An intelligent monitoring method for production of neodymium-iron-boron rare earth permanent magnet material

Abstract

The present application relates to the technical field of magnetic material detection, and discloses an intelligent monitoring method for production of neodymium-iron-boron rare earth permanent magnet material, comprising: using a magneto-optical Kerr image sequence driven by an applied magnetic field, obtaining surface avalanche cluster projection by difference between adjacent field steps, calculating area and shape tensor of the cluster, constructing power law relationship between cluster shape anisotropy ratio and cluster scale, and extracting shape anisotropy scale index of the surface projection of the avalanche cluster. Then, by establishing a scale spectrum varying with the field and forming a weighted integral index, nondestructive identification of internal low-coercivity weak layers and seepage channels generated by Tb / Dy grain boundary diffusion is realized. The present application breaks through the limitation of Kerr microscopy that can only observe the surface, realizes quantitative inversion of two-dimensional images to three-dimensional magnetic domain structure, and can be used for service state evaluation and failure warning of permanent magnets, and has high real-time performance, certainty and engineering implementability.

Description

Technical Field

[0001] This invention relates to the field of magnetic material testing technology, and more specifically, to an intelligent monitoring method for the production of neodymium iron boron rare earth permanent magnet materials. Background Technology

[0002] High-performance neodymium iron boron (Nd-Fe-B) rare-earth permanent magnets are core components for high-end applications such as new energy vehicles, wind power generation, and industrial motors. To improve coercivity while reducing the use of expensive heavy rare-earth elements (such as terbium (Tb) and dysprosium (Dy), grain boundary diffusion technology (GBDP) has become the mainstream process in the industry. This process involves penetrating heavy rare-earth elements from the magnet surface into the interior, forming a core-shell structure with high magnetocrystalline anisotropy on the grain surface, thereby significantly suppressing the formation of demagnetizing nuclei. However, when manufacturing large-size or bulk magnets, the non-uniformity of diffusion depth can easily lead to heavy rare-earth-depleted regions within the material, or even the formation of interconnected, low-coercivity weak layers or anti-core-shell structures. These microscopic defects are key factors causing magnet failure due to thermal demagnetization.

[0003] Currently, characterization of the microstructure of grain boundary diffused magnets mainly relies on destructive cross-section observation using electron microscopy (SEM / TEM) or non-destructive surface observation using magneto-optical Kerr microscopy (MEK). MEK utilizes the rotation of the polarization plane of light to dynamically image the magnetization state of the material surface in real time, enabling direct recording of the domain reversal process under an applied magnetic field. Existing domain image analysis methods are typically based on image processing techniques, assessing material uniformity by calculating the average grayscale of domains, the area ratio of reversal regions, or statistically counting the number of domains. For example, the magnetic performance quality of the material's surface layer can be qualitatively determined by identifying the migration velocity of domain walls during demagnetization or the reversal area threshold under a specific field strength.

[0004] However, existing monitoring methods based on the magneto-optical Kerr effect have significant technical limitations. First, the Kerr effect can only acquire a two-dimensional magnetization projection of the material surface, while the failure of grain boundary diffused magnets often originates from low-coercivity seepage channels in the internal three-dimensional space, making it difficult for surface observation to directly reflect the deep internal defect structure. Second, most existing image analysis algorithms only focus on scalar features such as the number or area of ​​magnetic domains, ignoring the crucial information contained in the morphological features of magnetic domains. In fact, magnetization reversal in NdFeB materials exhibits Barkhausen avalanche behavior, where the existing layered weak channels restrict the expansion of avalanche clusters in the thickness direction, forcing them to stretch along the weak layer direction, resulting in a specific shape anisotropy in the surface projection. Existing technologies lack the ability to quantitatively analyze the correlation between this surface projection shape and internal three-dimensional defects, and rely on a large number of manually set empirical thresholds (such as area thresholds), resulting in poor robustness of monitoring results in industrial production and making it difficult to achieve accurate, automated, and intelligent judgment of the health of the material's internal structure. Summary of the Invention

[0005] This invention provides an intelligent monitoring method for the production of neodymium iron boron rare earth permanent magnet materials, which solves the technical problems mentioned in the background art.

[0006] This invention provides an intelligent monitoring method for the production of neodymium iron boron rare earth permanent magnet materials, comprising: A monotonically varying demagnetizing field sequence is applied to NdFeB rare-earth permanent magnet materials, and a magnetic domain image sequence reflecting the evolution of the magnetization state on the material surface is acquired. The magnetization reversal process is analyzed based on the magnetic domain image sequence, and multiple magnetic domain reversal clusters occurring under each demagnetizing field intensity are extracted. For each magnetic domain reversal cluster, the corresponding magnetic domain cluster area is measured, and the magnetic domain morphological anisotropy characterizing geometric stretching characteristics is calculated. For each demagnetizing field intensity, an evolutionary correlation is established between the magnetic domain cluster area and the magnetic domain morphological anisotropy of all magnetic domain reversal clusters under that demagnetizing field intensity, and a magnetic domain morphological evolution scaling index reflecting the restricted characteristics of magnetic domain growth is extracted. Based on the magnetic domain morphological evolution scaling index, intelligent monitoring results of the magnetic domain structure characterizing the internal structural quality of the NdFeB rare-earth permanent magnet material are generated.

