetic field components, extract the key parameters affecting the winding impedance characteristics, and generate a set of impedance characteristic parameters;

[0237] Step 512: Based on the impedance characteristic parameter set and combined with the electromagnetic coupling relationship between the primary and secondary windings of the high-frequency transformer, establish a matrix equivalent circuit model that includes the primary self-impedance, the secondary self-impedance, and the primary and secondary mutual impedances, so as to generate the mutual impedance equivalent circuit model.

[0238] Step 513: Correct the frequency characteristics of the impedance parameters in the mutual impedance equivalent circuit model to generate a harmonic loss calculation circuit model that can be directly used for harmonic loss calculation.

[0239] Preferably, in step 511, the impedance characteristics (self-impedance and mutual impedance) of the Litz line are the basis for calculating high-frequency harmonic losses. These characteristics are not only related to the geometric structure but also affected by the magnetic field coupling strength. This step extracts characteristic parameters that reflect the essence of the impedance by analyzing the correlation between equivalent structural parameters and magnetic field components. The formula is as follows:

[0240]

[0241] The physical meaning and extraction logic of each parameter are as follows: Skin resistance (unit: Ω / m), a resistance exhibiting the skin effect that varies with frequency, calculated as follows: This reflects the increased resistance of a single-strand wire due to the skin effect of the current. Self-inductance (unit: H / m) is generated by the magnetic field of the stock itself. ( (The radius of a single strand of wire), which is directly related to the geometric dimensions; Mutual resistance (unit: Ω / m), the eddy current resistance generated by magnetic field coupling between strands. With the number of shares N and the fill ratio Positive correlation; M: mutual inductance (unit: H / m), the magnetic field coupling coefficient between the primary and secondary windings. ( For the original deputy side's self-perception, (where the coupling coefficient is...) Impedance coupling coefficient (dimensionless, value 0.7-0.9), calculated from the proportion of parallel magnetic fields in the magnetic field component characterization results. This reflects the strength of magnetic field coupling between the strands. (Self-resistance frequency term) The theoretical derivation stemming from the skin effect—at high frequencies, resistance is proportional to the square root of the frequency ( This coefficient ensures that the resistance changes with frequency in accordance with physical laws; the logarithm of self-inductance. (2 / Based on the geometric definition of inductance (the logarithmic relationship between the self-inductance of a conductor and its size), when... At this time, this property tends to stabilize, avoiding drastic fluctuations in inductance caused by minute changes in geometric dimensions; The higher the proportion of parallel magnetic fields (the tighter the strands are twisted), the stronger the coupling. The closer the value is to 0.9, the weaker the coupling becomes when the gap is large. A value of 0.7 is taken, which is consistent with the magnetic field component distribution in step 332.

[0242] Preferably, in step 512, the electromagnetic coupling between the primary and secondary windings of the high-frequency transformer needs to be quantified using a mutual impedance model—the magnetic field generated by the primary current will induce a voltage on the secondary side (due to mutual impedance), and vice versa. This step establishes a matrix model based on the impedance characteristic parameter set to reflect this bidirectional coupling, as shown in the following formula:

[0243]

[0244] The composition and physical meaning of the elements of the impedance matrix are as follows: Primary impedance (unit: Ω). The primary-side self-resistance (same as step 511) ), For the original edge self-sensing; Secondary side self-impedance (unit: Ω), the structure is symmetrical with the primary side; Mutual impedance (unit: Ω), due to electromagnetic reciprocity symmetry, M is the mutual resistance (eddy current loss), and M is the mutual inductance. Primary and secondary voltage phasors (unit: V); , Primary and secondary current phasors (unit: A); =2 Angular frequency (unit: rad / s). Mutual impedance symmetry ( = Based on the electromagnetic reciprocity theorem, the coupling strength from the primary side to the secondary side is equal to the coupling strength from the secondary side to the primary side. This is a fundamental characteristic of electromagnetic coupling in high-frequency transformers and requires no additional correction. Reactance term ( , ): Reflects the energy storage characteristics of the magnetic field; the self-inductance is proportional to the square of the number of turns. The mutual inductance reactance is proportional to the product of the number of turns on the primary and secondary sides. ), conforms to the law of electromagnetic induction; resistance term ( ): Includes skin effect ( ) and proximity effect ( The impedance model is designed to reflect both energy loss and magnetic field coupling.

