Method for regulating grain size and particle size distribution of coarse-grained layer of NdFeB magnet

Abstract

The present invention discloses a method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnets, which specifically relates to the technical field of NdFeB magnet material preparation, comprising: preparing a powder preform with uniform density through multi-stage pressure forming and densification treatment; collecting the local temperature change rate and stress change rate in real time during the sintering heating and insulation stages; triggering the coarse-grained layer tissue evolution evaluation based on monitoring data, and extracting a feature data group; calculating the diffusion coupling unevenness index and the grain boundary migration tensor index based on the feature data group, inputting the data into a machine learning model, outputting the abnormal risk level of the coarse-grained layer, and dynamically adjusting the sintering process parameters accordingly; the present invention dynamically perceives the abnormal evolution of the coarse-grained layer by combining real-time monitoring of local temperature and stress changes, and intelligently predicts the abnormal risk level of the coarse-grained layer grain size based on the feature data group, so as to timely suppress the abnormal growth of coarse grains and improve the microstructure consistency and service performance stability of the NdFeB magnet.

Description

Technical Field

[0001] The present invention relates to the technical field of preparation of NdFeB magnet materials, and more particularly to a method for regulating the grain size and particle size distribution of a coarse-grained layer of a NdFeB magnet. Background Art

[0002] Neodymium iron boron magnets are widely used in electronics, automotive, and new energy fields due to their excellent magnetic properties. However, the microstructure of neodymium iron boron magnets, especially the grain size and particle size distribution of the coarse-grained layer, has a significant impact on the magnet's ultimate performance stability and reliability. During the traditional sintering process, the grain size of the coarse-grained layer is easily affected by factors such as uneven local thermal field distribution, unreasonable sintering heating rates, and local stress concentration. This can lead to localized abnormal growth or uneven particle size distribution, which in turn causes fluctuations in magnetic properties and reduces the material's service reliability.

[0003] In the existing technology, the uniformity of the structure is usually improved by optimizing the initial forming pressure, adjusting the sintering curve or adding an external temperature control device. However, these methods are mostly based on static process settings and cannot respond in time to the risk of mutation caused by local abnormal evolution during the sintering process. In addition, the abnormal evolution of the grain size of the coarse-grained layer often occurs on the sintering temperature platform or during the cooling process, and has the characteristics of spatial locality and temporal suddenness. Traditional methods are difficult to effectively detect and intervene in the early stages of evolution. Therefore, the present invention proposes a method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnets in order to solve the above problems. Summary of the Invention

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] The method for regulating the grain size and particle size distribution of the coarse-grained layer of a NdFeB magnet comprises the following steps:

[0006] In the first step, a multi-stage pressure forming and densification process is used to prepare a powder preform with uniform density distribution and an overall density within a preset range;

[0007] In the second step, during the sintering heating and holding stages, multiple temperature sensors and strain sensors pre-installed inside and on the surface of the preform are used to obtain real-time data on local temperature changes and local stress changes at each location.

[0008] The third step is to determine whether to trigger the evaluation of the coarse-grained layer microstructure evolution based on the real-time monitored local temperature change rate and local stress change rate. When the local temperature change rate or local stress change rate exceeds the preset threshold, a feature data set including the local temperature change rate, local stress change rate, local diffusion rate estimate, and local grain boundary migration rate estimate is extracted.

[0009] In the fourth step, the diffusion coupling unevenness index and grain boundary migration tensor index are calculated based on the characteristic data set, and are fed into the pre-trained machine learning model as input to output the risk level of abnormal grain size in the coarse-grained layer. The sintering process is then dynamically adjusted based on the output risk level.

[0010] In a preferred embodiment, the multi-stage pressure forming and densification treatment method includes the following steps: first, the NdFeB powder raw material is preliminarily pressed under low pressure conditions to form a preliminary preform with a continuous structure and an internal porosity lower than that of the initial powder state; then, on the basis of applying uniform pressurization as a whole, multiple pressure loading processes are continuously superimposed, and the amplitude and duration of each loading are gradually increased according to a preset pressure increase curve, and the direction of the pressure applied during the pressurization process is kept stable and consistent; during the entire pressurization process, high-frequency micro-vibration is synchronously applied.

[0011] In a preferred embodiment, high-frequency micro-vibration is continuously applied throughout the multi-stage pressure forming and densification process, and the frequency of high-frequency micro-vibration is maintained in the range of 1,000 Hz to 100,000 Hz. The direction of applying high-frequency micro-vibration is set at a preset fixed angle with the pressurization direction. Continuous vibration energy input triggers local microscopic displacement and rotation adjustment of powder particles, enhances the contact probability and bonding strength between particles, reduces local internal porosity, and prevents powder particles from agglomerating or forming density-uneven aggregation areas during unidirectional pressurization. While applying high-frequency micro-vibration, the pressure direction of the overall pressurization process is maintained stable to ensure that the direction of particle rearrangement is consistent with the pressing direction, thereby promoting the high uniformity of the internal microstructure of the pressed powder preform, thereby laying the foundation for subsequent temperature and stress monitoring and dynamic regulation of the grain size of the coarse-grained layer.

[0012] In a preferred embodiment, in the process of determining whether to trigger the coarse-grained layer microstructure evolution evaluation based on the real-time monitoring of the local temperature change rate and the local stress change rate, the changing trends of the local temperature change rate and the local stress change rate are continuously monitored. When any local temperature change rate or local stress change rate exceeds the corresponding preset threshold condition, the coarse-grained layer microstructure evolution evaluation process is immediately triggered;

[0013] The preset threshold conditions are based on process settings, and the static calibration values ​​are set in different ranges according to different sintering stages to ensure that potential abnormal changes are captured in real time during key process sections, avoiding abnormal growth of grain size and particle size distribution in the coarse-grained layer due to delayed response, and improving the timeliness and accuracy of the trigger mechanism.

[0014] In a preferred embodiment, after triggering the coarse-grained layer microstructure evolution assessment, a feature data set is extracted, the feature data set consisting of a local temperature change rate, a local stress change rate, an estimated local diffusion rate, and an estimated local grain boundary migration rate;

[0015] The local diffusion rate estimate is obtained by obtaining local real-time temperature data and calculating it according to the diffusion rate calculation formula. The diffusion rate calculation formula is that the local diffusion rate is equal to the initial diffusion rate multiplied by an exponential function. The negative exponential term of the exponential function is obtained by dividing the diffusion activation energy by the product of the local real-time temperature and the gas constant. The local diffusion rate is calculated according to the following formula:

[0016] ;

[0017] Indicates that the monitoring point is at position Place, time The diffusion rate when is the reference initial diffusion rate constant, is the material diffusion activation energy, is the gas constant, take 8.314, Indicates that the monitoring point is at position Place, time Real-time temperature at the time;

[0018] The estimated value of the local grain boundary migration rate is obtained by multiplying the stress gradient measured by the local stress change rate by the linear response coefficient of the grain boundary migration driving force. The linear response coefficient of the grain boundary migration driving force is preset based on the internal grain boundary energy and grain size calibration data of the NdFeB magnet material;

[0019] Each data item in the feature data group is synchronously paired according to the timestamp.

[0020] In a preferred embodiment, the diffusion coupling heterogeneity index is derived based on the discrete spatiotemporal distribution of the local diffusion rate estimate, specifically comprising the following steps:

[0021] At each moment, the set of local diffusion rate estimates of all monitoring points in space is normalized. The normalization method is to subtract the global minimum diffusion rate from the local diffusion rate and divide it by the global diffusion rate range at the current moment to form a dimensionless normalized diffusion rate field.

[0022] Based on the normalized diffusion rate field, the spatial statistical variance of the normalized diffusion rate at that moment is calculated to quantitatively characterize the degree of dispersion of the local diffusion rate distribution. The larger the spatial statistical variance, the more uneven the spatial distribution of the diffusion rate.

