Graphene oxide / metal mixed powder uniformity quantitative evaluation method based on machine learning and image recognition

Abstract

The invention belongs to the field of composite material powder quality evaluation, and relates to a quantitative evaluation method for the uniformity of graphene oxide / metal mixed powder based on machine learning and image recognition. The method comprises the following steps: firstly, shooting a large number of scanning electron microscope images of mixed powder prepared by a solution stirring and mixing method; then metal powder and graphene oxide in each image are manually calibrated, machine learning recognition training is carried out, and a reasonable recognition model is determined; and finally, intelligently identifying the metal powder and the graphene oxide in the mixed powder at the characteristic position, calculating the area content of the graphene oxide in each image, and taking the average value, the standard deviation and the range of the area content of the graphene oxide in the two-dimensional scanning electron microscope image as uniformity evaluation indexes. According to the method, quantitative evaluation of the uniformity of the graphene oxide / metal mixed powder is realized for the first time, and the evaluation method based on machine learning and image recognition has the advantages of high recognition speed and high accuracy.

Description

Technical Field

[0001] The invention belongs to the field of composite material powder quality evaluation, and in particular relates to a method for quantitatively evaluating the uniformity of graphene oxide / metal mixed powder based on machine learning and image recognition. Background Art

[0002] Graphene oxide is a new type of high-strength two-dimensional nanomaterial with a tensile strength of more than 60GPa, which is much higher than that of metal materials. After adding a small amount of graphene oxide to metals and alloys such as Cu, Al, Ti and Ni to prepare graphene oxide-reinforced metal composites, the mechanical properties of the materials can be greatly improved. Studies have shown that after adding 0.5wt% of graphene oxide to Cu, the room temperature compressive strength increased by 28%; after adding 0.3vol% of graphene oxide to Al, the room temperature tensile strength increased by more than 23%; after adding 0.3wt% of graphene oxide to Ti, the room temperature microhardness increased by 25.6%. The preparation method of graphene oxide / metal composites is mainly powder metallurgy, that is, firstly mixing graphene oxide and metal powder into mixed powder, and then densifying the mixed powder into a composite material.

[0003] The two-dimensional flaky nanostructure of graphene oxide makes its layers have strong van der Waals forces, which makes it easy to agglomerate and difficult to disperse in metal powders. The uneven dispersion of graphene oxide in metal powders will sharply reduce the mechanical properties of the composite material, which is the main factor restricting the performance improvement of graphene oxide / metal composite materials. Effective evaluation of the uniformity of graphene oxide / metal mixed powders is of great significance to the optimization of powder mixing process and the improvement of composite material performance. However, there is currently no effective method for quantitatively evaluating the uniformity of graphene oxide / metal mixed powders. Summary of the invention

[0004] The purpose of the present invention is to provide a method for quantitatively evaluating the uniformity of graphene oxide / metal mixed powder based on machine learning and image recognition.

[0005] The technical solution of the present invention is: a method for quantitatively evaluating the uniformity of graphene oxide / metal mixed powder based on machine learning and image recognition, comprising the following steps: S1. Putting the graphene oxide anhydrous ethanol solution and the metal powder into a stirring tank to stir and mix the solution, and after the anhydrous ethanol evaporates and the mixture in the stirring tank is in a semi-dry state, selecting mixed powder samples at multiple random positions of the mixed powder in the stirring tank; S2. Place the obtained mixed powder sample in a vacuum drying oven and dry it for 4 to 6 hours at a drying temperature of 60 to 70°C and a vacuum degree of ≤10 kPa; S3, placing the dried mixed powder sample under a scanning electron microscope to take a secondary electron image, and randomly taking 3 to 6 images with different fields of view for the sample at each position; S4, randomly cutting the captured scanning electron microscope images into the same size, and manually calibrating the graphene oxide and metal powder pixels. Due to the solution stirring and mixing preparation method, the graphene oxide is adsorbed on the surface of the metal powder under the action of the surface tension of anhydrous ethanol, so the metal powder pixels contain graphene oxide; S5, enlarge the calibrated image, perform horizontal flipping, vertical flipping, 90° rotation, 180° rotation, 270° rotation and translation, and other data enhancements to expand the number of images in the data set; S6. Use the U-net network model to perform machine learning training on the dataset images. The training parameters are: learningrate = 1×10 -3 ~1×10 -7 , batch size=4~64, epoch=100; S7. Perform recognition tests on the trained metal powder recognition model and graphene oxide recognition model. If the model recognition accuracy is greater than 0.9, the trained model meets the requirements. Where T P is the number of samples correctly predicted as positive examples by the model, T N is the number of samples correctly predicted as negative examples by the model, F P is the number of samples that the model incorrectly predicts as positive examples, F N is the number of samples that the model incorrectly predicts as negative examples; S8. Use the above recognition model that meets the requirements to perform intelligent recognition of graphene oxide and metal powder on the powder images of 9 characteristic positions of the mixed powder in the mixing tank, including the bottom center, bottom edge, bottom 1 / 2 radius, middle center, middle edge, middle 1 / 2 radius, surface center, surface edge and surface 1 / 2 radius. The characteristic positions of powder sampling for uniformity evaluation are shown in Figure 1 , get the number of graphene oxide pixels in each image and the number of metal powder pixels i is the i-th image, metal powder recognition input image, metal powder recognition image, graphene oxide recognition input image and graphene oxide recognition image are shown in Figure 2 to Figure 5 ; S9. Calculate the area content of graphene oxide in each image S10, calculating the average, standard deviation and range of the area content of graphene oxide in the above 9 position images; S11. The average value of the area content of graphene oxide is used to characterize the average content of graphene oxide in the mixed powder. The standard deviation and range are used as evaluation indicators to quantitatively evaluate the uniformity of the graphene oxide / metal mixed powder. The smaller the standard deviation and range, the more uniform the dispersion.

