Method for evaluating orientation degree of sintered neodymium-iron-boron magnet

Abstract

The invention belongs to the field of rare earth permanent magnet materials, and particularly relates to a method for evaluating the orientation degree of a sintered neodymium-iron-boron magnet. The method comprises the following steps: carrying out electron backscatter diffraction (EBSD) analysis on the neodymium iron boron magnet, and carrying out electron backscatter diffraction (EBSD) analysis on the neodymium iron boron magnet; euler angle data obtained by the EBSD test are converted into an included angle theta i between a crystal grain c axis and an orientation field direction through mathematical transformation; the method comprises the following steps: carrying out automatic segmentation on an EBSD image by utilizing an image segmentation large model (SAM) to obtain a high-precision segmentation mask of each crystal grain; traversing the acquisition points in the mask of each crystal grain to obtain a staggered orientation angle of each crystal grain; and carrying out statistics on error orientation angles of all crystal grains, drawing a frequency distribution histogram and a cumulative distribution histogram of the error orientation angles, and outputting an orientation degree analysis report. The method for evaluating the orientation degree of the sintered neodymium-iron-boron magnet is established by combining EBSD and SAM image segmentation technologies, and the method has the advantages of high accuracy and high efficiency at the same time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material microstructure analysis, in particular to a method for evaluating the orientation degree of sintered neodymium-iron-boron magnets. BACKGROUND

[0002] Sintered neodymium-iron-boron magnets, as the third generation of rare earth permanent magnet materials, have become the core functional materials in the fields of consumer electronics, aerospace and new energy vehicles, etc. due to their ultra-high permanent magnet performance. Coercivity and remanence are two of the most important performance indicators, and the remanence is highly related to the orientation degree of the magnet. Since the Nd2Fe 14 B grains have strong uniaxial magnetic anisotropy, only when the easy magnetization axis (c-axis) is highly consistent, the directional output of the magnetic moment can be maximized, thereby obtaining high remanence. Therefore, achieving the highly consistent arrangement of the c-axis of the single crystal particles is the key target for preparing high-performance sintered neodymium-iron-boron magnets.

[0003] At present, the commonly used methods for evaluating the orientation degree of the magnet are X-ray method and remanence ratio method. The X-ray method reflects the orientation by comparing the ratio of the diffraction peak intensities of (105) and (006) crystal planes, but the peak intensity is significantly affected by stress and composition, and the ratio often cannot truly reflect the orientation degree, and the results obtained have large dispersion and limited accuracy; the remanence ratio method indirectly infers the orientation degree by calculating the remanence ratio of the easy axis and the hard axis, which is essentially a method of deducing the microstructure from the macro performance, and lacks crystallographic significance. These two methods can only give a single orientation degree value, and cannot reveal the orientation distribution characteristics of the sample. Therefore, these two methods can only be used for lateral comparison between samples, and lack absolute numerical significance.

[0004] Electron backscatter diffraction technology (EBSD) is a reliable method for analyzing the texture of metal materials, and the orientation distribution of a single grain in three-dimensional space is quantitatively characterized by Euler angles. Although this method is accurate, it is a microscopic statistical method that lacks statistical quantity and is heavily dependent on manual calibration. The present application proposes an accurate evaluation method for the orientation degree of sintered neodymium-iron-boron magnets based on the quantitative combination of EBSD grain orientation and SAM image segmentation, which can not only directly quantify the orientation state of a single grain in a crystallographic sense, but also obtain the orientation distribution of the whole sample, and has clear physical significance and statistical representativeness, thereby establishing a fast, accurate and quantifiable method for analyzing the orientation degree of the magnet. SUMMARY In order to accurately and quickly quantitatively evaluate the orientation degree of the magnet, the present application provides a method for evaluating the orientation degree of sintered neodymium-iron-boron magnets, which directly identifies the EBSD image by using a SAM large model, and calculates the misorientation angle of each grain by combining the EBSD Euler angle data and the SAM high-precision segmentation mask. , and further establish a method for evaluating the orientation degree of sintered Nd-Fe-B magnets.