[0023] The spatial statistical variance corresponding to each moment is cumulatively integrated or averaged along the time dimension to obtain the time-weighted cumulative index of spatial diffusion heterogeneity;

[0024] The diffusion coupling heterogeneity index is defined as the normalized value of the cumulative index of spatial diffusion heterogeneity. Finally, the diffusion coupling heterogeneity index quantifies the overall spatial discreteness level of the local diffusion process during the entire sintering cycle or cooling cycle. The higher the value, the worse the consistency of the diffusion behavior during the evolution of the coarse-grained layer organization and the higher the risk of abnormal coarse-grained growth.

[0025] In a preferred embodiment, the grain boundary migration tensor index is used to quantify the anisotropic distribution characteristics of the grain boundary migration rate within the coarse-grained layer of the NdFeB magnet on a spatial scale to evaluate the consistency level of grain boundary movement during the evolution of the coarse-grained layer. The calculation process includes the following steps:

[0026] Based on the estimated values ​​of local grain boundary migration rate collected at different monitoring positions, the local grain boundary migration velocity vector field is constructed, and the grain boundary migration velocity vector of each monitoring point is defined. for:

[0027] ;

[0028] represent the instantaneous migration components of the grain boundary along the X, Y, and Z directions, respectively;

[0029] Based on the local grain boundary migration velocity vector, the local grain boundary mobility tensor is constructed , defined as: ; This tensor reflects the gradient change of local grain boundary migration velocity in different spatial directions;

[0030] Calculate the symmetry deviation index of the local mobility tensor for the entire preform area , defined as the ratio of the sum of squares of the off-diagonal elements of the tensor to the sum of squares of the diagonal elements at each point, and is calculated as:

[0031] ; They are all direction indexes, taken from the X, Y, and Z directions respectively, representing the three coordinate directions of space. is the partial derivative of the grain boundary migration velocity in the i-th direction with respect to the j-th direction, indicating the gradient change of the grain boundary migration velocity with the spatial position. It represents the sum of squares of all non-diagonal elements, reflecting the degree of change of grain boundary migration rate in non-indigenous space and the discreteness of migration direction. It represents the sum of squares of all diagonal elements, reflecting the intensity of change in grain boundary migration rate along its own direction and the degree of migration direction. The symmetry deviation index is used to reflect the consistency of the direction of local grain boundary migration. The larger the value, the more chaotic the migration direction, and the smaller the value, the more isotropic the migration.

[0032] In the entire monitoring area and time period, the symmetry deviation index of all local points is spatially and temporally weighted averaged, and the grain boundary migration tensor index GBMTI is finally defined as:

[0033] ; Represents the volume of the entire monitoring area, It represents the total length of observation time. The larger the GBMTI value, the more uneven the grain boundary migration process is and the stronger the anisotropy is, which potentially leads to a higher risk of abnormal growth of the grain size and particle size distribution of the coarse-grained layer.

[0034] In a preferred embodiment, after completing the calculation of the diffusion coupling unevenness index and the grain boundary migration tensor index, the diffusion coupling unevenness index and the grain boundary migration tensor index are synchronously input into the pre-trained machine learning model in the form of feature input vectors. The machine learning model is trained based on the historical coarse-grained layer tissue evolution data set, and can infer the coarse-grained layer grain size anomaly risk level under the current preparation state based on the input diffusion behavior spatial discreteness characteristics and grain boundary migration directional discreteness characteristics. The output coarse-grained layer grain size anomaly risk level is a multi-level classification result or a continuous numerical regression result. The higher the coarse-grained layer grain size anomaly risk level, the greater the potential probability of abnormal growth or uneven growth of the coarse-grained layer tissue. The machine learning model is obtained by training based on regression analysis, classification tree, support vector machine or deep neural network.

[0035] In a preferred embodiment, according to the abnormal risk level of the coarse-grained layer grain size output by the machine learning model, the abnormal risk level is divided into three risk levels: low, medium and high, and the corresponding preset control operations are triggered according to different risk levels. When the abnormal risk level is low, the basic fine-tuning operation is performed, and the basic fine-tuning operation is used to adjust the sintering heating rate curve, the cooling curve change rate and the local heating power output parameters to achieve a slight optimization of the local thermal field; when the abnormal risk level is medium, the strategy switching operation is performed, and the strategy switching operation is used to dynamically replace the current process control strategy, introduce a slow-rise section processing in the heating stage and a multi-section buffer strategy in the cooling stage to weaken the stress concentration of the tissue; when the abnormal risk level is high, the nonlinear correction operation is performed, and the nonlinear correction operation is used to adaptively adjust the cooling rate curve according to the local abnormal expansion trend and extend the constant temperature time of the highest temperature platform section to maximize the suppression of the abnormal evolution trend of the grain size and particle size distribution of the coarse-grained layer, and improve the stability and consistency of the microstructure of the coarse-grained layer of the NdFeB magnet.

[0036] Technical effects and advantages of the present invention:

[0037] The present invention adopts a multi-stage pressure forming and densification treatment method, combined with the synergistic effect of low-pressure pre-pressing, continuous pressurization loading and high-frequency micro-vibration, to effectively improve the overall density and density uniformity of the NdFeB powder preform before sintering, significantly reduce the local porosity and particle agglomeration phenomenon, and suppress the abnormal growth trend of the coarse-grained layer caused by uneven heat diffusion and stress concentration in the subsequent sintering process from the source, improve the consistency of the coarse-grained layer tissue evolution, and provide a highly homogeneous physical basis for dynamic monitoring and precise regulation during the sintering process.

[0038] During the sintering heating and insulation stages, the present invention continuously collects local temperature change rate and local stress change rate data in real time by pre-arranging multiple temperature sensors and strain sensors, and extracts characteristic data groups including local diffusion rate estimates and local grain boundary migration rate estimates based on the super-threshold trigger mechanism. It can realize dynamic perception and high-timeliness early warning in the early stage of abnormal evolution of coarse-grained layer, establishes a real-time monitoring and abnormality capture mechanism for the evolution process of coarse-grained layer organization, and significantly improves the response speed and prediction accuracy to the abnormal growth trend of coarse-grained layer grain size.

[0039] The present invention calculates the diffusion coupling unevenness index and the grain boundary migration tensor index based on the characteristic data group, and feeds them as input to the pre-trained machine learning model synchronously, so as to infer the risk level of the abnormal grain size of the coarse-grained layer in real time, and further triggers basic fine-tuning operations, strategy switching operations or nonlinear correction operations according to different risk levels. It can realize dynamic adjustment of the sintering heating rate, cooling curve change rate and local heating power output parameters, effectively smooth out local thermodynamic field anomalies, accurately suppress the abnormal growth of coarse grains, and improve the consistency of the final microstructure of the NdFeB magnet and the stability of service performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0041] Figure 1 This is a schematic diagram of the method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnets in the present invention. DETAILED DESCRIPTION

[0042] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0043] Reference Figure 1 The following examples were obtained:

[0044] Example 1: The present invention addresses the problem of difficulty in controlling the grain size and particle size distribution of the coarse-grained layer during the sintering preparation process of NdFeB magnets, and proposes a dynamic control method based on intelligent predictive control. First, by adopting a multi-stage pressure forming and densification treatment method, combined with the synergistic effect of initial low-pressure pressing, continuous multiple pressurization and high-frequency micro-vibration, a powder preform with uniform density distribution and overall density within a preset range is prepared. In the subsequent sintering heating and insulation stages, the local temperature change data and local stress change data at each position are obtained in real time through multiple temperature sensors and strain sensors arranged inside and on the surface of the preform, and the local temperature change rate and the local stress change rate are used to determine whether the coarse-grained layer organization evolution evaluation is triggered, to ensure that potential anomalies can be captured in time in the key process sections, and to ensure the closed-loop integrity of the initial forming and sintering dynamic monitoring.