[0006] The secondary electron image magnification is 200 to 500 times.

[0007] At least 95% of the graphene oxide in the mixed powder after drying is adsorbed on the metal powder.

[0008] The metal powder is spherical and has a particle size of 15 to 250 μm.

[0009] The size of the graphene oxide sheet is no greater than the particle size of the metal powder.

[0010] The present invention has the following advantages and outstanding effects:

[0011] (1) According to the method proposed in the present invention, the dispersion uniformity of graphene oxide in metal powder and the mixing quality of graphene oxide / metal powder can be quantitatively evaluated, which plays an important role in the process optimization of graphene oxide / metal mixed powder and the preparation of high-performance graphene oxide reinforced metal matrix composites. (2) The present invention uses machine learning and image recognition to intelligently identify metal powder and graphene oxide in the mixed powder. Compared with manual naked eye recognition, the recognition speed is fast and the accuracy is high. (3) The present invention first proposes to use the area content of graphene oxide in the two-dimensional image of the mixed powder as the basic data for evaluating uniformity to achieve quantitative evaluation of the uniformity of graphene oxide / metal powder; (4) The powder sampling positions used for uniformity evaluation in the present invention are the bottom center, bottom edge, bottom 1 / 2 radius, middle center, middle edge, middle 1 / 2 radius, surface center, surface edge and surface 1 / 2 radius of the mixed powder in the stirring tank. The sampling positions include various characteristic positions of the powder after stirring, and have high statistical accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 Characteristic positions for powder sampling for uniformity evaluation, where: 1-mixing tank, 2-mixed powder, 3-sampling position.

[0013] Figure 2 An input image is provided for metal powder recognition, wherein: 1-metal powder containing graphene oxide, 2-graphene oxide adsorbed on the metal powder.

[0014] Figure 3 Image for metal powder recognition.

[0015] Figure 4 Input image for graphene oxide identification.

[0016] Figure 5 Identify images for graphene oxide. DETAILED DESCRIPTION

[0017] The following embodiments of the technical solution of the present invention will be described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0018] Embodiment 1:

[0019] 0.3g of graphene oxide was placed in 300ml of anhydrous ethanol and mixed with 100g of titanium alloy powder with a particle size of 53-105μm according to three different powder mixing processes: process one, process two and process three. After the mixture became semi-dry, samples were taken at 20 random locations of the three process mixed powder samples. The obtained powder samples were placed in a vacuum drying oven and dried for 4h, the drying temperature was 60℃, and the vacuum degree was ≤10KPa. The dried powder samples were placed in a scanning electron microscope and photographed in the secondary electron imaging mode with a magnification of 300 times and a photo size of 1280×960 pixels. The captured scanning electron microscope images were randomly cut into 640×480 pixels, and the graphene oxide and metal powder pixels were manually calibrated for each image. The calibrated images were enlarged, horizontally flipped, vertically flipped, 90° rotated, 180° rotated, 270° rotated, and translated, and the number of images in the expanded data set exceeded 1000 images. The U-net network model is used to perform machine learning training on the dataset images. The training parameters are: learning rate = 1×10 -3 ~1×10 -7 , batch size = 4 to 64, epoch = 100. The trained metal powder recognition model and graphene oxide recognition model are tested for recognition. When learning rate = 1 × 10 -6 When batch size = 8 and epoch = 35, the metal powder recognition accuracy reaches 0.95, which meets the recognition requirements; when learning rate = 1×10 -7When batch size=4 and epoch=20, the graphene oxide recognition accuracy reaches 0.91, which meets the recognition requirements, and the metal powder and graphene oxide recognition models are obtained. Sampling, drying and scanning electron microscope images are taken at 9 positions including the bottom center, bottom edge, bottom 1 / 2 radius, middle center, middle edge, middle 1 / 2 radius, surface center, surface edge and surface 1 / 2 radius of the mixed powders prepared by the three powder mixing processes. Three images are taken at each position. The metal powder and graphene oxide recognition models obtained above are used to perform intelligent recognition of metal powder and graphene oxide respectively, and the number of pixels of metal powder and graphene oxide in each image is obtained. According to the area content calculation formula of graphene oxide Calculate the area content C of graphene in each image i ,in is the number of graphene oxide pixels, is the number of metal powder pixels, and i is the i-th image. The average, standard deviation and range of the area content of graphene oxide in the above 27 images are calculated, and the preparation process with the smaller standard deviation and range is selected as the preferred process.

