Automatic calculation method for alpha-Al grain size in die-casting Al-Si alloy and storage medium
Through the YOLOv3 machine learning model and image processing method, the α-Al grains in the die-cast Al-Si alloy metallographic diagram are automatically identified, solving the problems of inaccurate measurement and inefficient efficiency in the prior art, and achieving efficient and accurate grain size calculations.
Patent Information
- Application Number
- CN202510554764.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-29
- Publication Date
- 2025-08-15
AI Technical Summary
The prior art is difficult to accurately distinguish the boundary between α-Al grains and eutectic Si in die-cast Al-Si alloys, resulting in inaccurate measurement results and require manual intervention, which is inefficient and easy to introduce artificial errors.
Using the trained YOLOv3 machine learning model and image processing method, the α-Al grain region and size in the die-cast Al-Si alloy metallographic diagram is automatically identified and calculated, and automated measurement is achieved through ruler identification and calibration.
The measurement efficiency and accuracy of α-Al grain size are improved, human errors are reduced, and work efficiency and measurement accuracy are improved.
Smart Images
Figure CN120495678A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of material characterization and analysis, in particular to automatic calculation of grain size in metallographic diagrams, in particular to an automatic calculation method of α-Al grain size in die-cast Al-Si alloys, and also to a storage medium. Background Art
[0002] With the growing demand for lightweighting in the automotive industry, Al-Si alloys have been widely used in the automotive die casting field due to their excellent specific strength, good casting properties and corrosion resistance. Especially in engine cylinder blocks, pistons and other key components, Al-Si alloys not only reduce vehicle weight but also improve fuel efficiency and emission performance. However, unlike the microstructure of conventional gravity casting, the Al-Si alloys produced by die casting contain both coarse α-Al (ESCs) grains and fine (α-Al) Ⅱ Grains. The presence of coarse grains significantly affects the mechanical properties, corrosion resistance and heat treatment response of die castings. Specifically, coarse α-Al (ESCs) grains may lead to increased brittleness of the material, reduced ductility and fatigue resistance, and affect the quality and reliability of the final product. On the contrary, fine (α-Al) Ⅱ Grains help to increase the strength and toughness of the material and improve its mechanical properties. Therefore, it is necessary to accurately count the coarse α-Al (ESCs) grains, fine (α-Al) Ⅱ The size of these α-Al grains is of great significance for optimizing production processes, improving product quality and developing high-performance alloys.
[0003] At present, the statistics of grain size are mainly calculated by professional software or manually calculated based on the straight line intercept method. Although these methods can achieve grain statistics, they still have certain limitations. It is difficult to accurately distinguish the boundaries between α-Al grains and eutectic Si in die-cast Al-Si alloys through software calculations, resulting in inaccurate measurement results. Moreover, the entire process often requires manual intervention to operate step by step, resulting in low overall work efficiency and increased possibility of human error. Manual calculation based on the intercept method is not only time-consuming and labor-intensive, but also unable to screen coarse α-Al (ESCs) grains from fine (α-Al) Ⅱ Grains are easily detected and human errors are introduced. In addition, granular impurities, pores and other defects in actual sample images may be mistaken for grains, further affecting measurement accuracy. Summary of the Invention
[0004] The first technical problem to be solved by the present invention is to provide an automatic calculation method for the α-Al grain size in the die-cast Al-Si alloy based on the above-mentioned existing technology, which can automatically identify and count the α-Al grains, thereby improving the measurement efficiency and accuracy.
[0005] The second technical problem to be solved by the present invention is to provide a computer-readable storage medium for the above-mentioned prior art, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the automatic calculation method of the α-Al grain size in the die-cast Al-Si alloy is realized.
[0006] The technical solution adopted by the present invention to solve the first technical problem is: a method for automatically calculating the α-Al grain size in a die-cast Al-Si alloy, characterized by comprising the following steps:
[0007] S1, initialization;
[0008] S2. Loading a trained YOLOv3 machine learning model, wherein the YOLOv3 machine learning model can identify and calibrate the scale in the metallographic image of the die-cast Al-Si alloy, and can also identify and calculate the α-Al grain area and size in the metallographic image of the die-cast Al-Si alloy;
[0009] S3, reading the metallographic image of the die-cast Al-Si alloy to be analyzed with a scale;
[0010] S4. Based on the YOLOv3 machine learning model and image processing method, the scale area in the die-cast Al-Si alloy metallographic image is identified and calibrated, and the α-Al grain area in the die-cast Al-Si alloy metallographic image is identified and selected;
[0011] S5. Calculate the size of each α-Al grain in the die-cast Al-Si alloy metallographic image based on the area of the α-Al grain region in the die-cast Al-Si alloy metallographic image and the calibration scale;
[0012] S6. Derive the size results of each α-Al grain in the metallographic image of the die-cast Al-Si alloy.
[0013] Preferably, the α-Al grains include α-Al(ESCs) grains and (α-Al) Ⅱ grains.