[0045] Neodymium iron boron (NdFeB) magnets are rare earth permanent magnets primarily composed of neodymium (Nd), iron (Fe), and boron (B). They possess an extremely high magnetic energy product and are widely used in motors, wind power generation, magnetic resonance imaging (MRI), acoustic equipment, and other fields. Due to their exceptional magnetic properties (such as high coercivity, high remanence, and high energy product), NdFeB magnets are known as the "King of Magnets." After evaluating the microstructural evolution of the triggered coarse-grained layer, the diffusion coupling heterogeneity index and grain boundary migration tensor index are calculated by extracting characteristic data sets including the local temperature change rate, local stress change rate, local diffusion rate estimates, and local grain boundary migration rate estimates. The diffusion coupling heterogeneity index is derived from the normalization of the local diffusion rate field and the integration of spatial statistical variance, while the grain boundary migration tensor index is calculated based on the cumulative symmetry deviation indicators of the local mobility tensor. These two indices are synchronously input into the pre-trained machine learning model as feature input vectors. The machine learning model is trained based on historical coarse-grained layer microstructure evolution data. It can infer the risk level of coarse-grained layer grain size anomaly under the current preparation state based on the spatial discreteness characteristics of diffusion behavior and the directional discreteness characteristics of grain boundary migration, thereby establishing a dynamic evaluation and prediction feedback mechanism.

[0046] According to the abnormal risk level of the grain size of the coarse-grained layer output by the machine learning model, the abnormal risk level is divided into three risk levels: low, medium and high, and the corresponding preset control operations are triggered according to different risk levels. When the abnormal risk level is low, basic fine-tuning operations are performed to make fine adjustments within the proportional range for the heating rate curve, cooling change rate and local heating power output parameters; when the abnormal risk level is medium, a strategy switching operation is performed to introduce a slow-rise section treatment in the heating stage and a multi-section buffer strategy in the cooling stage to dynamically optimize the process path; when the abnormal risk level is high, a nonlinear correction operation is performed to adaptively adjust the cooling rate curve based on the local abnormal expansion trend and extend the constant temperature time of the highest temperature platform section, thereby maximally suppressing the abnormal growth of grains in the coarse-grained layer and the discrete expansion of grain size, and ensuring the overall stability and consistency of the microstructure of the coarse-grained layer of the NdFeB magnet. Specifically including the following steps:

[0047] In the first step, a multi-stage pressure forming and densification treatment method is used to prepare a powder preform with uniform density distribution and an overall density within a preset range. The purpose of using a multi-stage pressure forming and densification treatment method is to significantly improve the internal density and density uniformity of the powder preform by optimizing the arrangement and contact state between the powder particles before sintering, and reduce the internal porosity, thereby laying a homogeneous foundation for the organizational evolution of the coarse-grained layer during the subsequent sintering process. The initial continuous structure is formed by low-pressure pre-pressing, and then through multiple incremental pressure loading and synchronous application of high-frequency micro-vibration, it can effectively promote the microscopic rearrangement of the powder, fill small pores, avoid the formation of local high-porosity areas or particle agglomeration areas, and make the powder preform as a whole reach the set density range. This stage ensures that under the subsequent temperature field and stress field, the internal diffusion process and grain boundary migration process of the material have high consistency and controllability, which directly affects the organizational stability and magnetic property consistency of the finished magnet after sintering.

[0048] In the second step, during the sintering heating and holding stages, multiple temperature sensors and strain sensors pre-installed inside and on the surface of the preform are used to obtain real-time local temperature change data and local stress change data at each location; during the sintering heating and holding stages, multiple temperature sensors and strain sensors are installed inside and on the surface of the powder preform to obtain real-time local temperature change data and local stress change data, aiming to build a dynamic monitoring network for the internal thermodynamic state of the material. By continuously collecting the changing trends of local temperature and stress in real time, it is possible to timely capture potential uneven diffusion phenomena, local thermal gradient accumulation effects, and areas of tissue stress concentration during the sintering process, providing a direct quantitative basis for the subsequent risk assessment of coarse-grained layer evolution. The significance of this step is to transform the traditional passive post-detection method into active real-time monitoring, thereby improving the predictability and intervention of the process for the change trend of the coarse-grained layer organization, ensuring that the response mechanism can be activated in time at key process nodes to avoid the irreversible phenomenon of abnormal coarse-grained growth.

[0049] The third step is to determine whether the coarse-grained layer microstructure evolution assessment is triggered based on the real-time monitored local temperature change rate and local stress change rate. When the local temperature change rate or local stress change rate exceeds the preset threshold condition, a feature data group including the local temperature change rate, local stress change rate, local diffusion rate estimate, and local grain boundary migration rate estimate is extracted. Based on the real-time monitored local temperature change rate and local stress change rate, it is determined whether the coarse-grained layer microstructure evolution assessment is triggered, and a feature data group including the local temperature change rate, local stress change rate, local diffusion rate estimate, and local grain boundary migration rate estimate is extracted when the abnormal threshold is triggered. The purpose is to accurately lock in the precursor characteristics of the coarse-grained layer microstructure change in the early stage of potential abnormality development in a data-driven manner. Through the over-limit detection of temperature change rate and stress change rate, risk sources such as uneven thermal diffusion and local stress anomalies during the sintering process can be effectively identified. Subsequently, key features that comprehensively reflect the diffusion dynamics and grain boundary migration behavior can be extracted to provide complete and timely data input for subsequent intelligent prediction models, ensuring that the coarse-grained layer evolution assessment is not only real-time, but also relevant to physical mechanisms, thereby improving the scientificity and accuracy of overall prediction and regulation.

[0050] The fourth step is to calculate the diffusion coupling heterogeneity index and the grain boundary migration tensor index based on the feature data set. These are fed into a pre-trained machine learning model as input, outputting a risk level for grain size anomaly in the coarse-grained layer. The sintering process is then dynamically adjusted based on the risk level. This approach aims to establish a real-time adaptive, data-driven dynamic optimization control loop. The diffusion coupling heterogeneity index reflects the spatial dispersion of the local diffusion process, while the grain boundary migration tensor index quantitatively characterizes the directional heterogeneity of grain boundary migration. Together, they reflect the microscopic dynamics of the coarse-grained layer evolution. By predicting the risk level of complex nonlinear behavior using a machine learning model, dynamic optimization of the sintering heating curve, local heating power, and cooling curve can be achieved at the initial stage of the coarse-grained layer evolution trend, minimizing the growth of abnormal grain size in the coarse-grained layer and improving the consistency and reliability of the final magnet performance.

[0051] The multi-stage pressure forming and densification treatment method includes the following steps: first, the NdFeB powder raw material is preliminarily pressed under low pressure conditions to form a preliminary preform with a continuous structure and an internal porosity lower than the initial powder state; then, on the basis of applying uniform pressurization as a whole, multiple pressure loading processes are continuously superimposed, and the amplitude and duration of each loading are gradually increased according to the preset pressure increase curve, and the direction of the pressure applied during the pressurization process is kept stable and consistent; during the entire pressurization process, high-frequency micro-vibration is applied synchronously.

[0052] High-frequency micro-vibration is continuously applied throughout the entire multi-stage pressure forming and densification process. The frequency of high-frequency micro-vibration is maintained in the range of 1,000 Hz to 100,000 Hz, and the direction of applying high-frequency micro-vibration is set at a preset fixed angle to the pressurization direction. Continuous vibration energy input triggers local microscopic displacement and rotation adjustment of powder particles, enhances the contact probability and bonding strength between particles, reduces local internal porosity, and prevents powder particles from agglomerating or forming density-uneven aggregation areas during unidirectional pressurization. While applying high-frequency micro-vibration, the pressure direction of the overall pressurization process is maintained stable to ensure that the direction of particle rearrangement is consistent with the pressing direction, thereby promoting the high uniformity of the internal microstructure of the pressed powder preform, thereby laying the foundation for subsequent temperature and stress monitoring and dynamic regulation of the grain size of the coarse-grained layer.

[0053] The multi-stage pressure forming and densification process first involves preliminary pressing of the NdFeB powder raw material under low-pressure conditions. Low-pressure conditions refer to a pressing pressure less than 20% to 50% of the powder particles' yield strength, typically controlled within the range of 50 MPa to 120 MPa, to prevent powder particle breakage and promote the initial formation of a continuous, connected structure. NdFeB powder raw material refers to a composite metal magnetic powder material containing neodymium, iron, boron, and rare earth elements with a specific particle size distribution. The purpose of this initial pressing step is to create initial contact and overlap between the powder particles through uniform external pressure without destroying the integrity of the particles, thereby forming a continuous three-dimensional skeleton structure. This also significantly reduces the internal porosity of the preliminary preform to a value significantly lower than that of the initial loose powder state, typically by more than 10%, providing a good foundation for subsequent densification.

