High-temperature alloy precipitated phase segmentation and analysis method and system based on deep learning

Abstract

The invention relates to the technical field of high-temperature alloy precipitated phase segmentation, in particular to a high-temperature alloy precipitated phase segmentation and analysis method and system based on deep learning. The method comprises the following steps: performing a thermal compression experiment based on a high-temperature alloy sample to collect SEM image data; marking according to the features of the segmented target object, and performing image enhancement on the data; obtaining a loss function weight of the reference precipitated phase region by using a loss function according to shape features, structural features and the like of the reference precipitated phase region relative to other regions; calculating according to the weight of the loss function to obtain model feedback; training based on a YOLOv8 model to obtain a model; carrying out precipitated phase segmentation on the high-temperature alloy SEM image according to the model; and performing data statistics and analysis according to a segmentation result. The method can be used for identifying, segmenting and counting the phases of the same type and different precipitation positions of the high-temperature alloy, and is of great significance for improving the alloy performance.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of high-temperature alloy precipitated phase segmentation, and particularly relates to a high-temperature alloy precipitated phase segmentation and analysis method and system based on deep learning. BACKGROUND

[0002] The high-temperature alloy precipitated phase refers to a second phase particle with a certain crystal structure and chemical composition formed by alloy elements precipitating from a matrix under specific conditions in a high-temperature alloy. Excessive delta phase of Inconel 718 high-temperature alloy can accelerate the thermal corrosion of the alloy and deteriorate the high-temperature tensile properties of the material; and carbide can accelerate the sliding deformation of the alloy and thus weaken the strength of the alloy. Therefore, timely observation and reasonable control of precipitated phases at different positions are of great significance to the optimization of material design and the improvement of alloy performance.

[0003] In the prior art, a method based on machine learning is commonly used to complete the identification and segmentation of precipitated phases. Similar researches on machine learning or deep learning methods have obtained good identification and classification performance, but there is no research on using machine learning methods to identify, segment and count phases (such as delta phase) of the same type and different precipitated positions, although the regulation and control of precipitated phases at different positions are of great significance to the improvement of alloy performance. SUMMARY

[0004] To solve the problem that the prior art cannot identify, segment and count phases (such as delta phase) of the same type and different precipitated positions, the application aims to provide a high-temperature alloy precipitated phase segmentation and analysis method and system based on deep learning, and the technical scheme is as follows: a high-temperature alloy precipitated phase segmentation and analysis method based on deep learning, the method comprises the following steps: performing a hot compression experiment based on a high-temperature alloy sample, and collecting image data of compression deformation rates 0, 0.1 and 0.3 by using an SEM during the experiment; slicing the collected data according to the resolution; labeling according to the features of the segmentation target object, performing image enhancement on the data, and obtaining a data set; obtaining the loss function weight of the reference precipitated phase region by using a loss function according to the shape features and structure features of the reference precipitated phase region compared with other regions; obtaining model feedback according to the loss function weight calculation; training a model based on a YOLOv8 model to obtain the model; performing precipitated phase segmentation on the SEM image of the high-temperature alloy based on the model; and performing data statistics and analysis according to the segmentation result.

[0005] Further, the hot compression experiment step includes: sequentially arranging boron nitride, mica sheet and other heat insulation and lubricating materials on the surface of the sample in contact with the pressure rod to reduce the friction and heat loss between the sample and the pressure rod; installing the sample between the pressure head and the base of the thermal simulation testing machine, and fixing the extensometer on the sample with the extensometer clamp for measuring the displacement and strain during deformation; starting the heating system of the testing machine to heat the sample to the set experimental temperature; after the heat preservation is completed, the compression deformation of the sample is started, and the loading is carried out according to the set compression deformation rate until the predetermined deformation amount is reached; and observing the microstructure of the sample by using a scanning electron microscope and collecting image data.

[0006] Further, the data set obtaining method includes: image slicing according to the required resolution; labeling according to the characteristics of the three types of precipitated phases in the image using labeling software; and data enhancement according to the contrast and brightness of the image, and the data enhancement principle formula is as follows: In the formula, g out represents the image gray value result of data enhancement; g in represents the input image pixel point gray value; and γ represents the related index of image enhancement.

[0007] Further, the model feedback obtaining method includes: obtaining the weight according to the loss function, and the calculation formula is as follows: In the formula, y represents the label of data labeling; p represents the probability of inference segmentation; and β represents the balance parameter.

[0008] Further, the model is obtained by training based on the YOLOv8 model, including: dividing the three types of precipitated phase regions into 0, 1 and 2, and taking the remaining regions of the image as background images; training according to the model loss function until the loss function reaches a preset value to obtain a trained model; segmenting the precipitated phase of the high-temperature alloy according to the trained model; and obtaining specific precipitated phase data according to the mathematical statistics of the segmentation result.

[0009] A high-temperature alloy precipitated phase segmentation and analysis system based on deep learning, the system includes a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to realize the steps of the above-mentioned high-temperature alloy precipitated phase segmentation and analysis method based on deep learning.