[0014] As an improvement, in step S5, the area of the α-Al grain region selected by the frame is used to determine the relationship between the α-Al (ESCs) grain region and the (α-Al) in the metallographic image of the die-cast Al-Si alloy. Ⅱ Grain area, the α-Al (ESCs) grain area and (α-Al) in the metallographic diagram of die-cast Al-Si alloy ⅡThe grain area is filled with different colors, and in step S6, the metallographic image of the die-cast Al-Si alloy after the color filling is simultaneously exported.
[0015] As an improvement, step S4 includes the following steps S4.1 to S4.4;
[0016] S4.1. Preprocess the metallographic image of the die-cast Al-Si alloy to be analyzed, convert it into single-precision floating-point numbers, and normalize it to meet the image size input requirements of the YOLOv3 machine learning model.
[0017] S4.2. Input the preprocessed die-cast Al-Si alloy metallographic image into the YOLOv3 machine learning model and perform target detection using the YOLOv3 machine learning model. The targets include the scale area and α-Al grain area in the die-cast Al-Si alloy metallographic image. Obtain the target's bounding box, confidence score, and category label. If a target is detected, draw the target's bounding box; otherwise, display the original image.
[0018] S4.3. Extracting a region of interest containing α-Al grains according to the detection target, and performing grayscale and binarization processing on the region of interest;
[0019] S4.4. Remove noise from the region of interest and mark connected regions;
[0020] Step 5 includes the following steps S5.1 to S5.2;
[0021] S5.1. Calculate a scale factor based on the pixels in the scale area of the detection target;
[0022] S5.2. Calculate the area of each connected region based on the scale factor and determine the properties of the α-Al grains.
[0023] As an improvement, the training method of the YOLOv3 machine learning model is as follows;
[0024] S100, initializing the training environment;
[0025] S200, obtaining a die-cast Al-Si alloy image training set, wherein each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set has a scale;
[0026] S300, selecting a scale area in each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set;
[0027] S400, based on the YOLOv3 machine learning method, trains the images in the die-cast Al-Si alloy image training set by setting hyperparameters, and then performs region recognition and calibration on the scale of each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set;
[0028] S500, based on the YOLOv3 machine learning method and image processing method, automatically identifies, selects, and calculates the area and size of each α-Al grain in each die-cast Al-Si alloy image, and exports the calculation results of each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set.
[0029] S600: Obtain the trained YOLOv3 machine learning model.
[0030] Preferably, the hyperparameters include a learning rate, a number of iterations, a regularization parameter, a confidence threshold, and an overlap threshold in non-maximum suppression.
[0031] Preferably, the criterion for selecting the hyperparameters is: taking the loss function graph of the training results as a reference, if after multiple iterations, the training loss tends to be stable and close to 0, it means that the YOLOv3 machine learning model has basically converged, and the YOLOv3 machine learning model formed by the corresponding hyperparameters will be adopted; if the training loss fluctuates greatly or is not close to 0, the hyperparameters will be adjusted and the YOLOv3 machine learning model will be retrained.
[0032] Preferably, the size results of each α-Al grain in the die-cast Al-Si alloy image are exported in *.xlsx format.
[0033] The technical solution adopted by the present invention to solve the above-mentioned second technical problem is: a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the aforementioned method for automatically calculating the α-Al grain size in the die-cast Al-Si alloy.
[0034] Compared with the prior art, the advantages of the present invention are: the automatic calculation method of the α-Al grain size in the die-cast Al-Si alloy in the present invention processes the metallographic image of the die-cast Al-Si alloy based on the trained YOLOv3 machine learning model and image processing method, and can automatically realize the selection, identification and size calculation of the α-Al grain area, and then export the results of all α-Al grains. In this process, no human intervention is required, which improves the measurement efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 Flowchart of the method for automatically calculating the α-Al grain size in the die-cast Al-Si alloy according to an embodiment of the present invention.
[0036] Figure 2 This is the metallographic diagram of die-cast Al-Si alloy.
[0037] Figure 3 To use image pro software Figure 2 Calibration scale diagram.
[0038] Figure 4 This is a map of the α-Al region automatically identified using Image Pro software based on the automatic calculation method of α-Al grain size in die-cast Al-Si alloy.
[0039] Figure 5 This is the loss function diagram of the YOLO3 model on the training set.
[0040] Figure 6 For Figure 3 Diagram for identification and calibration of scale areas in metallographic images.
[0041] Figure 7 For Figure 3 Grayscale processing and binarization processing of metallographic images.
[0042] Figure 8 For Figure 3 Calculation result diagram of the calculation area identified in the metallographic image.
[0043] Figure 9 For Figure 3 Colored image of the location of calculation results in the metallographic image. DETAILED DESCRIPTION
[0044] The present invention will be described in further detail below with reference to the accompanying drawings and embodiments.
[0045] like Figure 1 As shown, the automatic calculation method of the α-Al grain size in the die-cast Al-Si alloy in this embodiment includes the following steps S1 to S6.
[0046] S1. Initialization, which includes clearing the command window, workspace variables, and closing all graphics windows to ensure a clean environment and avoid interference from old data.