[0054] Subsequently, based on the overall uniform pressurization, multiple pressure loading processes are continuously superimposed, with the amplitude and duration of each loading stepwise increasing according to a preset pressure ramp curve. Uniform pressurization refers to the uniform distribution of the pressing force applied across the entire preform surface in all directions, avoiding localized high or low pressure areas. The sequential superposition of multiple pressure loading processes involves gradually increasing the applied pressure in stages after the initial pressing process by controlling the press or pressure-applying device. For example, starting with the initial pressing pressure, the pressure is increased by 10 to 20 percent each time, with two to four successive pressurization steps, each lasting between 10 and 30 seconds. The preset pressure ramp curve is a reference trajectory of pressure variation over time, pre-set based on the material's forming characteristics. It typically exhibits an increasing trend to achieve progressive densification while avoiding particle breakage or structural inhomogeneity caused by a single high pressure application. Throughout the multiple pressurization processes, the applied pressure direction remains stable and consistent, meaning that the direction of the pressing force remains constant in the spatial coordinate system and does not shift. This ensures that the powder particles rearrange and deform in a fixed direction during the pressing process, thereby promoting an overall uniform density distribution.

[0055] During the entire pressurization process, high-frequency micro-vibration is applied synchronously. High-frequency micro-vibration refers to the vibration frequency controlled in the range of 1,000 Hz to 100,000 Hz, and the vibration amplitude is very small, usually in the micron level. The way to apply high-frequency micro-vibration is to transfer high-frequency vibration energy to the interior of the powder preform through a mechanical vibration source while pressurizing. The direction of high-frequency micro-vibration is set at a preset fixed angle with the pressurization direction. The angle is generally controlled in the range of 15 to 45 degrees to induce local microscopic displacement, rotation and rearrangement of the powder particles. In this way, the contact probability and interface reconstruction ability between the powder particles are significantly improved, and at the same time, the local internal porosity can be reduced, effectively preventing the occurrence of particle agglomeration or local density unevenness of the powder during the unidirectional pressurization process, thereby further improving the consistency of the overall density.

[0056] For example, in a specific implementation, the following process parameters can be used: First, NdFeB powder with a particle size ranging from 3 to 50 microns is initially pressed at a pressure of 50 MPa to form a preliminary preform; then, the preform is loaded three times, with each loading time of 20 seconds, at a pressure increase of 15 percent, until the final pressure reaches approximately 180 MPa. During the pressurization process, a high-frequency micro-vibration source with a vibration frequency of approximately 5 kHz and an angle of 30 degrees is continuously applied to the preform. After this series of treatments, a powder preform with a density distribution standard deviation of less than 3 percent and an overall density of over 95 percent is obtained, providing a homogeneous and reliable physical basis for monitoring and controlling the evolution of the coarse-grained layer structure during the subsequent sintering heating, insulation, and cooling processes.

[0057] In the process of determining whether to trigger the coarse-grained layer microstructure evolution assessment based on the real-time monitoring of the local temperature change rate and the local stress change rate, the changing trends of the local temperature change rate and the local stress change rate are continuously monitored. When any local temperature change rate or local stress change rate exceeds the corresponding preset threshold condition, the coarse-grained layer microstructure evolution assessment process is immediately triggered;

[0058] The preset threshold conditions are based on process settings, and the static calibration values ​​are set in different ranges according to different sintering stages to ensure that potential abnormal changes are captured in real time during key process sections, avoiding abnormal growth of grain size and particle size distribution in the coarse-grained layer due to delayed response, and improving the timeliness and accuracy of the trigger mechanism.

[0059] During the evaluation process of determining whether the evolution of the coarse-grained layer microstructure has been triggered based on the real-time monitoring of the local temperature change rate and the local stress change rate, the changing trends of the local temperature change rate and the local stress change rate are continuously monitored. The local temperature change rate refers to the rate of change of temperature per unit time at a designated monitoring location inside or on the surface of the NdFeB powder preform, usually measured in degrees Celsius per second. The local stress change rate refers to the rate of change of stress within the material per unit time at the corresponding monitoring location, usually measured in megapascals per second. Continuous monitoring refers to the real-time recording of the dynamic changes in the local temperature change rate and the local stress change rate through high-frequency, uninterrupted data acquisition throughout the sintering heating and holding stages to ensure that small fluctuations or abnormal trends in the internal thermodynamic state of the material can be captured in a timely manner.

[0060] When any local temperature change rate or local stress change rate exceeds the corresponding preset threshold conditions, the coarse-grained layer microstructure evolution assessment process is immediately triggered. The preset threshold conditions refer to the critical change rate values ​​set for different monitoring locations and different stages based on the predetermined process flow and material properties, which serve as the trigger threshold for determining whether there is a potential abnormal evolution trend of the coarse-grained layer. Exceeding the preset threshold conditions means that abnormal diffusion rate, abnormal grain boundary migration rate, or thermal stress concentration effects may occur in local areas of the material, necessitating a subsequent immediate coarse-grained layer microstructure evolution risk assessment.

[0061] The preset threshold conditions are based on process settings. Specifically, they are determined based on process parameters such as the NdFeB magnet's sintering heating rate, holding temperature platform, and cooling rate curve, combined with powder particle size, initial density distribution, and material thermophysical properties (such as thermal diffusivity, thermal expansion coefficient, and yield strength). Static calibration values ​​refer to reasonable ranges for local temperature and stress change rates set at different sintering stages, verified through small-batch experiments before sintering. For example, a higher temperature change rate threshold is set during the initial heating stage to accommodate rapid overall heating, while a lower temperature change rate threshold is set during the high-temperature plateau stage to accurately monitor local thermal stability.

[0062] By setting preset threshold conditions in different ranges at different sintering stages, the sensitivity of anomaly capture can be dynamically adjusted based on the different microstructural evolution characteristics at each stage of the sintering process. For example, in actual applications, the local temperature change rate threshold can be set to five degrees Celsius per second and the local stress change rate threshold to one megapascal per second during the initial sintering temperature rise phase; while at the highest temperature plateau phase, the local temperature change rate threshold is lowered to one degree Celsius per second and the local stress change rate threshold is lowered to 0.5 megapascals per second. This setting ensures that potential abnormal changes are captured in real time during critical process sections, avoiding irreversible abnormal growth of the grain size and particle size distribution of the coarse-grained layer due to response lag.

[0063] Through the combination of the above-mentioned continuous monitoring, dynamic judgment and staged threshold setting, the present invention can greatly improve the timeliness and accuracy of the trigger mechanism, realize accurate early warning and intervention control of the evolution of the coarse-grained layer structure during the sintering process, and ensure that the final NdFeB magnet product has higher consistency and reliability.

[0064] In the specific implementation process, for example, when preparing a batch of NdFeB powder preforms with a particle size distribution ranging from three microns to fifty microns, they are first preliminarily pressed under a low pressure of fifty MPa to form a preliminary continuous structure, with the porosity reduced by about fifteen percent compared to the original loose powder. Subsequently, under the guidance of a preset pressure increase curve, three consecutive pressurizations are applied, with each pressurization increasing by twenty percent, and each pressurization lasting twenty seconds, until a final pressure of one hundred and eighty MPa is reached. During this pressurization process, the direction of the applied pressure is kept stable, and high-frequency micro-vibration with a frequency of approximately five kilohertz and a thirty-degree angle to the pressurization direction is simultaneously applied, with the vibration amplitude controlled within a few microns. Through this multi-stage pressure forming and densification treatment method, a highly uniform powder preform with a density distribution standard deviation of less than three percent and an overall density exceeding ninety-five percent is obtained.