[0007] The beneficial effects of this invention include: quantitatively reflecting the three-dimensional low-coercivity percolation channels and weak layer structures within NdFeB permanent magnets caused by Tb / Dy grain boundary diffusion processes through surface Kerr magneto-optical imaging; and elevating traditional two-dimensional surface observations using Kerr microscopy to a quantifiable inversion of three-dimensional magnetic domain structures by constructing avalanche cluster projection shape anisotropy scaling exponent and extracting deep three-dimensional domain inversion features from a scale-anisotropy power law varying with the external field. Therefore, it can non-destructively and in real-time identify potential demagnetizing weak regions, significantly improving the reliability prediction capability of NdFeB permanent magnets and enabling intelligent monitoring of internal structures. Attached Figure Description

[0008] Figure 1 This is a flowchart of an intelligent monitoring method for the production of neodymium iron boron rare earth permanent magnet materials according to the present invention; Figure 2 This is a graph showing the evolution of the scaling exponent as a function of the magnetic field in this invention. Detailed Implementation

[0009] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0010] like Figure 1 As shown, an intelligent monitoring method for the production of neodymium iron boron rare earth permanent magnet materials includes: A monotonically varying demagnetizing field sequence is applied to NdFeB rare-earth permanent magnet materials, and a magnetic domain image sequence reflecting the evolution of the magnetization state on the material surface is acquired. The magnetization reversal process is analyzed based on the magnetic domain image sequence, and multiple magnetic domain reversal clusters occurring under each demagnetizing field intensity are extracted. For each magnetic domain reversal cluster, the corresponding magnetic domain cluster area is measured, and the magnetic domain morphological anisotropy characterizing geometric stretching characteristics is calculated. For each demagnetizing field intensity, an evolutionary correlation is established between the magnetic domain cluster area and the magnetic domain morphological anisotropy of all magnetic domain reversal clusters under that demagnetizing field intensity, and a magnetic domain morphological evolution scaling index reflecting the restricted characteristics of magnetic domain growth is extracted. Based on the magnetic domain morphological evolution scaling index, intelligent monitoring results of the magnetic domain structure characterizing the internal structural quality of the NdFeB rare-earth permanent magnet material are generated.

[0011] It should be noted that, since the demagnetization process of rare earth permanent magnets exhibits nonlinear avalanche characteristics, using linearly spaced field strength sampling can ensure the consistency of the time step when calculating the magnetic susceptibility or differential image, thereby avoiding human calculation errors introduced by changes in the sampling interval; and by normalizing the reference image of positive and negative saturation states, the linear response characteristics of the magneto-optical Kerr effect are utilized to decouple the light intensity signal into a magnetic signal.

[0012] In a preferred embodiment, the NdFeB magnet sample subjected to grain boundary diffusion treatment is first placed in the stage of a magneto-optical Kerr microscope, and the objective lens focal length is adjusted to obtain a clear image of the surface morphology. When setting the demagnetizing field sequence, the system first defines the initial positive saturation magnetic field strength. With the termination of the negative magnetic field strength .in, Based on the nominal coercivity of the material Values ​​are typically set. To ensure the sample reaches full magnetic saturation at the initial moment, for example, for a sample with a nominal coercivity of 1.5T, take... , Usually taken as Or a negative value sufficient to cause reverse saturation of the sample. The demagnetization process is discretized as... For the nth linearly equally spaced sampling points, sampling points ( The corresponding external magnetic field strength The calculation formula is: In the formula, Indicates the first The instantaneous magnetic field strength applied to the sample surface, expressed in Tesla (T) or Oersted (Oe); This is the preset total number of sampling steps, and its value must satisfy the requirements of Shannon's sampling theorem for capturing avalanche events. It is typically set to... To ensure sufficiently high field resolution, a CCD or CMOS camera simultaneously acquires images during the application of the magnetic field. Specifically, the system... Images are continuously acquired and stored as positive reference images, denoted as... ;exist (Right now Images are acquired and stored at specific times as negative reference images, denoted as... In each field step in the middle The system acquires the current raw reflection intensity image. In the image, Represents the two-dimensional discrete coordinates of pixels on an image sensor. This represents the grayscale value of the light intensity received by that pixel, typically an 8-bit (0-255) or 16-bit (0-65535) integer. Subsequently, the processor processes each frame... Perform normalization operation to calculate the normalized grayscale image. The calculation formula is as follows: In the formula, It is a range of values The dimensionless floating-point numbers between these values, by subtracting the negative background and dividing by the maximum magneto-optical contrast difference, completely eliminate non-magnetic background noise caused by scratches, unevenness, or uneven brightness distribution at the center of the light source on the sample surface. Finally, based on the material's saturation Kerr angular constant, the normalized grayscale is mapped to a magneto-optical Kerr angular distribution image. The mapping formula is: In the formula, Representing coordinates In the field strength The local magneto-optical Kerr rotation angle is expressed in degrees (deg) or milliradians (mrad). The preset material saturation Kerr angular constant is determined by the magneto-optical coefficient of the material itself. It can be pre-calibrated using an elliptic polarization spectrometer or taken directly from an empirical value (for example, for neodymium iron boron materials, it can be taken as...). ).