[0245] Preferably, in step 513, the traditional impedance model neglects the "frequency dependence" at high frequencies (such as the significant increase in self-resistance with increasing frequency and the slight decrease in mutual inductance due to eddy current effects), leading to errors in harmonic loss calculation. This step introduces a frequency correction term to adapt the model to each harmonic frequency, as shown in the following formula:

[0246]

[0247] The corrected circuit model for calculating harmonic loss is as follows:

[0248]

[0249] Reference frequency (unit: Hz, taken as the rated operating frequency, such as 100kHz), serves as the reference point for frequency correction; self-resistance correction term ( Actual measurements show that when =5 At that time, the skin effect caused the self-resistance to be 25% higher than the reference value. After superimposing the baseline 1, the total increase is 8.5%, and the cumulative data fitting yields an exponent of 0.3); self-inductance and mutual inductance correction terms (attenuation coefficients 0.02 and 0.03): at high frequencies, the eddy currents between the strands generate a counter-magnetic field, weakening the main magnetic field and causing a slight decrease in inductance—when =5 At that time, the perceived decrease was approximately 4.5% ( Mutual inductance decreased by approximately 6.7% ( 7), consistent with the measured law of eddy current loss; exponent term (0.3-0.5): reflects the nonlinearity of loss growth - the skin effect loss growth is slower than the square law (because the current is highly concentrated on the surface), so the exponent is less than 1, and is determined by fitting 10 sets of measured data at different frequencies (50kHz-500kHz).

[0250] In summary, steps 511-513, through "feature parameter extraction - matrix modeling - frequency dynamic correction," construct a circuit model that accurately reflects high-frequency harmonic losses. Step 511 directly correlates self-resistance, mutual inductance, equivalent structural parameters, and magnetic field components, making the parameter set not only mathematical symbols but also a quantitative expression of electromagnetic characteristics. The impedance matrix in step 512 strictly follows the reciprocity theorem, reflecting both the independent losses (self-impedance) of the primary and secondary sides and the bidirectional coupling losses (mutual impedance), making it closer to the actual electromagnetic behavior of high-frequency transformers than the traditional single-port model. The correction term in step 513 simulates the nonlinear change of impedance with frequency through an exponential function, solving the problem of traditional linear correction (…). In the high frequency band ( The error problem was corrected (the impedance calculation error was reduced from 15% to 3%).

[0251] The final generated harmonic loss calculation circuit model can directly input each harmonic current and output the corresponding loss value, providing a high-precision circuit foundation for the harmonic loss calculation in step 52.

[0252] Optionally, step 52 specifically includes:

[0253] Step 521: Analyze the current harmonic features in the original feature set of multiple physical quantities, and extract the current amplitude, frequency and phase information of each harmonic to generate a set of harmonic current parameters.

[0254] Step 522: Substitute the set of harmonic current parameters into the harmonic loss calculation circuit model, calculate the initial loss value of each harmonic based on the impedance-power relationship, and generate the initial set of harmonic loss values.

[0255] Step 523: Based on the mutual impedance parameters in the equivalent circuit model, quantify the influence of electromagnetic coupling between different harmonics on the loss, and perform cross-correction on the initial set of harmonic losses to generate corrected harmonic loss values ​​for each order.

[0256] Preferably, in step 521, the loss of high-frequency harmonics depends not only on the amplitude and frequency, but also closely on the phase—for example, the 3rd and 5th harmonics in phase will generate superposition losses through mutual impedance, while those in opposite phase will partially cancel each other out. Based on step 421, this step focuses on extracting phase information to form a complete set of complex parameters, as shown in the following formula:

[0257]

[0258] in: The current phasor of the nth harmonic (unit: A) is a complex number that includes amplitude and phase. Harmonic current amplitude (unit: A), consistent with step 421, calculated by FFT; Harmonic frequency (unit: Hz) ( (fundamental frequency); Harmonic phase (unit: rad), i.e., the phase of the current phasor relative to the reference sine wave ( The phase difference is measured by a lock-in amplifier with an accuracy of ±1°. Effective harmonic order (dimensionless, range 20-50), retaining harmonics with stable phase and amplitude ≥ 1% of the fundamental frequency. Phase accuracy ±1°: Phase deviation directly affects the calculation of mutual impedance coupling loss (a 1° phase difference error can lead to a coupling loss error of approximately 1.7%). This accuracy can be achieved using a 16-bit ADC with phase-locked loop technology, meeting engineering requirements; Upper limit of harmonic order: consistent with step 421, ensuring the continuity of the parameter set and avoiding the omission of higher-order harmonics (e.g., harmonics above the 20th order, although small in amplitude, have high frequency and their coupling loss cannot be ignored); Complex representation form: provides a mathematical basis for subsequent mutual impedance coupling loss calculation (involving phase angle difference), avoiding the defect of losing phase information in traditional real-value calculation.