[0065] During the sintering heating and holding stages, a total of twenty temperature sensors and strain sensors are installed internally and on the surface to collect continuous data once per second. The local temperature change rate threshold is set at five degrees Celsius per second in the initial heating stage, and the local stress change rate threshold is set at one megapascal per second. During the high-temperature platform stage, the local temperature change rate threshold is set at one degree Celsius per second, and the local stress change rate threshold is set at 0.5 megapascals per second. When the local temperature change rate in a certain area exceeds one degree Celsius per second in the high-temperature platform stage, or the local stress change rate exceeds 0.5 megapascals per second, the coarse-grained layer microstructure evolution assessment process is immediately triggered, and real-time feature data sets for that location and its surrounding areas are extracted, providing basic data support for subsequent prediction of coarse-grained layer abnormality risks and dynamic regulation.

[0066] After triggering the coarse-grained layer microstructure evolution assessment, a feature data set is extracted, which consists of a local temperature change rate, a local stress change rate, an estimated local diffusion rate, and an estimated local grain boundary migration rate;

[0067] The local diffusion rate estimate is obtained by obtaining local real-time temperature data and calculating it according to the diffusion rate calculation formula. The diffusion rate calculation formula is that the local diffusion rate is equal to the initial diffusion rate multiplied by an exponential function. The negative exponential term of the exponential function is obtained by dividing the diffusion activation energy by the product of the local real-time temperature and the gas constant. The local diffusion rate is calculated according to the following formula:

[0068] ;

[0069] Indicates that the monitoring point is at position Place, time The diffusion rate when is the reference initial diffusion rate constant, is the material diffusion activation energy, is the gas constant, take 8.314, Indicates that the monitoring point is at position Place, time Real-time temperature at the time;

[0070] The estimated value of the local grain boundary migration rate is obtained by multiplying the stress gradient measured by the local stress change rate by the linear response coefficient of the grain boundary migration driving force. The linear response coefficient of the grain boundary migration driving force is preset based on the internal grain boundary energy and grain size calibration data of the NdFeB magnet material;

[0071] Each data item in the feature data group is synchronously paired according to the timestamp.

[0072] After triggering the coarse-grained layer microstructure evolution assessment, a feature data set is extracted. The feature data set consists of the local temperature change rate, the local stress change rate, an estimated local diffusion rate, and an estimated local grain boundary migration rate. The local temperature change rate refers to the actual rate of temperature change per unit time at a specified monitoring location, typically measured in degrees Celsius per second, and is used to characterize local heating or cooling trends during the sintering process. The local stress change rate refers to the actual rate of stress change per unit time at the same location, typically measured in megapascals per second, and is used to reflect the evolution of local mechanical strain during the sintering process.

[0073] The local diffusion rate estimate is obtained by obtaining local real-time temperature data and calculating it according to the diffusion rate calculation formula. Local real-time temperature data refers to the actual temperature value of the current monitoring point collected in real time by the temperature sensor, in Kelvin. The diffusion rate calculation formula is that the local diffusion rate is equal to the reference initial diffusion rate constant multiplied by an exponential function, and the negative exponential term of the exponential function is obtained by dividing the material diffusion activation energy by the product of the local real-time temperature and the gas constant. The reference initial diffusion rate constant is the standard diffusion rate of the material under reference conditions, usually expressed in square meters per second; the material diffusion activation energy refers to the energy required for particles to overcome energy barriers during the diffusion process, in joules per mole; the gas constant is taken as 8.314 joules per mole per Kelvin, which is a universal natural constant.

[0074] The local diffusion rate is derived using the following formula: The local diffusion rate is equal to the reference initial diffusion rate constant multiplied by an exponential function, where the negative exponential term is the material's activation energy for diffusion divided by the product of the gas constant and the local real-time temperature. This formula reflects the exponential relationship in which the diffusion process is highly sensitive to temperature: higher temperatures result in faster diffusion rates.

[0075] In specific implementations, for example, a batch of preforms made from NdFeB powder with particle sizes ranging from three to fifty microns is subjected to internal temperature sensors, which collect real-time temperature data at different spatial locations once per second. By reading the real-time temperature at each monitoring point and substituting the known reference initial diffusion rate constant of one times ten to the power of negative eighth square meters per second, the material diffusion activation energy of 180 kilojoules per mole, and the gas constant of 8.314 joules per mole per Kelvin, an estimated local diffusion rate at each monitoring point is calculated in real time according to the above derivation formula. This estimated value, as an important component of the characteristic data set, provides the fundamental physical quantity support for the subsequent calculation of the diffusion coupling heterogeneity index and the prediction of the risk level of coarse-grained layer evolution.

[0076] The diffusion coupling heterogeneity index is derived based on the discrete spatiotemporal distribution of the local diffusion rate estimates, which includes the following steps:

[0077] At each moment, the set of local diffusion rate estimates of all monitoring points in space is standardized. The local diffusion rate estimate refers to the estimated amount of the diffusion capacity of material particles at different positions in the sintering process, which is inferred from the temperature monitoring data, usually in square meters per second. The standardization method is to subtract the minimum diffusion rate value among all monitoring points at the same moment from each local diffusion rate, and then divide it by the diffusion rate range of all monitoring points at that moment, that is, the difference between the maximum diffusion rate value and the minimum diffusion rate value, thereby forming a dimensionless normalized diffusion rate field. The dimensionless normalized diffusion rate field means that the diffusion rate data of all monitoring positions are normalized to a unified interval of zero to one, eliminating the influence of different absolute numerical magnitudes, so that subsequent statistical analysis only focuses on the spatial distribution characteristics of the diffusion behavior, and is not affected by the absolute value.

[0078] Based on the normalized diffusion rate field, the spatial statistical variance of the standardized diffusion rate at that moment is calculated. Spatial statistical variance quantifies the degree of dispersion in the spatial distribution of the normalized diffusion rate. It is typically calculated by summing the squared differences between the normalized diffusion rate values ​​at all monitoring points and their average value and taking the average. Larger spatial statistical variance values ​​indicate more dramatic spatial fluctuations in the local diffusion rate, meaning a more uneven diffusion rate distribution, reflecting increasing inconsistencies in diffusion activity at different locations during the sintering process.

[0079] Along the time dimension, the spatial statistical variance corresponding to each moment is cumulatively integrated or discrete-time averaged. Cumulative integration refers to the continuous integration of the spatial statistical variance values ​​at all time points throughout the sintering process or a selected stage; discrete-time averaging refers to the average of the spatial statistical variances at each discrete sampling moment. In this way, a time-weighted cumulative index of spatial diffusion heterogeneity is obtained, which not only reflects spatial discreteness but also comprehensively considers the changing trend of discreteness in the time dimension, forming a comprehensive evaluation of the consistency of diffusion behavior throughout the sintering or cooling cycle.

[0080] The diffusion coupling heterogeneity index is defined as the normalized value of the cumulative index of spatial diffusion heterogeneity. The normalization process is to ensure that the index is comparable and uniform in different sintering batches and under different process parameters. Ultimately, the diffusion coupling heterogeneity index quantifies the overall spatial discreteness level of the local diffusion process during the entire sintering cycle or cooling cycle. The higher the value of the diffusion coupling heterogeneity index, the worse the consistency of the local diffusion behavior during the evolution of the coarse-grained layer structure, which means that it is more likely to have undesirable phenomena such as abnormal growth of the coarse-grained layer and uneven particle size distribution, which suggests that a more stringent dynamic control strategy needs to be implemented in the future.

[0081] In the specific implementation process, for example, after a batch of preliminary preforms are prepared, the sintering heating stage is carried out. It is assumed that at a certain moment, the local real-time temperature data of twenty monitoring points are collected by sensors, and the corresponding local diffusion rate estimates are obtained by calculation. After standardization, the normalized diffusion rate value of each monitoring point is obtained, and its spatial statistical variance is further calculated. By cumulatively integrating the spatial statistical variance of the sampling points once per second during the entire sintering heating stage, the spatial diffusion heterogeneity cumulative index is finally obtained, and normalized to obtain the diffusion coupling unevenness index. If the diffusion coupling unevenness index reaches above the preset high-risk threshold, it indicates that there is a large heterogeneity risk in the coarse-grained layer structure during the sintering process, and it is necessary to start the subsequent dynamic sintering temperature control adjustment mechanism for compensation optimization.