[0013] In some possible embodiments, taking into account electromagnetic interference or light source flicker that may exist in the industrial environment, when acquiring a forward reference image... With negative reference image At that time, instead of acquiring only a single frame image, a multi-frame temporal averaging method is used. Specifically, when the magnetic field stabilizes at... At that time, the system continuously collects data. Frames (e.g.) The original image is used to calculate its arithmetic mean. Similarly, handle At this point, the normalization calculation formula... In fact This processing method can reduce random shot noise in the reference image and prevent it from being affected by the denominator ( The algorithm improves its robustness in low signal-to-noise ratio environments by addressing the issue that excessive noise fluctuations in certain pixels can lead to singularities in the normalization results.

[0014] In some possible embodiments, the method for setting the demagnetizing field sequence is modified to accommodate materials with different coercivity distributions. When selecting multiple demagnetizing field intensities, although the overall trend remains monotonically decreasing, a piecewise linear, equally spaced approach can be used. Specifically, the system pre-defines a key focus area based on material parameters. Set a small magnetic field step size within this range. A larger magnetic field step size is set in the saturation region outside this range. At this time, the first Magnetic field at each sampling point The calculation no longer uses a linear formula, but is based on... The corresponding linear recursive formula is used for each paragraph. This allows for more precise capture of avalanche behavior near the coercivity with less data while maintaining the closed nature of the computational process. In subsequent steps involving the calculation of magnetic field intervals, a variable step size must be used accordingly. .

[0015] It should be noted that since the actual magnetization reversal occurs instantaneously in an avalanche-like manner, it is difficult to distinguish the reversed area from the background in static images. However, by calculating the difference images of adjacent field steps, the interference of static background textures (such as surface scratches and non-magnetic phase impurities) can be effectively eliminated, retaining only the dynamic region where magnetic moment deflection occurs within that field step interval. Furthermore, the introduction of adaptive normalization and threshold segmentation based on global standard deviation is to address the inherent photon shot noise and readout noise problems in imaging systems. By converting the Kerr rotation angle change into a statistically significant indicator (signal-to-noise ratio), the algorithm exhibits universal robustness on samples with different illumination intensities and different Kerr coefficients.

[0016] In a preferred embodiment, the processor receives a magnetic domain image sequence. Then, the temporal difference operation is performed first. Specifically, for the first... sampling points ( ), calculate the strength of two adjacent demagnetizing fields and The corresponding image differences are used to generate a difference image. The calculation formula is: In the formula, and The first Step and the first Step at pixel coordinates The magneto-optical Kerr rotation angle at that location, This characterizes the increment of magnetization state within the field interval. Next, the system performs statistical analysis on the difference image, first calculating the overall average value. Then the global standard deviation of the pixel values ​​is calculated. The calculation formula is as follows: In the formula, Image regions representing those involved in the statistics Total number of pixels within; The mean gray level of the difference image is theoretically close to zero, but it needs to be measured in practice to eliminate system drift. This represents the background noise level of the differenced image frame. After obtaining the noise baseline, the system uses this global standard deviation to normalize the differenced image, generating a dimensionless flip intensity field. The calculation formula is: .at this time, This represents the multiple of the signal strength at that point relative to the standard deviation of the background noise (i.e., (Multiple). Then, the system calls the preset intensity threshold. and Compare and generate a binarized mask. The comparison logic is: if ,but ,otherwise .in, It is a pre-defined dimensionless constant, whose value is based on statistical principles. The criteria are preferably set as follows: Therefore, only when the signal strength exceeds three standard deviations of the background noise is it considered a genuine domain reversal event, thus eliminating random noise interference with a 99.7% confidence level. Finally, the generated binarized mask... Perform Connected Component Labeling (CCL) analysis. The system uses an 8-neighborhood connectivity determination rule to scan the mask image and analyze all pixel values. Furthermore, pixels spatially connected by 8 neighborhoods (top, bottom, left, right, top left, top right, bottom left, bottom right) are grouped into a single object and assigned an index number. Thus, the first step of the field can be extracted. Magnetic domain flipping cluster ,in This is the set of all pixel coordinates that make up the cluster.