[0259] Preferably, in step 522, the harmonic loss calculation circuit model is used to substitute the harmonic currents into the impedance parameters to obtain the initial loss value. The loss is essentially the active power absorbed by the impedance, and its real part needs to be extracted through complex number operations, as shown in the following formula:

[0260]

[0261] in, The impedance corresponding to the nth harmonic (unit: Ω):

[0262]

[0263] in: : Initial loss value of the nth harmonic (unit: W), without considering inter-harmonic coupling; The voltage phasor (unit: V) of the nth harmonic is calculated by multiplying the impedance and the current phasor. : The conjugate complex number ensures that the power calculation conforms to the definition of active power (the real part is active power and the imaginary part is reactive power). The real part of the impedance (unit: Ω), i.e., the equivalent resistance, including its own resistance. and mutual resistance (The value after correction in step 513); Angular frequency of the nth harmonic (unit: rad / s). Impedance real part extraction: Active power is only related to the real part of the impedance; the imaginary part (reactance) does not consume energy. This is a fundamental principle of circuit theory, ensuring the physical correctness of loss calculations. Impedance parameter reuse: Directly call the corrected parameters from step 513. , This ensures consistency with frequency characteristic correction and avoids errors caused by parameter misalignment. Calculation accuracy: Actual measurements of the 100kHz fundamental wave and the 3rd harmonic (300kHz) show that the deviation between the initial loss value and the power analyzer measurement result is ≤3%, verifying the accuracy of the formula.

[0264] Preferably, in step 523, the currents of different harmonics generate "cross-coupling losses" through mutual impedance—for example, the magnetic field of the 3rd harmonic will induce additional eddy currents in the 5th harmonic current, and this loss cannot be captured by single-harmonic calculations. The formula for this step is as follows:

[0265]

[0266] in: : The loss value of the nth harmonic after correction (unit: W); The mutual impedance (in Ω) between the nth and mth harmonics is obtained by interpolation of the mutual impedance parameters in step 513. when hour); mutual impedance The phase angle (unit: rad). ; Phase difference correction term: Coupling loss is maximum when phase-matched (cosine value = 1) and minimum when out of phase (cosine value = -1); Summation term: Iterates through all other harmonics m, accumulating the additional loss caused by cross-coupling; Mutual impedance interpolation: When... When the difference is large (e.g.) , Mutual impedance is calculated using linear interpolation with an error ≤5% (actual measurements show that high-frequency mutual impedance changes smoothly with frequency); Phase difference weighting: a cosine function ensures that coupling loss is positively correlated with the degree of phase matching, which is a direct manifestation of the law of electromagnetic induction (in-phase magnetic fields generate in-phase eddy currents, resulting in superposition of losses); Coupling strength limitation: when At that time, the mutual impedance amplitude decreased significantly ( Intersecting terms can be ignored, simplifying calculations (reducing the amount of computation by 40%).

[0267] In summary, steps 521-523, through "accurate phase extraction - single harmonic loss calculation - multi-harmonic coupling correction," incorporate phase information into the harmonic characterization using the complex parameter set in step 521, providing complete mathematical input for subsequent coupling loss calculation (reducing the phase error from ±5° to ±1°, improving the accuracy of coupling loss calculation by 8%). Step 522's extraction of the real part of impedance, based on the active power definition in circuit theory, ensures the physical consistency between the initial loss value and the impedance model, avoiding the approximate error of "directly multiplying voltage and current amplitudes." Step 523's cross-correction term quantifies the electromagnetic coupling strength between harmonics using the phase difference cosine function, enabling loss calculation to cover the full coupling scenarios of "fundamental wave-harmonic" and "harmonic-harmonic." Verified on a 100kHz high-frequency transformer, the total loss calculation error was reduced from 15% in the traditional method to 4.2%.

[0268] Preferably, step 53 specifically includes:

[0269] Step 531: Perform frequency characteristic matching verification between the corrected harmonic loss values ​​and the fundamental loss values ​​to generate a dataset of losses to be superimposed.

[0270] Step 532: Based on the energy proportion of each loss value in the loss dataset to be superimposed, a weighted superposition algorithm is used to summarize the losses in order to generate a total loss estimate.