[0082] The grain boundary migration tensor index is used to quantify the anisotropic distribution characteristics of the grain boundary migration rate within the coarse-grained layer of NdFeB magnets on a spatial scale, so as to evaluate the consistency of grain boundary movement during the evolution of the coarse-grained layer. The calculation process includes the following steps:

[0083] Based on the estimated values ​​of local grain boundary migration rate collected at different monitoring positions, the local grain boundary migration velocity vector field is constructed, and the grain boundary migration velocity vector of each monitoring point is defined. for:

[0084] ;

[0085] represent the instantaneous migration components of the grain boundary along the X, Y, and Z directions, respectively;

[0086] Based on the local grain boundary migration velocity vector, the local grain boundary mobility tensor is constructed , defined as: ; This tensor reflects the gradient change of local grain boundary migration velocity in different spatial directions;

[0087] Calculate the symmetry deviation index of the local mobility tensor for the entire preform area , defined as the ratio of the sum of squares of the off-diagonal elements of the tensor to the sum of squares of the diagonal elements at each point, and is calculated as:

[0088] ; They are all direction indexes, taken from the X, Y, and Z directions respectively, representing the three coordinate directions of space. is the partial derivative of the grain boundary migration velocity in the i-th direction with respect to the j-th direction, indicating the gradient change of the grain boundary migration velocity with the spatial position. It represents the sum of squares of all non-diagonal elements, reflecting the degree of change of grain boundary migration rate in non-indigenous space and the discreteness of migration direction. It represents the sum of squares of all diagonal elements, reflecting the intensity of change in grain boundary migration rate along its own direction and the degree of migration direction. The symmetry deviation index is used to reflect the consistency of the direction of local grain boundary migration. The larger the value, the more chaotic the migration direction, and the smaller the value, the more isotropic the migration.

[0089] In the entire monitoring area and time period, the symmetry deviation index of all local points is spatially and temporally weighted averaged, and the grain boundary migration tensor index GBMTI is finally defined as:

[0090] ; Represents the volume of the entire monitoring area, It represents the total length of observation time. The larger the GBMTI value, the more uneven the grain boundary migration process is and the stronger the anisotropy is, which potentially leads to a higher risk of abnormal growth of the grain size and particle size distribution of the coarse-grained layer.

[0091] The grain boundary migration tensor index is used to quantify the anisotropic distribution characteristics of the grain boundary migration rate within the coarse-grained layer of NdFeB magnets on a spatial scale, so as to evaluate the consistency of grain boundary movement during the evolution of the coarse-grained layer. The calculation process of the grain boundary migration tensor index includes the following steps:

[0092] Based on the estimated local grain boundary migration rates collected at different monitoring locations, a local grain boundary migration velocity vector field is constructed. The estimated local grain boundary migration rate refers to the instantaneous movement rate of the grain boundary in all directions obtained by inferring the local stress change rate during sintering or cooling, usually measured in meters per second. The local grain boundary migration velocity vector field refers to the definition of the grain boundary migration velocity vector at each monitoring point in space, which represents the instantaneous direction and amplitude changes of the grain boundary movement. The grain boundary migration velocity vector of each monitoring point is defined as the combination of the components along the three coordinate directions of the grain boundary, including the component along the X direction of space, the component along the Y direction of space, and the component along the Z direction of space. Each component uses the position coordinate and time as a function, representing the instantaneous movement rate of the grain boundary at different spatial positions and different time points.

[0093] Based on the local grain boundary migration velocity vector, a local grain boundary mobility tensor is constructed. The local grain boundary mobility tensor refers to the gradient variation of the grain boundary migration velocity in different spatial directions, expressed as a tensor matrix. Specifically, it is defined as taking the partial derivatives of each component of the grain boundary migration velocity with respect to each spatial coordinate in the X, Y, and Z spatial directions, and arranging them to form a three-by-three matrix. This tensor reflects the gradient of the local grain boundary migration velocity in different spatial directions, reflecting the anisotropic characteristics of the local grain boundary migration process.

[0094] The symmetry deviation index of the local mobility tensor is calculated for the entire preform region. This symmetry deviation index is the ratio of the sum of squares of the off-diagonal elements to the sum of squares of the diagonal elements in the mobility tensor at each monitoring point. It is used to quantify the degree of deviation in the consistency of motion between directions during grain boundary migration. The specific calculation method is: calculate the sum of squares of all off-diagonal partial derivatives, then calculate the sum of squares of all diagonal partial derivatives, and take the ratio of the two. The direction indices i, j, and k are taken from the X, Y, and Z directions, respectively, representing the three orthogonal coordinate axes of three-dimensional space.

[0095] In the above definition, the partial derivative of the grain boundary migration velocity in the i-th direction with respect to the j-th spatial coordinate represents the gradient of the grain boundary migration velocity with respect to spatial position. The sum of the squares of all off-diagonal terms reflects the degree of variation in the grain boundary migration velocity in non-orientational directions, reflecting the high dispersion of the migration direction. The sum of the squares of all diagonal terms reflects the intensity of variation in the grain boundary migration velocity along the oriented direction, reflecting the degree of isotropy of the migration direction. The symmetry deviation index is used to reflect the degree of consistency in the directionality of local grain boundary migration. Larger values ​​indicate more chaotic grain boundary migration directions, while smaller values ​​indicate more isotropic migration.

[0096] Within the entire monitoring area and time period, the symmetry deviation indicators of all local points are spatially and temporally weighted averaged to ultimately define the grain boundary migration tensor index. Spatial weighting refers to the integration or summation of local indicators at different spatial positions, and temporal weighting refers to the integration or averaging of indicators at each moment within the entire observation time period. The final grain boundary migration tensor index is defined as the weighted integral average of the local symmetry deviation indicators per unit observation volume and observation time length. The larger the value of the grain boundary migration tensor index, the more uneven the changes in all directions during the grain boundary migration process, the stronger the anisotropy of the local microstructure evolution, and the higher the risk of potentially causing abnormal growth in the grain size and particle size distribution of the coarse-grained layer.

[0097] In the specific implementation process, for example, twenty spatial monitoring points are set during the sintering process, and the local stress change rate is measured in real time to infer the local grain boundary migration rate. Further, based on the gradient changes of the velocity components in different directions, a local mobility tensor matrix is ​​constructed. Sampling is performed once every second, and the local symmetry deviation index of each monitoring point is calculated. Then, a weighted average is performed along the time axis and spatial area to finally obtain the grain boundary migration tensor index value. If the grain boundary migration tensor index reaches a preset high-risk threshold value or above in the middle and late stages of sintering, it indicates that the directionality of the grain boundary migration process in the coarse-grained layer is out of control, and dynamic temperature control and cooling adjustment measures need to be initiated to prevent abnormal coarse grain growth and uneven expansion of grain size.

[0098] After completing the calculation of the diffusion coupling unevenness index and the grain boundary migration tensor index, the diffusion coupling unevenness index and the grain boundary migration tensor index are synchronously input into the pre-trained machine learning model in the form of feature input vectors. The machine learning model is trained based on the historical coarse-grained layer tissue evolution data set. It can infer the risk level of the coarse-grained layer grain size anomaly under the current preparation state based on the input diffusion behavior spatial discreteness characteristics and grain boundary migration directional discreteness characteristics. The output coarse-grained layer grain size anomaly risk level is a multi-level classification result or a continuous numerical regression result. The higher the coarse-grained layer grain size anomaly risk level, the greater the potential probability of abnormal growth or uneven growth of the coarse-grained layer tissue. The machine learning model is obtained by training based on regression analysis, classification tree, support vector machine or deep neural network.