[0005] The technical scheme adopted by the present application is: a method for evaluating the orientation degree of sintered Nd-Fe-B magnets, comprising the following steps: Step 1: prepare an EBSD sample with a clean, smooth and stress-free surface, and collect an EBSD image; Step 2: mathematically transform the EBSD Euler angle data to obtain the angle between the c-axis of each collection point and the orientation field direction, defined as the single-point misorientation angle ; Step 3: segment the EBSD image using the image segmentation large model (SAM) to obtain the high-precision mask of each grain; The specific steps of step 3 are: Step 3.1: SAM is trained by 110 million high-definition images of various types and 10 billion mask images, with up to 630 million parameters, which can realize zero-shot transfer to new image segmentation tasks, and has very strong generalization ability; the SAM architecture is simple and efficient, consisting of an Image Encoder, a Prompt Encoder and a Mask Encoder. The Image Encoder is an encoder based on the Vision Transformer architecture, pre-trained using Masked Autoencoder (MAE) technology, which is good at extracting global and local features of input images. This encoder only needs to run once on the image and output a dense image embedding. This design distributes the computational overhead, so that no matter how many prompts the user provides, the image features only need to be extracted once, greatly improving the overall efficiency, especially for interactive applications. The Prompt Encoder is a prompt encoder that can handle various interactive prompts provided by the user, including: text prompts, point prompts and box prompts; the Mask Decoder is a lightweight but efficient module that generates high-quality target segmentation masks based on image embeddings from the image encoder and prompt embeddings from the prompt encoder. The SAM model can be directly obtained from the official website https: / / github.com / facebookresearch / segment-anything, which provides three pre-trained models with different generalization abilities, which can be selected as needed. The EBSD image is input into the SAM model for rapid segmentation.

[0006] Step 3.2: collect all the segmentation masks for preprocessing, first set the pixel value threshold to delete the segmentation masks of the Nd-rich phase and contaminants; then set the IoU threshold to delete the grain masks that are repeatedly segmented. Finally, high-quality segmentation masks of all grains are obtained.

[0007] Step 4: Traverse the sampling points inside each grain mask and calculate the misalignment angle of each grain. ; The specific steps of step 4 are as follows: Step 4.1: Establish the mapping relationship between the mask coordinates and the coordinates of the EBSD acquisition points.

[0008] Step 4.2: Based on the definition in Step 2 For all valid point sets within the k-th grain mask S k Calculate the misorientation angle :

[0009] Step 5: Repeat steps 2-4 above to calculate the misorientation angle of the grains in multiple EBSD images. To ensure statistical accuracy; Step 6: Calculate the misorientation angles of all grains. Draw the corresponding frequency distribution histogram and cumulative distribution histogram and output the orientation analysis report; The specific steps of step 6 are as follows: definition: A 10 For the wrong orientation angle This value represents 10% of the grains; A 50 For the wrong orientation angle This value represents 50% of the grains; A 90 For the wrong orientation angle This value represents 90% of the grains; A 90 / A 10 for A 90 and A 10 The ratio of the average misorientation angle is used to measure the dispersion of the misorientation angle distribution; the average misorientation angle For the wrong orientation angle The average value; the magnet orientation degree is the average misorientation angle. The cosine value.

[0010] This invention combines EBSD and SAM image segmentation techniques to establish a method for evaluating the orientation degree of sintered NdFeB magnets, which has the advantages of both high accuracy and high efficiency. Attached Figure Description

[0011] Figure 1 This is the segmentation result of the EBSD data of magnet A and magnet B in an example of the present invention; Figure 2 Frequency distribution histograms and cumulative distribution histograms of the magnets A and B drawn for embodiments of the present application. DETAILED DESCRIPTION

[0012] The specific embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0013] A method for evaluating the orientation degree of sintered neodymium-iron-boron magnets, comprising the following steps: Step 1: respectively cut samples on bulk magnets A and B, prepare EBSD samples with clean, smooth and stress-free surface and collect EBSD images; Step 2: mathematically transform the EBSD Euler angle data to obtain the angle between the c-axis of each collection point and the orientation field direction, defined as single-point misorientation angle ; Step 3: segment the image using the SAM image segmentation large model to obtain the high-precision mask of each grain. Figure 1 The recognition results of the SAM large model on the EBSD images of the magnets A and B;

[0014] Step 3.1: SAM is trained on 110 million high-definition images of various types and 10 billion mask images, with up to 630 million parameters, enabling zero-shot transfer to new image segmentation tasks and strong generalization capabilities. The SAM architecture is highly efficient and consists of an Image Encoder, a Prompt Encoder, and a Mask Encoder. The Image Encoder is a Vision Transformer-based encoder pre-trained using Masked Autoencoder (MAE) technology, which is good at extracting global and local features of input images. This encoder only needs to run once on an image and output a dense image embedding. This design distributes the computational overhead, so that regardless of the number of prompts provided by the user, the image features only need to be extracted once, greatly improving overall efficiency, especially for interactive applications. The Prompt Encoder is a prompt encoder that can handle various interactive prompts provided by the user, including text prompts, point prompts, and box prompts. The Mask Decoder is a lightweight but efficient module that generates high-quality target segmentation masks based on image embeddings from the image encoder and prompt embeddings from the prompt encoder. The SAM model is available on the official website https: / / github.com / facebookresearch / segment-anything, which provides three pre-trained models with different generalization capabilities that can be selected as needed. The EBSD images of magnet A and magnet B are input into the SAM model for rapid segmentation.