[0010] The present application has the following excellent effects: the present application carries out hot compression experiment based on high-temperature alloy sample, collects image data of compression deformation rates 0, 0.1 and 0.3 by using SEM in the experiment process, so as to facilitate subsequent segmentation of high-temperature alloy precipitated phase regions by using a deep learning network model; since the resolution of the original image data is too large, in order to reduce the computer single operation amount and enrich the original data set, the original image data set is sliced, and then labeled and image enhanced according to the precipitated phase characteristics, so as to obtain the high-temperature alloy precipitated phase data set; since different types of precipitated phases have unique image characteristics and structure information, the weight is calculated by using the loss function, and the model feedback is obtained; the model is obtained by training based on YOLOv8; the high-temperature alloy precipitated phase is segmented according to the trained model; since the data information such as area and diameter of the precipitated phase is needed for material performance optimization, the data statistics are carried out according to the segmentation results, and the specific data of the precipitated phase is obtained. The present application considers the characteristics of different types of high-temperature alloy precipitated phases, so that the high-temperature alloy precipitated phase samples after data enhancement are balanced and rich, the generalization ability of the deep learning segmentation model is increased, and the segmentation accuracy is improved. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the scheme in the embodiments of the present application or the prior art, the drawings needed in the implementation examples or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0012] Figure 1 It is a flow chart of a high-temperature alloy precipitated phase segmentation and analysis method based on deep learning provided by an embodiment of the present application. DETAILED DESCRIPTION

[0013] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of a high-temperature alloy precipitated phase segmentation and analysis method and system according to the present application will be described in detail as follows. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0014] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0015] The specific scheme of a high-temperature alloy precipitated phase segmentation and analysis method and system based on deep learning provided by the present application will be specifically described below with reference to the drawings.

[0016] Referring to Figure 1 The method for separating and analyzing precipitated phases of a high-temperature alloy based on deep learning provided by the embodiment of the application comprises the following steps: S1: performing a hot compression experiment based on a high-temperature alloy sample, and collecting image data of the high-temperature alloy sample at compression deformation rates of 0, 0.1 and 0.3 during the experiment.

[0017] The embodiment of the application is mainly applied to the field of separating and analyzing images of precipitated phases of a high-temperature alloy. In order to obtain image data of the precipitated phases of the high-temperature alloy, a hot compression experiment of the high-temperature alloy is first performed, and image data of the high-temperature alloy sample during the experiment is collected. In the embodiment of the application, a deep learning model is used to separate the precipitated phases of the high-temperature alloy, and a large number of data set samples, i.e., SEM images containing the precipitated phases of the high-temperature alloy, are required to ensure that the data set samples of the model network are sufficient.

[0018] In one embodiment of the application, a YOLOv8 deep learning model is used to separate the precipitated phases of the high-temperature alloy. It should be noted that in other embodiments of the application, the same operation can be performed using a U-Net model, and both the YOLOv8 deep learning model and the U-Net model are technical means known to those skilled in the art, and will not be limited or described again.

[0019] S2: slicing the collected data according to the resolution; labeling according to the characteristics of the target objects to be separated, and performing image enhancement on the data to obtain a data set.

[0020] In one embodiment of the application, the original data set is an original SEM image with a size of 16384x12287 pixels, and the magnification of each image is 48000. The original image is cropped into an image with a size of 512x512 pixels. Since there is a large difference in the pixel grayscale values between the target objects to be separated and other regions, artificial labeling is performed on the pixel points of different types of target object regions. Since the number of original image data is limited, data enhancement is performed from the angles of contrast and brightness to enrich the samples and enhance the generalization ability of the model, and a data set is obtained.

[0021] Preferably, in one embodiment of the application, the data enhancement method comprises: realizing data enhancement according to a data enhancement principle formula, and the data enhancement principle formula is as follows: In the formula, g out represents the grayscale value of the image to be enhanced; g in represents the input image pixel grayscale value; and γ represents a related index of image enhancement.

[0022] Step S3: obtaining a loss function weight of the reference precipitated phase region according to shape features, structural features, etc. of the reference precipitated phase region compared with other regions by using the loss function; and obtaining model feedback according to loss function weight calculation;

[0023] Preferably, in an embodiment of the present application, the loss function weight calculation formula is as follows: In the formula, y represents a label of data labeling; p represents a probability of inference segmentation; and β represents a balance parameter.

[0024] In the loss function weight calculation formula, when y>0, the loss function is a variant of binary cross-entropy loss, and when y is large, even if there is a small difference between the predicted probability p and the true label y, the loss value will become relatively large due to the amplification of y, emphasizing the loss calculation for positive samples; when y=0, the loss is only related to the predicted probability p, and the shape of the loss is controlled by α and β, and when β is large, if the predicted probability p is large, the loss will increase sharply.

[0025] Step S4: training a model based on the YOLOv8 model to obtain the model; performing precipitated phase segmentation on the SEM image of the high-temperature alloy according to the model; and performing data statistics and analysis according to the segmentation result.