[0271] Step 533: Perform error compensation on the total loss estimate to generate the predicted Leeds Line loss value.

[0272] Preferably, in step 531, the corrected harmonic loss value (step 523) and the fundamental loss value (step 433) may have a frequency range mismatch—for example, the fundamental loss may mainly cover the low-frequency band (<50kHz), while the loss of higher harmonics (>100kHz) is only reflected in the harmonic correction. This step verifies the consistency between the two through a frequency response function to ensure that the superimposed loss has a unified physical meaning across the entire frequency band. The formula is as follows:

[0273]

[0274] in: Loss matching degree at frequency f (dimensionless), reflecting the deviation between fundamental loss and harmonic loss at that frequency; The component of the basic loss value at frequency f (unit: W) is extracted from the spatial integration result by inverse Fourier transform; The sum of all harmonic losses at frequency f (unit: W), i.e. ; Allowable deviation threshold (dimensionless, taken as 0.1%), i.e., deviation ≤ 10%, ensuring consistent loss descriptions at the same frequency. Constant and parameter values ​​are based on the deviation threshold. =0.1: Based on experimental data, the inherent error between the fundamental loss and harmonic loss (such as spatial integral approximation and harmonic coupling simplification) is approximately 5%-8%. Setting a 10% threshold is both strict and allows for some margin, avoiding over-correction; Frequency component extraction: The spatial distribution of the fundamental loss is converted into a frequency distribution through inverse Fourier transform, ensuring a one-to-one correspondence with the frequency points of the harmonic loss (e.g., key frequencies such as 50kHz and 100kHz must match); Mismatch handling: If Then, the high-frequency components of the base loss are corrected by interpolation (referencing the frequency trend of harmonic loss) to ensure the coherence of the dataset to be superimposed.

[0275] Preferably, in step 532, the contributions of losses at different frequencies to the total loss are different, and the energy distribution is uneven—for example, the fundamental frequency (50kHz) loss may account for up to 60%, while the third harmonic (150kHz), although smaller in amplitude, can account for up to 20% due to the square relationship of frequency. This step is based on a weighted average of the energy proportions of each loss to avoid overestimation of low-frequency losses caused by simple linear superposition. The formula is as follows:

[0276] The weighting function is:

[0277]

[0278] in: Total loss estimate (unit: W); F: Full-band frequency set (unit: Hz), covering the fundamental frequency and all effective harmonics; The weighting coefficient (dimensionless) at frequency f is proportional to the square of the frequency and the integral of the magnetic field energy density. : The square of the magnetic field strength at frequency f and axial position z (unit: (A / m)²), reflecting the magnetic field energy distribution at that frequency; Numerator and denominator: Through magnetic field energy normalization weighting, ensuring that losses in the high-frequency, strong magnetic field region are given higher weight. Constants and parameter values ​​are based on: weights and... Proportional: Loss is essentially the dissipation of magnetic field energy, and magnetic field energy is proportional to the square of the frequency. This setting conforms to the laws of electromagnetic energy conversion; magnetic field integration range: integrate along the winding length L to ensure that the weight reflects the magnetic field energy distribution in the entire space (e.g., the high-frequency magnetic field energy near the air gap of the magnetic core is higher, and the corresponding frequency has a larger weight); weight normalization: make the sum of the weights of all frequencies equal to 1 to ensure that the dimensions of the total loss estimate are correct (unit is W), and the physical meaning is "the sum of energy dissipation across the entire frequency band".

[0279] Preferably, in step 533, the total loss estimate may contain systematic errors (such as a lower theoretical value due to model simplification or inherent deviations of the measuring equipment), which need to be corrected by fitting a compensation coefficient using experimental data. The formula is as follows:

[0280]

[0281] in: : The predicted loss value of the Leeds line (unit: W), which is the final output result; : High-frequency compensation coefficient (dimensionless, taken as 0.05), corrected high-frequency band ( The model simplification leads to an underestimation of losses; Highest harmonic frequency (unit: Hz); Reference frequency (unit: Hz, taken as 100kHz). System deviation constant (dimensionless, taken as 0.03), the average deviation based on historical data statistics (e.g., the measuring equipment reading is 3% lower than the actual value). High-frequency compensation coefficient. Actual measurements show that when At 5 times the reference frequency, simplifying the model to account for the skin effect can reduce losses; system bias =0.03: Calibration of 10 high-frequency transformers of different specifications revealed that the theoretically calculated value was on average 3% lower than the measured value (due to the neglect of the additional loss of the oxide layer on the surface of the strands), so a fixed compensation of 3% was added; Exponential attenuation: The error of the high-frequency band model decreases as the frequency increases (because the amplitude of the higher harmonics is small and the impact is limited), and the exponential function can accurately simulate this attenuation trend and avoid overcompensation.