[0099] After calculating the diffusion coupling heterogeneity index and the grain boundary migration tensor index, they are simultaneously input into the pre-trained machine learning model as feature input vectors. The diffusion coupling heterogeneity index quantifies the spatial discreteness of the local diffusion process within the coarse-grained layer, while the grain boundary migration tensor index quantifies the anisotropy of grain boundary migration within the coarse-grained layer. The feature input vector combines these two quantitative indicators with different physical meanings into a unified data input format for processing and inference by the machine learning model.

[0100] The machine learning model is trained on a historical dataset of the evolution of the coarse-grained layer structure. This dataset refers to long-term monitoring data collected from different batches and under different process conditions related to the evolution of the coarse-grained layer structure. This includes information such as local diffusion rate, local grain boundary migration rate, sintering temperature curve, cooling temperature variation, and the final coarse-grained layer grain size and particle size distribution. By leveraging this rich data sample, the machine learning model can learn the nonlinear mapping relationship between input features and output results.

[0101] The machine learning model can infer the risk level of abnormal grain size in the coarse-grained layer under the current preparation state based on the spatial discreteness characteristics of the input diffusion behavior and the directional discreteness characteristics of grain boundary migration. The risk level of abnormal grain size in the coarse-grained layer is a quantitative score of whether there is a risk of abnormal growth in the microstructure evolution of the coarse-grained layer during the current sintering process. It is usually classified as a multi-level classification result or a continuous numerical regression result. The higher the output risk level of abnormal grain size in the coarse-grained layer, the greater the potential probability of abnormal growth or uneven growth of the coarse-grained layer microstructure, indicating the need for timely process parameter adjustments.

[0102] Machine learning models can be trained using different types of methods:

[0103] Regression analysis refers to establishing a mathematical regression relationship between input variables and output continuous numerical values ​​by minimizing the sum of squared errors between input features and output risk levels. It is suitable for scenarios where the output is a continuous risk score.

[0104] A classification tree recursively divides the feature space into several subregions, outputting a class label within each subregion to form a tree-structured model. This model is suitable for classifying the risk level of coarse-grained layer anomalies into discrete levels. The classification tree training process is achieved by minimizing the impurity index (such as the Gini index or information entropy) after each node is partitioned.

[0105] Support vector machine refers to finding the optimal hyperplane in the feature space to separate samples of different risk levels with the maximum interval. It is suitable for training input data with small samples, high dimensions and certain nonlinear characteristics. The training process maximizes the classification boundary interval and introduces the kernel function for feature mapping.

[0106] A deep neural network refers to a system that uses a multi-layer nonlinear transformation structure to establish a highly nonlinear relationship between complex inputs and outputs through weighted connections between a large number of neuron nodes. It is suitable for processing large-scale data sets and can explore the deep impact of the complex interaction between diffusion behavior and grain boundary migration on the abnormal evolution of coarse-grained layers. The training process usually uses a backpropagation algorithm combined with gradient descent to optimize the loss function.

[0107] In specific implementations, for example, a deep neural network can be used as the machine learning model structure, with input nodes corresponding to the diffusion coupling heterogeneity index and the grain boundary migration tensor index. The number of hidden layers is set to three, with 64, 32, and 16 nodes per layer, respectively. The activation function uses the rectified linear unit function, and the loss function uses the mean square error loss. By training on one thousand sets of historical sintering sample data, it is ultimately possible to accurately predict the risk level of grain size anomalies in the coarse-grained layer. In actual production, sintering parameters can be dynamically adjusted based on the prediction results to achieve consistent control of the microstructural evolution of the coarse-grained layer.

[0108] According to the abnormal risk level of the coarse-grained layer grain size output by the machine learning model, the abnormal risk level is divided into three risk levels: low, medium and high, and the corresponding preset control operations are triggered according to different risk levels. Among them, when the abnormal risk level is low, the basic fine-tuning operation is performed. The basic fine-tuning operation is used to adjust the sintering heating rate curve, the cooling curve change rate and the local heating power output parameters to achieve a slight optimization of the local thermal field; when the abnormal risk level is medium, the strategy switching operation is performed. The strategy switching operation is used to dynamically replace the current process control strategy, introduce a slow rise section processing in the heating stage and a multi-section buffer strategy in the cooling stage to weaken the stress concentration of the tissue; when the abnormal risk level is high, the nonlinear correction operation is performed. The nonlinear correction operation is used to adaptively adjust the cooling rate curve according to the local abnormal expansion trend and extend the constant temperature time of the highest temperature platform section to maximize the suppression of the abnormal evolution trend of the grain size and particle size distribution of the coarse-grained layer, and improve the stability and consistency of the microstructure of the coarse-grained layer of the NdFeB magnet.

[0109] According to the abnormal risk level of grain size in the coarse-grained layer output by the machine learning model, the abnormal risk level is divided into three risk levels: low, medium and high, and the corresponding preset control operations are triggered according to different risk levels.

[0110] When the abnormal risk level is low, perform basic fine-tuning operations. Basic fine-tuning operations refer to the dynamic adjustment of the heating rate curve, the cooling curve change rate and the local heating power output parameters in a subtle range during the sintering process. The sintering heating rate curve refers to the set temperature increase rate trajectory that changes with time, the cooling curve change rate refers to the temperature change rate during the cooling process after sintering is completed, and the local heating power output parameter refers to the power intensity applied by the heater in different areas. The specific use of the basic fine-tuning operation is: when the abnormal risk level is detected to be low, the sintering heating rate is proportionally reduced, for example, by less than 5%, and at the same time, the cooling curve change rate is fine-tuned to adjust the cooling rate to increase or decrease by no more than 5%, and reduce or increase the heating power output by 3% to 5% in the local area, so as to slightly optimize the local thermal field uniformity, suppress the subtle uneven diffusion trend, and ensure the smooth evolution of the overall organization.

[0111] When the abnormal risk level is medium, a strategy switching operation is performed. The strategy switching operation refers to dynamically replacing the current process control strategy according to the real-time detection results during the sintering process, introducing a slow-rise segment treatment in the heating stage and a multi-segment buffer strategy in the cooling stage. The slow-rise segment treatment in the heating stage refers to inserting a transition section with a lower heating rate in the heating curve, such as reducing the original heating rate to 70% to 80% of the original rate, and slowly heating to reduce the risk of local thermal stress concentration. The multi-segment buffer strategy in the cooling stage refers to dividing the original continuous cooling curve into multiple small sections with different cooling rates, and the cooling rate between each section transitions smoothly. For example, slow cooling is used in the high to medium temperature section, and moderately accelerated cooling is used in the medium to low temperature section, forming a staged progressive cooling mode as a whole. Through the strategy switching operation, the internal stress concentration and uneven diffusion of the organization during the sintering process are effectively weakened, thereby reducing the moderate risk of abnormal evolution of the coarse-grained layer.

[0112] When the abnormal risk level is high, a nonlinear correction operation is performed. The nonlinear correction operation refers to the adaptive adjustment of the cooling rate curve and the extension of the constant temperature time of the highest temperature platform section according to the local abnormal expansion trend detected in real time, so as to minimize the abnormal evolution of the grain size and particle size distribution of the coarse grain layer. The adaptive adjustment of the cooling rate curve means that after the high-risk area is monitored, the original rapid cooling rate is adjusted to a gradual slow decline curve. The cooling rate reduction can reach 30% to 50% of the original set value, and the cooling speed is dynamically fine-tuned according to the local thermal field state. Extending the constant temperature time of the highest temperature platform section means setting an additional holding time in the highest temperature section, for example, adding 20% ​​to 50% of the constant temperature time on the basis of the original platform holding, so as to fully release the local thermal stress and diffusion driving force, smooth the evolution process of the coarse grain boundary, and suppress the abnormal growth trend of the coarse grains.

[0113] During the specific implementation process, for example, in the preparation process of a batch of NdFeB magnets, the machine learning model outputs the risk level of abnormal grain size in the coarse-grained layer in real time. If the risk level is detected to be at a low level at a certain stage, the sintering heating curve is automatically adjusted, the heating rate is reduced by three percent, and the heating power output is fine-tuned in the local area; if the risk level rises to a medium level, the strategy switch is triggered, and a slow-rise segment is inserted when the temperature reaches one thousand degrees Celsius, the heating rate is reduced to seventy-five percent of the original planned rate, and multi-stage buffer cooling is started when cooling to eight hundred degrees Celsius; if the risk level further rises to a high level, the holding time of the highest sintering temperature platform is extended by thirty percent, and the cooling curve is changed to a three-stage gradual decline, with the cooling rate in each stage gradually decreasing, effectively suppressing the abnormal growth of the coarse-grained layer and ensuring the uniform and stable structure of the final magnet.