[0026] Preferably, in an example of the present application, the model is obtained by training based on the YOLOv8 model, which includes: taking the δ phase on the grain boundary as the 0th type of precipitated phase, taking the δ phase in the grain boundary as the 1st type of precipitated phase, and taking the carbide as the 2nd type of precipitated phase; and training the model until the weight calculated by the loss function reaches a preset value, and the model training is completed to obtain a model file.

[0027] The model obtained by training is used to perform segmentation processing on the high-temperature alloy precipitated phase image, which can avoid the problems of incomplete segmentation result and category confusion caused by similar features of the target object and the background image to the greatest extent, improve the accuracy of segmentation, and complete the segmentation of the high-temperature alloy precipitated phase.

[0028] Thus far, the segmentation of the high-temperature alloy precipitated phase is completed.

[0029] In summary, based on the high-temperature alloy sample, a hot compression experiment is performed, and during the experiment, SEM is used to collect image data at compression deformation rates of 0, 0.1 and 0.3; the collected data is sliced according to the resolution; the data is labeled according to the features of the segmentation target object, and image enhancement is performed on the data to obtain a data set; a loss function weight of a reference precipitated phase region is obtained according to shape features, structural features, etc. of the reference precipitated phase region compared with other regions by using the loss function; model feedback is obtained according to loss function weight calculation; a model is obtained by training based on the YOLOv8 model; and precipitated phase segmentation is performed on the SEM image of the high-temperature alloy according to the model.

[0030] The embodiment of the present application also provides a high-temperature alloy precipitated phase segmentation and analysis system based on deep learning, which comprises a memory, a processor and a computer program, wherein the memory is used for storing the corresponding computer program, the processor is used for running the corresponding computer program, and the computer program can realize the method described in steps S1-S4 when running in the processor.

[0031] It should be noted that the above-mentioned embodiment sequence of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or can be advantageous.

[0032] Each of the embodiments in the specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment mainly explains the difference from other embodiments.

Claims

1. A method for segmenting and analyzing precipitation phases of high-temperature alloys based on deep learning, characterized in that: The method includes: conducting a hot compression experiment on a high-temperature alloy sample, and using a SEM to collect image data with compression deformation rates of 0, 0.1, and 0.3 during the experiment; slicing the collected data according to resolution; annotating the data according to the characteristics of the segmented target object, performing image enhancement on the data, and obtaining a data set; using a loss function based on the shape characteristics and structural characteristics of the reference precipitate phase region compared with other regions, obtaining a loss function weight of the reference precipitate phase region; obtaining model feedback based on the loss function weight calculation; training based on the YOLOv8 model to obtain a model; performing precipitate phase segmentation on the high-temperature alloy SEM image according to the model; and performing data statistics and analysis based on the segmentation results.

2. The method for segmenting and analyzing high-temperature alloy precipitates based on deep learning according to claim 1, characterized in that: The thermal compression test steps include: sequentially placing thermal insulation and lubricating materials such as boron nitride and mica sheets on the contact surface of the sample and the compression rod to reduce friction and heat loss between the sample and the compression rod; installing the sample between the pressure head and the base of a thermal simulation tester, and fixing the extensometer on the sample with an extensometer chuck to measure displacement and strain during deformation; starting the heating system of the tester to heat the sample to a set experimental temperature; after the insulation is completed, starting to compress and deform the sample, and loading it according to the set compression deformation rate until a predetermined deformation amount is reached; and observing the microstructure of the sample with a scanning electron microscope to collect image data.

3. The method for segmenting and analyzing high-temperature alloy precipitates based on deep learning according to claim 1, characterized in that: The data set acquisition method includes: slicing the image according to the required resolution; annotating the image using annotation software according to the characteristics of the three types of precipitated phases; and enhancing the data according to the image contrast and brightness. The data enhancement principle formula is as follows: Where g out Indicates the image grayscale value result of data enhancement; g in Represents the grayscale value of the input image pixel; γ represents the correlation index of image enhancement.

4. The method for segmenting and analyzing high-temperature alloy precipitates based on deep learning according to claim 1, characterized in that: The model feedback obtaining method includes: obtaining weights according to a loss function, and the calculation formula is as follows: Where y represents the label of the data; p represents the probability of inference segmentation; and β represents the balance parameter.

5. The method for segmenting and analyzing high-temperature alloy precipitates based on deep learning according to claim 1, characterized in that: The model is obtained by training based on the YOLOv8 model, including: dividing the three types of precipitation phase areas into three categories: 0, 1, and 2, and the rest of the image area is used as the background image; training is performed according to the model loss function until the loss function reaches a preset value, and the trained model is obtained; the high-temperature alloy precipitation phase is segmented according to the trained model; mathematical statistics are performed on the segmentation results to obtain specific precipitation phase data.

6. A high-temperature alloy precipitate phase segmentation and analysis system based on deep learning, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the high-temperature alloy precipitate phase segmentation and analysis method based on deep learning as described in any one of claims 1 to 5 are implemented.

Images