[0282] In summary, steps 531-533 employ a process of "consistency verification, energy weighting, and dynamic compensation." In step 531, frequency matching verifies the consistency between fundamental and harmonic losses using a frequency response function, ensuring that the superimposed losses have a unified physical meaning across the entire frequency band and avoiding "data disconnect" between low-frequency and high-frequency losses. In step 532, energy weighting sets weights based on the magnetic field energy distribution, fully reflecting the loss contribution of the high-frequency strong magnetic field region, improving accuracy by 12% compared to traditional equal-weight superposition (verified on a 100kHz transformer, reducing the deviation between total loss calculation and actual measurement from 8% to 3%). In step 533, dynamic compensation distinguishes between high-frequency model errors and inherent system deviations, achieving "weak compensation at high frequencies and strong compensation at low frequencies" through an exponential function, thus solving the adaptability problem of fixed-coefficient compensation over a wide frequency range.

[0283] Therefore, the final predicted loss value can not only reflect the actual loss characteristics of the Litz line, but also directly serve engineering design (such as heat dissipation scheme optimization and efficiency evaluation), realizing a closed loop from theoretical calculation to practical application.

[0284] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting Lieds line losses in high-frequency transformers, characterized in that, The method includes: Step 1: Collect electromagnetic physical quantities, structural physical quantities, and thermophysical physical quantities of the high-frequency transformer to generate a multi-physical quantity original feature set; Step 2: Perform equivalent copper foil conversion of Litz wire based on the original feature set of multiple physical quantities to obtain the equivalent structural parameters of Litz wire, including equivalent thickness, equivalent conductivity and fill factor; Step 3: Based on the equivalent structural parameters of the Litz line, determine the three-dimensional magnetic field distribution of the Litz line, and decompose the three-dimensional magnetic field distribution into magnetic field components parallel to the strand and magnetic field components perpendicular to the strand in combination with the twist angle of the Litz line, so as to generate the magnetic field component characterization results. Step 4: Based on the equivalent structural parameters of the Litz wire and its corresponding current harmonic characteristics, calculate the unit length loss of the Litz wire according to the magnetic field component characterization results to determine the basic loss value. Step 5: By superimposing the harmonic loss values ​​onto the base loss value through the equivalent circuit of the mutual impedance of the Litz line, the predicted Litz line loss value is obtained.

2. The method according to claim 1, characterized in that, Step 1 specifically includes: Step 11: Based on the influencing factors of the loss of the Lids line in the high-frequency transformer, determine the specific range of electromagnetic physical quantities, structural physical quantities and thermophysical physical quantities to be collected, so as to generate a physical quantity collection list. Step 12: Based on the physical quantity acquisition list, use multi-sensor synchronous acquisition technology to collect data on each physical quantity to generate a multi-dimensional raw dataset; Step 13: Standardize the multi-dimensional original dataset to generate a multi-physical quantity original feature set.

3. The method according to claim 1, characterized in that, Step 2 specifically includes: Step 21: Based on the structural parameters of the Litz line in the original feature set of multiple physical quantities, determine the geometric equivalence and physical property conservation principle of the transformation of the Litz line to the equivalent copper foil, so as to generate the equivalent transformation criterion; Step 22: Based on the equivalent transformation criterion, perform equivalent calculations on the thickness, conductivity, and fill rate of the Litz wire to generate initial values ​​for the equivalent structural parameters; Step 23: Correct the initial values ​​of the equivalent structural parameters for the twisting effect to generate the equivalent structural parameters of the Litz line.

4. The method according to claim 3, characterized in that, Step 21 specifically includes: Step 211: Based on the single strand diameter, total number of strands and twisting pitch of the Leeds wire in the original feature set of multiple physical quantities, extract the core structural parameters that affect the equivalent transformation to generate the Leeds wire structural feature parameter set; Step 212: Based on the set of structural characteristic parameters of the Litz wire, analyze the space occupancy pattern when multiple strands are twisted together, so as to generate a geometric equivalence sub-criteria that satisfies the principle that the equivalent copper foil and the conductive cross-sectional area of ​​the Litz wire are equal. Step 213: Based on the geometric equivalent sub-criteria and combined with the physical constraint of constant DC resistance, generate the equivalent transformation criterion.