[0017] In some possible embodiments, considering that some industrial cameras may have dead pixels or salt-and-pepper noise, which could cause abnormally high difference values ​​for individual pixels, thus triggering a threshold and forming pseudo-avalanche clusters, a morphological filtering or area screening step is added after performing connected component labeling analysis. Specifically, after identifying the initial magnetic domain reversal clusters... Then, the system will count the number of pixels contained in each cluster. and compare it with the preset minimum area threshold. Compare them. To define the optical resolution limit of a microscopic system, for example, setting (pixels). If If the cluster is identified as noise, it will be removed from the mask (i.e., removed from the mask). (0). Thus, by introducing spatial scale constraints, the purity of magnetic domain reversal event extraction is further improved.

[0018] In some possible embodiments, the generation method of the flip intensity field is adjusted to accommodate situations with significant illumination gradients. When calculating the statistical properties of the difference image, instead of directly calculating the global standard deviation of the entire image, a block-based statistical strategy is employed. The image is divided into... (For example For each of the several sub-regions (pixels), calculate the local standard deviation within each sub-region. ,in Indexing for sub-regions. During normalization, pixels... Overturning strength at the point It is derived by dividing the pixel value of that point by the local standard deviation of its sub-region. This effectively addresses the problem of uneven noise distribution caused by the bright center and dark edges of the field of view (generally, the higher the light intensity, the greater the photon shot noise), ensuring that the weak avalanche signal at the edge of the field of view is not masked by the high-noise substrate at the center of the field of view, thereby improving the consistency of the entire field of view monitoring.

[0019] It should be noted that in magneto-optical Kerr microscopy, the three-dimensional Barkhausen avalanche event is projected as a connected region on a two-dimensional plane. Since the digital image sensor itself is composed of discrete photosensitive unit arrays, the total number of pixels constituting this connected region is the most direct and error-free statistical measure of the projected scale of the avalanche event. By accurately counting pixels, the complex topological geometry is compressed into a scalar. The area metric based on pixel count has translational and rotational invariance, ensuring that the monitoring results are not biased by the different positions or orientations of the avalanche clusters in the field of view.

[0020] In a preferred embodiment, after completing the connected component labeling analysis, the system obtains the first... The first demagnetizing field strength Magnetic domain flipping cluster This cluster is defined as a set containing several two-dimensional pixel coordinates. To determine the cluster area of ​​the magnetic domains of this cluster. The processor does not rely on any approximate geometric fit (such as the area of ​​a circumscribed rectangle or fitted ellipse), but instead performs a strict pixel traversal counting operation. The specific calculation formula is constructed as a summation of feature functions: In the formula, Represents the pixel coordinate index on the image plane; and These represent the width and height of the image, respectively. The indicator function is logically defined as: when the coordinates Belongs to set hour, ,otherwise The final calculation yielded It is a dimensionless non-negative integer representing the total number of photosensitive units occupied by the magnetic domain inversion cluster on the target surface of the imaging sensor. For example, if a cluster consists of 500 pixels, then... .

[0021] In some possible embodiments, considering that in actual industrial monitoring it may be necessary to convert the pixel area to the actual area for comparison with the microstructural dimensions of the material (such as grain size), the system introduces the calibration coefficient of the microscopic system for physical dimension conversion after counting the total number of pixels. Specifically, the calculation formula is extended as follows: .in, Pixel count (unit: pixels); This is the spatial resolution calibration coefficient of the system, which means the actual length corresponding to the side length of a single pixel, usually expressed in micrometers per pixel. ); The converted physical area of ​​the magnetic domain cluster is expressed in square micrometers. ).parameter The measurement is pre-calculated based on the microscope's objective lens magnification, relay lens magnification, and camera pixel size, or calibrated by taking images of a standard micrometer. For example, if the camera pixel size is 3.45... If the total amplification of the system is 10, then .

[0022] In some possible embodiments, for system environments employing run-length encoding (RLE) for image compression or storage, determining the magnetic domain cluster area does not require reconstructing a two-dimensional matrix for point-by-point counting; instead, it is calculated directly based on the compressed data to improve efficiency. Specifically, magnetic domain flipping clusters... It is stored as a series of run segments, i.e. ,in For the index of the run segment, This represents the total number of run segments contained in the cluster. At this point, the magnetic domain cluster area is... The calculation formula is adjusted as follows: In the formula, and Representing the first The difference between the end coordinates and start coordinates (including the boundary) of each run segment in the row direction, plus 1, is the number of pixels contained in the continuous line segment. This represents the total number of discrete line segments that make up the cluster. Therefore, when dealing with large, complex avalanche clusters, it can significantly reduce the number of memory accesses and greatly improve the algorithm's running speed on embedded monitoring devices.