[0015] Step 3.2: Collect all segmentation masks, first set the pixel threshold to remove the segmentation masks of Nd-rich phases and contaminants; then set the IoU threshold to compare the IoU between masks and delete duplicate grains. Finally, obtain high-quality segmentation masks for all grains in the image.

[0016] Step 4: Traverse the collection points inside each grain mask and calculate the misorientation angle of each grain ; Step 4.1, establish the mapping relationship between the mask coordinates and the EBSD collection point coordinates.

[0017] Step 4.2, according to the definition of , calculate the misorientation angle S k of all valid points in the kth grain mask :

[0018] Step 5: Repeat steps 2-4 above to calculate the misorientation angles of all grains in the EBSD map to ensure the statistics; Step 6: According to the results, draw the corresponding frequency distribution histogram and cumulative distribution histogram respectively and output the orientation degree analysis report, as shown in Figure 2 and Table 1.

[0019] In this statistics, magnet A and magnet B contain 646 and 695 grains respectively. The SAM-based image analysis algorithm is completed on a single NVIDIA RTX 4090 graphics card, and the calculation time is 11 minutes and 49 seconds, which reflects the efficiency of the method. From the statistical results, it can be seen that the average misorientation angle of magnet B is 14.6 °, A 50 14.0 °, which are significantly lower than 16.8 ° and 15.4 ° of magnet A. In addition, the orientation distribution of magnet B is more concentrated, with a standard deviation of 7.3 °, A 90 / A 10 3.9, both of which are smaller than magnet A, indicating that the grain orientation consistency of magnet B is better and the distribution is narrower. Therefore, magnet B has smaller grain orientation deviation and more concentrated distribution characteristics, and the overall orientation degree is better than magnet A.

[0020] Table 1 is the orientation analysis report of magnet A and magnet B output by the example of the present application;

Claims

1. A method for evaluating the orientation degree of sintered NdFeB magnets, characterized in that, The method includes the following steps: Step 1: Prepare a metallographic sample with a clean, smooth surface and no residual stress, and acquire electron backscatter diffraction (EBSD) images; Step 2: Perform mathematical transformation on the EBSD Euler angle data to obtain the angle between the c-axis of the grain and the orientation field direction at each acquisition point, which is defined as the single-point fault orientation angle. θ i ; Step 3: Use the Segment Anything Model (SAM) to segment the EBSD image, and preprocess the output to obtain a high-precision segmentation mask for each grain; Step 4: Traverse the sampling points inside each grain mask and calculate the misorientation angle of each grain. ; Step 5: Repeat steps 2-4 above to calculate the grain misorientation angles in multiple EBSD images. To ensure statistical accuracy; Step 6: Calculate the misorientation angles of all grains. Draw frequency distribution histograms and cumulative distribution histograms of misalignment angles, and output an analysis report on orientation degree.

2. The method as described in claim 1, characterized in that, Step 2 specifically includes: Because the c-axis of each grain in the sintered NdFeB magnet is not perfectly aligned, its distribution forms a conical surface around the orientation field direction; based on the Euler angles acquired by EBSD... Through mathematical transformation, the angle between the c-axis of the grain and the orientation field direction at each sampling point is obtained, defined as the single-point fault orientation angle. θ i ,in, θ i ∈[0°,90°].

3. The method as described in claim 1, characterized in that, Step 3 specifically includes: Step 3.1: SAM is an open-source image segmentation model released by Meta AI, consisting of Image Encoder, Prompt Encoder, and Mask Encoder. It is known as the "Chatgpt" of the computer vision field due to its powerful generalization ability and zero-shot transfer ability. The model provides three pre-trained models with different generalization abilities: vit_b, vit_l, and vit_h, which can be selected according to the needs of the task. Step 3.2: Input the EBSD image into the SAM model for automatic full-image pixel-level segmentation; then preprocess the recognition results, including: setting pixel value thresholds and deleting segmentation masks of Nd-rich phases and contaminants; setting the intersection-over-union (IoU) threshold and deleting grain masks of repeated segmentation.

4. The method as described in claim 1, characterized in that, Step 4 specifically includes: Step 4.1: Establish the mapping relationship between the mask coordinates and the coordinates of the EBSD acquisition points; Step 4.2: Based on the definition in Step 2 For all valid point sets within the k-th grain mask S k Calculate the misorientation angle : 。 5. The method as described in claim 1, characterized in that, Step 6 is defined as follows: A 10 For the wrong orientation angle This value represents 10% of the grains; A 50 For the wrong orientation angle This value represents 50% of the grains; A 90 For the wrong orientation angle This value represents 90% of the grains; A 90 / A 10 for A 90 and A 10 The ratio of the average misorientation angle is used to measure the dispersion of the misorientation angle distribution; the average misorientation angle... For the wrong orientation angle The average value; the magnet orientation degree is the average misorientation angle. The cosine value.

Images