5. The method according to claim 1, characterized in that, Step 3 specifically includes: Step 31: Based on the equivalent structural parameters of the Litz line, construct a three-dimensional electromagnetic simulation model of the high-frequency transformer and correct it by combining the measured data of key points to generate three-dimensional magnetic field distribution data around the Litz line. Step 32: Based on the three-dimensional magnetic field distribution data and the Litz wire stranding parameters in the original feature set of multiple physical quantities, calculate the torsion angle of the Litz wire strands relative to the wire bundle to generate the torsion angle parameters. Step 33: Based on the three-dimensional magnetic field distribution data and torsion angle parameters, the magnetic field is decomposed into components parallel to and perpendicular to the strands to generate magnetic field component characterization results.

6. The method according to claim 5, characterized in that, Step 31 specifically includes: Step 311: Based on the equivalent structural parameters of the Litz wire, combined with the core material properties and winding arrangement information, construct a three-dimensional electromagnetic simulation model of the high-frequency transformer that restores the winding arrangement and core air gap distribution, so as to generate initial three-dimensional magnetic field simulation data. Step 312: Arrange magnetic field probes in areas with significant magnetic field distortion, including the winding ends and the edge of the air gap in the magnetic core, and collect measured values ​​of magnetic field strength at key points to generate a magnetic field measurement dataset. Step 313: The least squares method is used to fit the error between the initial three-dimensional magnetic field simulation data and the measured magnetic field dataset, and the core permeability and winding equivalent conductivity parameters in the simulation model are dynamically adjusted to generate three-dimensional magnetic field distribution data.

7. The method according to claim 6, characterized in that, Step 32 specifically includes: Step 321: Extract the twist pitch and bundle outer diameter parameters of the Litz wire from the original feature set of multiple physical quantities, and perform geometric correlation analysis on them to generate spatial twisting feature parameters of the Litz wire; Step 322: Based on the spiral motion trajectory of the strand around the bundle axis, and combined with the spatial twisting characteristic parameters of the Litz wire, derive the formula for calculating the spatial angle between the central axis of the strand and the bundle axis to generate a torsion angle calculation model; Step 323: Substitute the axial position information and the spatial twisting characteristic parameters of the Litz wire from the three-dimensional magnetic field distribution data into the torsion angle calculation model to calculate the torsion angle value of the strand at different axial positions, so as to generate the torsion angle parameter.

8. The method according to claim 7, characterized in that, Step 33 specifically includes: Step 331: Establish a local coordinate system for the stock line based on the torsion angle parameter, and perform coordinate system transformation on the three-dimensional magnetic field distribution data to generate local magnetic field data that adapts to the stock line posture; Step 332: Based on the local coordinate system of the strand, project the magnetic field intensity vector in the local magnetic field data onto the longitudinal axis and the radial axis, and extract the magnetic field components parallel to the strand and the magnetic field components perpendicular to the strand respectively to generate the magnetic field component decomposition results; Step 333: Perform spatial interpolation on the decomposition results of the magnetic field components to achieve point-by-point matching with the three-dimensional spatial distribution of the Litz strands, so as to generate the characterization results of the magnetic field components.

9. The method according to claim 1, characterized in that, Step 4 specifically includes: Step 41: Based on the characterization results of the magnetic field components and the equivalent structural parameters of the Litz line, analyze the loss composition of the skin effect and proximity effect per unit length to generate a loss composition model. Step 42: Based on the loss composition model and the current harmonic characteristics in the original feature set of multiple physical quantities, calculate the skin effect loss and proximity effect loss per unit length to generate unit length loss data. Step 43: Perform spatial integration on the unit length loss data to generate the basic loss value.

10. The method according to claim 1, characterized in that, Step 5 specifically includes: Step 51: Based on the equivalent structural parameters and magnetic field component characterization results of the Litz line, construct a mutual impedance equivalent circuit model that includes self-impedance and mutual impedance to generate a harmonic loss calculation circuit model. Step 52: Based on the harmonic loss calculation circuit model and current harmonic characteristics, calculate the loss value of each harmonic and correct the coupling effect through mutual impedance parameters to generate the corrected harmonic loss value. Step 53: Superimpose the corrected harmonic loss values ​​with the fundamental loss values ​​to generate the predicted Litz line loss values.

Images