[0114] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0115] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0116] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0117] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0118] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnets, characterized in that: The following steps are involved: In the first step, a multi-stage pressure forming and densification process is used to prepare a powder preform with uniform density distribution and an overall density within a preset range; In the second step, during the sintering heating and holding stages, multiple temperature sensors and strain sensors pre-installed inside and on the surface of the preform are used to obtain real-time data on local temperature changes and local stress changes at each location. The third step is to determine whether to trigger the evaluation of the coarse-grained layer microstructure evolution based on the real-time monitored local temperature change rate and local stress change rate. When the local temperature change rate or local stress change rate exceeds the preset threshold, a feature data set including the local temperature change rate, local stress change rate, local diffusion rate estimate, and local grain boundary migration rate estimate is extracted. In the fourth step, the diffusion coupling heterogeneity index and grain boundary migration tensor index are calculated based on the characteristic data set and fed into a pre-trained machine learning model to output the risk level of grain size anomaly in the coarse-grained layer. The sintering process is then dynamically adjusted based on the output risk level. The multi-stage pressure forming and densification process includes the following steps: first, the NdFeB powder raw material is initially pressed under a preset low pressure to form a preliminary preform with a continuous structure and an internal porosity lower than that of the initial powder state; then, on the basis of applying uniform pressure to the entire body, multiple pressure loading processes are continuously superimposed, and the amplitude and duration of each loading are gradually increased according to a preset pressure increase curve. The direction of the pressure applied during the pressurization process is kept stable and consistent, and high-frequency micro-vibration is synchronously applied throughout the pressurization process; After triggering the coarse-grained layer microstructure evolution assessment, a feature data set is extracted, which consists of a local temperature change rate, a local stress change rate, an estimated local diffusion rate, and an estimated local grain boundary migration rate; The local diffusion rate estimate is obtained by obtaining local real-time temperature data and calculating it according to the diffusion rate calculation formula. The diffusion rate calculation formula is that the local diffusion rate is equal to the initial diffusion rate multiplied by an exponential function. The negative exponential term of the exponential function is obtained by dividing the diffusion activation energy by the product of the local real-time temperature and the gas constant. The local diffusion rate is calculated according to the following formula: ； Indicates that the monitoring point is at position Place, time The diffusion rate when is the reference initial diffusion rate constant, is the material diffusion activation energy, is the gas constant, take 8.314, Indicates that the monitoring point is at position Place, time Real-time temperature at the time; The estimated value of the local grain boundary migration rate is obtained by multiplying the stress gradient measured by the local stress change rate by the linear response coefficient of the grain boundary migration driving force. The linear response coefficient of the grain boundary migration driving force is preset based on the internal grain boundary energy and grain size calibration data of the NdFeB magnet material; Each data item in the feature data group is synchronously paired according to the timestamp.

2. The method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnet according to claim 1, characterized in that: High-frequency micro-vibration is continuously applied throughout the entire multi-stage pressure forming and densification process. The frequency of the high-frequency micro-vibration is maintained in the range of 1,000 Hz to 100,000 Hz, and the direction of applying the high-frequency micro-vibration is set at a preset fixed angle to the direction of pressurization.

3. The method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnet according to claim 2, characterized in that: In the process of determining whether to trigger the coarse-grained layer microstructure evolution assessment based on the real-time monitoring of the local temperature change rate and the local stress change rate, the changing trends of the local temperature change rate and the local stress change rate are continuously monitored. When any local temperature change rate or local stress change rate exceeds the corresponding preset threshold condition, the coarse-grained layer microstructure evolution assessment process is immediately triggered; The preset threshold conditions are based on process settings, and the static calibration values ​​are set in different intervals according to different sintering stages.

4. The method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnet according to claim 3, characterized in that: The diffusion coupling heterogeneity index is derived based on the discrete spatiotemporal distribution of the local diffusion rate estimates, which includes the following steps: At each moment, the set of local diffusion rate estimates of all monitoring points in space is normalized. The normalization method is to subtract the global minimum diffusion rate from the local diffusion rate and divide it by the global diffusion rate range at the current moment to form a dimensionless normalized diffusion rate field. Based on the normalized diffusion rate field, the spatial statistical variance of the normalized diffusion rate at that moment is calculated to quantitatively characterize the degree of dispersion of the local diffusion rate distribution. The spatial statistical variance corresponding to each moment is cumulatively integrated or averaged along the time dimension to obtain the time-weighted cumulative index of spatial diffusion heterogeneity; The diffusion coupling heterogeneity index is defined as the normalized value of the cumulative index of spatial diffusion heterogeneity.

5. The method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnet according to claim 4, characterized in that: The grain boundary migration tensor index is used to quantify the anisotropic distribution characteristics of the grain boundary migration rate within the coarse-grained layer of NdFeB magnets on a spatial scale. The calculation process includes the following steps: Based on the estimated values ​​of local grain boundary migration rate collected at different monitoring positions, the local grain boundary migration velocity vector field is constructed, and the grain boundary migration velocity vector of each monitoring point is defined. for: ； represent the instantaneous migration components of the grain boundary along the X, Y, and Z directions, respectively; Based on the local grain boundary migration velocity vector, the local grain boundary mobility tensor is constructed , defined as: ; Calculate the symmetry deviation index of the local mobility tensor for the entire preform area , defined as the ratio of the sum of squares of the off-diagonal elements of the tensor to the sum of squares of the diagonal elements at each point, and is calculated as: ; They are all direction indexes, taken from the X, Y, and Z directions respectively, representing the three coordinate directions of space. is the partial derivative of the grain boundary migration velocity in the i-th direction with respect to the j-th direction, indicating the gradient change of the grain boundary migration velocity with the spatial position. It represents the sum of squares of all non-diagonal elements, reflecting the degree of change of grain boundary migration rate in non-indigenous space and the discreteness of migration direction. It represents the sum of squares of all diagonal elements, reflecting the intensity of change in grain boundary migration rate along its own direction and reflecting the degree of migration directionality; In the entire monitoring area and time period, the symmetry deviation index of all local points is spatially and temporally weighted averaged, and the grain boundary migration tensor index GBMTI is finally defined as: ; Represents the volume of the entire monitoring area, Indicates the total length of observation time.

6. The method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnet according to claim 5, characterized in that: After completing the calculation of the diffusion coupling unevenness index and the grain boundary migration tensor index, the diffusion coupling unevenness index and the grain boundary migration tensor index are synchronously input into the pre-trained machine learning model in the form of feature input vectors. The machine learning model is trained based on the historical coarse-grained layer tissue evolution data set. It can infer the risk level of the coarse-grained layer grain size anomaly under the current preparation state based on the input diffusion behavior spatial discreteness characteristics and grain boundary migration directional discreteness characteristics. The output coarse-grained layer grain size anomaly risk level is a multi-level classification result or a continuous numerical regression result. The higher the coarse-grained layer grain size anomaly risk level, the greater the potential probability of abnormal growth or uneven growth of the coarse-grained layer tissue. The machine learning model is obtained by training based on regression analysis, classification tree, support vector machine or deep neural network.

7. The method for regulating the grain size and particle size distribution of the coarse-grained layer of NdFeB magnets according to claim 6, characterized in that: According to the abnormal risk level of the coarse-grained layer grain size output by the machine learning model, the abnormal risk level is divided into three risk levels: low, medium and high, and the corresponding preset control operations are triggered according to different risk levels. Among them, when the abnormal risk level is low, the basic fine-tuning operation is performed; when the abnormal risk level is medium, the strategy switching operation is performed; when the abnormal risk level is high, the nonlinear correction operation is performed.

Images