[0023] It should be noted that in NdFeB magnets treated with grain boundary diffusion (GBD), the high coercivity shell structure formed by the infiltration of heavy rare earth elements (such as Tb / Dy) along the grain boundaries causes the originally isotropic three-dimensional avalanche behavior to undergo restricted growth when encountering these hard shells or undiffused soft channels. This restricted growth manifests as significant stretching of magnetic domain clusters along specific directions (usually the direction of weak magnetic channels) on a two-dimensional projection plane. By constructing a shape tensor and solving for eigenvalues, we are essentially looking for the principal axis (maximum extension direction) and secondary axis (minimum extension direction) of the material distribution of the magnetic domain cluster. The ratio of these two axes reflects the anisotropic strength of the pinning or guiding effect of the microstructure on the movement of the magnetic domain walls, and is a quantitative indicator for judging whether the internal microstructure of the material is uniform.

[0024] In a preferred embodiment, the system targets each extracted area as... Magnetic domain flipping clusters First, calculate the coordinates of its geometric center (i.e., its centroid). Suppose this cluster contains... The pixel, the The coordinates of the pixels are ,in Geometric center coordinates The calculation formula is: as well as In the formula, and For column and row indices in the image coordinate system, and These are the calculated centroid floating-point coordinates. Then, based on the spatial distribution of all pixels within this cluster relative to the geometric center, a... The real symmetric shape tensor (also known as the second-order central moment matrix or rotation tensor). The matrix representation of this tensor is as follows: The formulas for calculating each element in the matrix are as follows: , , .in, Characterizes the dispersion of pixels along the X-axis. Characterizes the dispersion in the Y-axis direction. The correlation (i.e., covariance) between the X and Y directions is characterized. Next, the shape tensor is solved. The eigenvalues ​​of . According to linear algebra theory, eigenvalues ​​are... Satisfy the characteristic equation The specific calculation formula is as follows: The system defines the larger value in the calculation result as the first eigenvalue. The smaller value is defined as the second eigenvalue. (Right now The corresponding "+" term in the formula (Corresponding to the - sign item). These two eigenvalues ​​characterize the moments of inertia along the major and minor axes of the equivalent ellipse corresponding to the fitted domain cluster, respectively. Finally, the domain morphological anisotropy is calculated. The calculation formula is: In the formula, It is a dimensionless scalar greater than or equal to 1. When the magnetic domain cluster is circular, ,but When magnetic domain clusters appear as long, thin strips, ,but The numerical value increases, thus quantitatively characterizing the degree of geometric stretching of the cluster.

[0025] In some possible embodiments, considering that magnetic domain flipping is not a simple binary behavior and the degree of flipping may vary at different locations, a weighted shape tensor is introduced to reflect the centroid of magnetic moment variation. Specifically, a flipping intensity field is introduced when calculating the geometric center and the shape tensor. As a weighting factor At this point, the formula for calculating the geometric center is adjusted to: The formula for calculating shape tensor elements has been adjusted as follows: (The same applies to other elements). Among them, Representing the The flip significance weight of each pixel. Specifically, the more drastic the flip, the greater the contribution to morphological anisotropy, thereby effectively suppressing the interference of blurred pixels (low weight) at the edge of the flip cluster on the overall shape judgment, so that the anisotropy degree can better reflect the growth direction of magnetic domains.

[0026] In some possible embodiments, to enhance the stability of the algorithm in extreme cases, especially for certain extremely thin or only single-pixel-wide linear avalanche clusters (where the second eigenvalue is...), (Potentially close to 0), a small regularization constant is introduced when calculating the final ratio. At this point, the anisotropy degree The calculation formula is revised as follows: .in, For a preset, extremely small positive number (e.g.) The value of this parameter is based on the floating-point precision limit of the computing platform. The introduction of this parameter prevents the risk of program crashes due to a zero denominator, while smoothing out extremely small-scale noise fluctuations, ensuring the robustness and continuous operation capability of the algorithm in fully automated industrial monitoring software.

[0027] It should be noted that the demagnetization process of grain boundary diffused NdFeB permanent magnets belongs to a typical domain wall pinning and avalanche model in disordered media. Theoretical predictions indicate a nonlinear self-similarity relationship between the avalanche scale (area) and its geometric morphology (anisotropy), i.e. Analyzing this nonlinear relationship directly in the original linear coordinate system is extremely difficult and sensitive to noise. However, through a double log-log transformation, this potential power-law relationship can be transformed into a linear relationship, thus making the scaling exponent... Extraction of the slope of a straight line becomes possible.