Claims

1. A quantitative evaluation method for the uniformity of graphene oxide / metal mixed powder based on machine learning and image recognition, characterized in that: The following steps are involved: S1. Putting the graphene oxide anhydrous ethanol solution and the metal powder into a stirring tank to stir and mix the solution, and after the anhydrous ethanol evaporates and the mixture in the stirring tank is in a semi-dry state, selecting mixed powder samples at multiple random positions of the mixed powder in the stirring tank; S2. Place the obtained mixed powder sample in a vacuum drying oven and dry it for 4 to 6 hours at a drying temperature of 60 to 70°C and a vacuum degree of ≤10 kPa; S3, placing the dried mixed powder sample under a scanning electron microscope to take a secondary electron image, and randomly taking 3 to 6 images with different fields of view for the sample at each position; S4, randomly cutting the captured scanning electron microscope images into the same size, and manually calibrating the graphene oxide and metal powder pixels. Due to the solution stirring and mixing preparation method, the graphene oxide is adsorbed on the surface of the metal powder under the action of the surface tension of anhydrous ethanol, so the metal powder pixels contain graphene oxide; S5, enlarge the calibrated image, perform horizontal flipping, vertical flipping, 90° rotation, 180° rotation, 270° rotation and translation, and other data enhancements to expand the number of images in the data set; S6. Use the U-net network model to perform machine learning training on the dataset images. The training parameters are: learning rate = 1×10 -3 ~1×10 -7 , batch size=4~64, epoch=100; S7. Perform recognition tests on the trained metal powder recognition model and graphene oxide recognition model. If the model recognition accuracy is greater than 0.9, the trained model meets the requirements. Where T P is the number of samples correctly predicted as positive examples by the model, T N is the number of samples correctly predicted as negative examples by the model, F P is the number of samples that the model incorrectly predicts as positive examples, F N is the number of samples that the model incorrectly predicts as negative examples; S8. Use the above recognition model that meets the requirements to perform intelligent recognition of graphene oxide and metal powder on powder images at nine characteristic positions of the mixed powder in the mixing tank, including the bottom center, bottom edge, bottom 1 / 2 radius, middle center, middle edge, middle 1 / 2 radius, surface center, surface edge and surface 1 / 2 radius, and obtain the number of graphene oxide pixels in each image. and the number of metal powder pixels i is the i-th image; S9. Calculate the area content of graphene oxide in each image S10, calculating the average, standard deviation and range of the area content of graphene oxide in the above 9 position images; S11. The average value of the area content of graphene oxide is used to characterize the average content of graphene oxide in the mixed powder. The standard deviation and range are used as evaluation indicators to quantitatively evaluate the uniformity of the graphene oxide / metal mixed powder. The smaller the standard deviation and range, the more uniform the dispersion.

2. The method for quantitatively evaluating the uniformity of graphene oxide / metal mixed powder based on machine learning and image recognition according to claim 1, characterized in that: The secondary electron image magnification is 200 to 500 times.

3. The method for quantitatively evaluating the uniformity of graphene oxide / metal mixed powder based on machine learning and image recognition according to claim 1, characterized in that: At least 95% of the graphene oxide in the dried mixed powder is adsorbed on the metal powder.

4. The method for quantitatively evaluating the uniformity of graphene oxide / metal mixed powder based on machine learning and image recognition according to claim 1, characterized in that: The metal powder is spherical and has a particle size of 15 to 250 μm.

5. The method for quantitatively evaluating the uniformity of graphene oxide / metal mixed powder based on machine learning and image recognition according to claim 1, characterized in that: The size of the graphene oxide sheet is no greater than the particle size of the metal powder.

Images