[0028] In a preferred embodiment, the system at a specific demagnetizing field strength Below, it has been identified and calculated The domain inversion cluster, targeting the first one Clusters ( Its area value has been stored in the system memory. and morphological anisotropy value The processor iterates through all fields under that field strength. Each cluster is processed using its own natural logarithm. Specifically, the logarithm of the area is calculated first. The calculation formula is: In the formula, Dimensionless pixel count or area value (must be guaranteed) ), For the natural constant Logarithmic function with base 0. The transformed abscissa variable represents the order-of-magnitude exponent of the domain flipping scale. Next, the logarithmic value of the anisotropy degree is calculated. The calculation formula is: In the formula, Anisotropy ratios greater than or equal to 1 Also a natural logarithm function The transformed ordinate variable must have a value greater than or equal to 0 (because...) This represents the order of magnitude of the degree of morphological stretching. After completing the calculation, the system will assign each pair... Combined into a two-dimensional data point, and all data under that field strength... The data points corresponding to each cluster are aggregated to form a data point set. Data point set This constitutes the training set for the linear regression model, where It is defined as the independent variable (explanatory variable). Defined as the dependent variable (response variable), the growth scale of the magnetic domain cluster is the cause, while the morphological stretching caused by structural constraints is the result of scale variation.

[0029] In some possible embodiments, to improve the signal-to-noise ratio of the data point set, the set is constructed... Previously, a data cleaning step was added. Due to pixel discretization effects, the shape tensor of extremely small magnetic domain clusters (e.g., clusters with an area less than 5 pixels) is affected far more by the pixel arrangement (e.g., cross-shaped, square) than by anisotropy, potentially introducing spurious quantum noise. Therefore, before performing logarithmic operations, the system introduces an effective area lower bound. The specific filtering logic is as follows: If If the cluster does not participate in subsequent logarithmic operations and set construction, then the cluster will not participate in subsequent logarithmic operations and set construction. The value is determined based on the point spread function (PSF) size of the imaging system, and is typically set to [value missing]. (Right now (pixel region). Only if it meets the following conditions. Only clusters will be converted into point pairs. And add it to the set.

[0030] It should be noted that the slope (i.e., scaling exponent) of the linear regression model reflects the degree to which the growth mechanism of magnetic domain clusters is constrained by geometric space. If the internal microstructure of the material is homogeneous and isotropic, the magnetic domains will tend to expand in a circular pattern, and the logarithmic relationship between their area and shape factor should show a weak correlation or zero slope. However, in grain boundary diffused NdFeB magnets, due to the presence of a highly coercive shell perpendicular to the diffusion direction, large-sized magnetic domain clusters are forced to grow flat along low-barrier channels (i.e., undiffused soft magnetic phase cores or planar weak magnetic regions). Therefore, the slope value extracted from the experimental data by the least squares method is essentially an average measure of the constraint strength of the anisotropic defect channels within the material on the motion of the magnetic domain walls. This average measure has statistical robustness and can effectively filter out outliers caused by lattice defects or imaging noise.

[0031] In a preferred embodiment, the processor receives the first Demagnetization intensity The constructed data point set ,in This represents the total number of sample points under this field strength (i.e., the number of domain-flipping clusters). This is used to establish a linear regression model describing the evolutionary correlation. The system executes a standard univariate linear least squares fitting algorithm. First, it calculates the arithmetic mean of all data points on both the independent and dependent variables, using the following formulas: as well as .in, This represents the logarithmic mean of the size of the magnetic domain clusters under that field strength. This represents the logarithmic mean of the morphological anisotropy. Subsequently, based on the objective function of minimizing the Residual Sum of Squares (RSS), the slope of the regression line is calculated. The formula for calculating the closed-form solution of the slope is: In the formula, the numerator term... The sample covariance (unnormalized) representing the logarithmic values ​​of area and anisotropy reflects the synergistic effect of their changes; the denominator term This represents the sample variance (unnormalized) of the logarithmic area, used to normalize the dispersion of the independent variable. The calculated... This is the magnetic domain morphology evolution scaling index. The index is a dimensionless real number that typically fluctuates between 0 and 1. The larger the value, the more the morphology of the magnetic domain cluster tends to undergo severe stretching deformation as the area of ​​the magnetic domain cluster increases, thus revealing the existence of more significant interconnected weak magnetic channels inside the material.

[0032] In some possible implementations, considering the potential presence of significant outliers in the experimental data, such as pseudo-large clusters caused by surface dust or scratches, which could severely distort the least squares fitting results, a Theil-Sen Estimator is used instead of ordinary least squares to extract the slope. Specifically, the system first calculates the set... Any two data points and The slope between (in For those containing A set of points can yield at most [number] points. A pair of slope values. The system's final output scaling exponent. Take all these paired slopes The median is used to accurately reconstruct the mainstream morphological evolution trend even with up to 29% arbitrary contamination in the dataset. This makes it particularly suitable for monitoring scenarios with harsh industrial environments and unstable imaging quality.

[0033] In some possible implementations, to simultaneously consider the measurement errors inherent in both the independent variable (area) and the dependent variable (anisotropy) (i.e., error-in-variables models), the system employs Orthogonal Distance Regression (ODR) or Total Least Squares (TLS) instead of Ordinary Least Squares. Ordinary Least Squares assumes that only... There is an error in the axial direction, and The axis (area) is precise, but this is not entirely accurate in pixelated imaging. Specifically, the goal of the fit is no longer to minimize... The residuals are not in the direction of the regression line, but rather the sum of squares of the perpendicular distances from the data points to the regression line are minimized. Slope The calculation involves the eigenvalue decomposition of the data covariance matrix, specifically by solving the equation The root is obtained. This allows for structural parameter estimation that is closer to the true law than ordinary least squares method, and is especially suitable for the research and development of high-end magnets where monitoring accuracy is extremely important.

[0034] It should be noted that since the magnetic field points sampled in the experiment are discrete, directly using discrete points for evaluation may miss the details of rapid changes near the coercivity. Therefore, an interpolation algorithm is introduced to construct a continuous function to reconstruct the full picture of the physical process. Furthermore, considering that the performance degradation of NdFeB magnets (such as the decrease in squareness) is mainly related to domain wall pinning failure near the coercivity, while the domain behavior in the saturation region has a smaller impact on the final performance, a weighted integral mechanism centered on the nominal coercivity is designed to amplify the signal weight in the critical region and suppress noise in non-critical regions. Finally, a negative exponential decay mapping is adopted to transform the unbounded cumulative quantity into a health score (0 to 1 range) that conforms to cognition, establishing a clear quality grading standard.

[0035] In a preferred embodiment, the system first calls the discrete data pairs stored in memory. The fitting was performed using the cubic spline interpolation algorithm. This algorithm constructs a cubic polynomial between every two adjacent data points, ensuring the continuity of the first and second derivatives at the connection points, thus obtaining a result covering the entire demagnetization range. Continuous evolution curve function Next, the system constructs a coercivity in the name of materials. center normal distribution weight function The formula for calculating the normal distribution weight function is: In the formula, For magnetic field variables; The pre-entered standard coercivity value (e.g., 2.0T) of this grade of magnet is used as the peak position of the weight; The standard deviation parameter determines the width of the interval of interest, and its value is usually set to 1. A fixed ratio, for example To ensure that the weights cover approximately the coercivity area The fluctuation range is then determined. Subsequently, the evolution curve function is weighted and integrated using this weighting function to calculate the cumulative anisotropy index. The integral formula is designed in normalized form: At the computational level, this integral is achieved through discretization and summation on a subdivided mesh using Composite Simpson's Rule. The calculated... The value represents the weighted average intensity of domain morphological anisotropy during the most critical demagnetization stage. Finally, to obtain intuitive monitoring results, the system utilizes a negative exponential decay function to... Mapped to The calculation formula is: In the formula, The preset scale constant (ToleranceScale) is based on a large number of historical high-quality qualified products. Set a statistical average or upper limit for the value (e.g., twice the average value of qualified products). The rigor of the evaluation system is determined by its standards. This is the final output of the intelligent monitoring result of the magnetic domain structure, and its value range is within... Between these values, the closer the value is to 1, the more complete the internal microstructure of the material (extremely low anisotropy), while a value close to 0 indicates the presence of severely interconnected weak magnetic channels inside.

[0036] In some possible embodiments, the choice of interpolation algorithm is adjusted, using least-squares polynomial fitting instead of spline interpolation. Specifically, the system uses a higher-order polynomial (e.g., 5th order). To globally approximate the discrete point sequence, where, This represents the weighting coefficient of the nth term in a higher-order polynomial. This method not only connects data points but also smooths out random measurement noise in experimental data, preventing interpolation curve oscillations (Runge phenomenon) caused by individual abnormally high points. At this point, the evolution curve function... That is, the polynomial .

[0037] In some possible embodiments, the form of the weighting function is modified, using a Lorentzian function instead of a Gaussian function. Specifically, the weighting function... The calculation formula is adjusted as follows: .in, The half-width at half-height (HWHM) parameter is used to set the width of the region of interest, for example, taking... The Lorentz function has a longer tail than the Gaussian function, thus still focusing on the magnetic domain behavior near the coercivity point, but also assigning relatively higher weights to regions far from the coercivity point (i.e., the early and late stages of demagnetization). This makes it suitable for scenarios with extremely high product consistency requirements, where even morphological anomalies caused by minute structural defects far from the coercivity point should not be too quickly attenuated and ignored, thus providing a more stringent monitoring standard than Gaussian weighted methods.

[0038] like Figure 2 As shown, Figure 2The graph shows the evolution curve of the magnetic domain morphology scaling exponent as a function of the demagnetizing field strength. The horizontal axis represents the normalized demagnetizing field strength, and the vertical axis represents the anisotropic scaling exponent. The fluctuations in the curve visually reflect the dynamic law of the restricted magnetic domain growth characteristics of NdFeB materials during demagnetization, with the peak value typically corresponding to the coercive region where magnetization reversal is most intense.

[0039] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A method for intelligent monitoring of production of neodymium-iron-boron rare earth permanent magnet material, characterized in that, include: A monotonically varying demagnetizing field sequence was applied to neodymium iron boron rare earth permanent magnet materials, and a magnetic domain image sequence reflecting the evolution of the magnetization state on the material surface was collected. The magnetization reversal process is analyzed based on the magnetic domain image sequence, and multiple magnetic domain reversal clusters occurring under each demagnetization field strength are extracted; for each magnetic domain reversal cluster, the corresponding magnetic domain cluster area is measured, and the magnetic domain morphological anisotropy characterizing the geometric stretching features is calculated; For each demagnetizing field intensity, an evolutionary correlation is established between the domain cluster area and the domain morphological anisotropy of all said domain-flipping clusters under that demagnetizing field intensity, including: Perform natural logarithmic calculations on the area of ​​each domain inversion cluster and the anisotropy of the domain morphology for each domain inversion cluster to obtain the corresponding area logarithm and anisotropy logarithm. Use the area logarithm as the independent variable and the anisotropy logarithm as the dependent variable to construct a data point set containing all domain inversion clusters under the demagnetization field strength. Extracting a magnetic domain morphology evolution scaling index that reflects the restricted growth characteristics of magnetic domains includes: performing a linear least squares fitting operation on the data point set to establish a linear regression model describing the evolutionary correlation; extracting the slope value of the linear regression model and determining the slope value as the magnetic domain morphology evolution scaling index; Based on the magnetic domain morphology evolution scaling index, intelligent monitoring results of the magnetic domain structure, characterizing the internal structural quality of the NdFeB rare earth permanent magnet material, are generated, including: An interpolation algorithm is used to fit the domain morphology evolution scaling index under various demagnetization field intensities to establish an evolution curve function describing the continuous change of the domain morphology evolution scaling index with the demagnetization field intensity. A normal distribution weight function centered on the nominal coercivity value of the NdFeB rare earth permanent magnet material is constructed. The evolution curve function is weighted and integrated over the entire demagnetization field variation range using the normal distribution weight function to obtain the anisotropic cumulative index. Based on a preset scaling constant, the anisotropic cumulative index is numerically mapped using a negative exponential decay function to obtain a normalized mapping value, which is then used as the intelligent monitoring result of the domain structure.

2. The intelligent monitoring method for the production of Nd-Fe-B rare earth permanent magnet material according to claim 1, characterized in that, A monotonically varying demagnetizing field sequence was applied to neodymium iron boron rare-earth permanent magnet materials, and a magnetic domain image sequence reflecting the evolution of the magnetization state on the material surface was acquired, including: A demagnetization process is set up, starting from a positive saturation magnetic field and monotonically decreasing to a negative magnetic field. Multiple demagnetization field intensities are selected in a linear and equally spaced manner. At each of the demagnetization field intensities, the original reflection intensity image of the material surface is acquired, and a positive reference image of the material in the positive saturation state and a negative reference image in the negative saturation state are obtained respectively. A normalized grayscale image is calculated using the original reflection intensity image, the positive reference image, and the negative reference image, and the normalized grayscale image is mapped to a magneto-optical Kerr angular distribution image representing the local magnetization direction of the surface, which is used as the magnetic domain image sequence. 3.The intelligent monitoring method for production of Nd-Fe-B rare earth permanent magnet material according to claim 2, characterized in that, Based on the analysis of the magnetic domain image sequence, the magnetization reversal process is analyzed, and multiple magnetic domain reversal clusters occurring under various demagnetization field intensities are extracted, including: Calculate the difference image between the magneto-optical Kerr angular distribution images corresponding to two adjacent demagnetization field intensities in the magnetic domain image sequence; calculate the global standard deviation of the pixel values ​​in the difference image, and normalize the pixel values ​​in the difference image using the global standard deviation to obtain the flip intensity field; compare the flip intensity field with a preset intensity threshold to generate a binary mask identifying the magnetization flip region; perform connected component labeling analysis on the binary mask to identify and label the spatially connected pixel sets as the magnetic domain flip clusters.

4. The intelligent monitoring method for the production of Nd-Fe-B rare earth permanent magnet material according to claim 3, characterized in that, For each of the aforementioned domain inversion clusters, the corresponding domain cluster area is measured, including: The total number of pixels constituting each domain inversion cluster is counted, and the total number is used as the area of ​​the domain cluster.

5. The intelligent monitoring method for the production of Nd-Fe-B rare earth permanent magnet material according to claim 4, characterized in that, Calculate the anisotropy of magnetic domain morphology characterizing geometric stretching features, including: Calculate the geometric center coordinates of each domain-flipping cluster; construct a shape tensor describing the geometry of the domain-flipping cluster based on the spatial distribution of all pixels within the domain-flipping cluster relative to the geometric center coordinates; solve for the first and second eigenvalues ​​of the shape tensor, wherein the first eigenvalue is greater than the second eigenvalue; calculate the square root of the ratio of the first eigenvalue to the second eigenvalue, and use the square root as the anisotropy of the domain morphology.

Images