[0047] S2. Load the trained YOLOv3 machine learning model, where the YOLOv3 machine learning model can identify and calibrate the scale in the metallographic image of the die-cast Al-Si alloy, and can also identify and calculate the α-Al grain area and size in the metallographic image of the die-cast Al-Si alloy.
[0048] S3. Reading a metallographic image of the die-cast Al-Si alloy to be analyzed with a scale, wherein the metallographic image of the die-cast Al-Si alloy is selected and determined by the user.
[0049] S4. Based on the YOLOv3 machine learning model and image processing method, the scale area in the die-cast Al-Si alloy metallographic image is identified and calibrated, and the α-Al grain area in the die-cast Al-Si alloy metallographic image is identified and selected.
[0050] Specifically, step S4 includes the following steps S4.1 to S4.4.
[0051] S4.1. To facilitate the YOLOv3 machine learning model to process the exported die-cast Al-Si alloy metallographic image, the die-cast Al-Si alloy metallographic image to be analyzed is preprocessed before the YOLOv3 machine learning model calculates the die-cast Al-Si alloy metallographic image. After preprocessing, the die-cast Al-Si alloy metallographic image is converted into a single-precision floating-point number and normalized to meet the image size input requirement of the YOLOv3 machine learning model.
[0052] S4.2. Input the preprocessed die-cast Al-Si alloy metallographic image into the YOLOv3 machine learning model and use the YOLOv3 machine learning model to detect the target. The target includes the scale area and α-Al grain area in the die-cast Al-Si alloy metallographic image. The bounding box, confidence score and category label of the target are obtained. When the target is detected, the bounding box of the target is drawn. Otherwise, the original image is displayed. In addition, the α-Al grains include coarse α-Al (ESCs) grains and fine (α-Al) Ⅱ Grains. Coarse α-Al (ESCs) grains correspond to a relatively large area in the metallographic diagram of die-cast Al-Si alloys, while fine (α-Al) Ⅱ The area of the grain corresponding to the metallographic diagram of the die-cast Al-Si alloy is relatively small.
[0053] S4.3. After obtaining the target in the die-cast Al-Si alloy metallographic image, extract the region of interest containing α-Al grains according to the detection target, and perform grayscale and binarization processing on the region of interest.
[0054] S4.4. Remove noise from the region of interest. Specifically, remove small area noise and retain larger connected regions. Then, mark the connected regions on the processed die-cast Al-Si alloy metallographic image.
[0055] S5. Calculate the size of each α-Al grain in the die-cast Al-Si alloy metallographic image based on the area of the α-Al grain region in the box-marked die-cast Al-Si alloy metallographic image and the calibration scale. Determine the properties of each α-Al grain based on the different size ranges of each α-Al grain, that is, determine whether each α-Al grain region is an α-Al(ESCs) grain region or an (α-Al) Ⅱ Grain area. In order to more intuitively view the α-Al (ESCs) grain area and (α-Al) in the metallographic diagram of die-cast Al-Si alloy Ⅱ The distribution of grain area, the α-Al (ESCs) grain area and (α-Al) in the metallographic diagram of die-cast Al-Si alloyⅡ The grain area is filled with different colors.
[0056] Step 5 includes the following steps S5.1 to S5.2;
[0057] S5.1. Calculate a scale factor based on the pixels in the scale area of the detection target;
[0058] S5.2. Calculate the area of each connected region based on the scale factor and determine the properties of the α-Al grains.
[0059] S6. Export the size results of each α-Al grain in the die-cast Al-Si alloy metallographic image, and also export the color-filled die-cast Al-Si alloy metallographic image. For ease of viewing, in this embodiment, the size results of each α-Al grain in the die-cast Al-Si alloy image are exported in *.xlsx format.
[0060] The automatic calculation method of α-Al grain size in die-cast Al-Si alloy in the present invention processes the metallographic image of the die-cast Al-Si alloy based on the trained YOLOv3 machine learning model and image processing method. It can automatically realize the selection, identification and size calculation of the α-Al grain area, and then export the results of all α-Al grains. In this process, no human intervention is required, which improves the measurement efficiency and accuracy.
[0061] The training method of the YOLOv3 machine learning model is as follows S100~S600.
[0062] S100: Initialize the training environment.
[0063] S200, obtain the die-cast Al-Si alloy image training set, each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set has a ruler. In order to ensure the accuracy of the model calculation, when preparing the die-cast Al-Si alloy image training set, use Image pro software to Figure 2 The α-Al grains in the metallographic image are counted, and the scale length is selected through the calibration toolbar to calibrate the scale.
[0064] S300: Select the scale area in each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set. Specifically, select the scale position of the die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set using the Image Labeler plug-in in MATLAB software.
[0065] S400, based on the YOLOv3 machine learning method, by setting hyperparameters, where the hyperparameters include learning rate, number of iterations, regularization parameter, confidence threshold, and overlap threshold in non-maximum suppression.
[0066] The images in the die-cast Al-Si alloy image training set are trained, and then the region recognition and calibration of the scale of each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set are performed.
[0067] The standard for parameter selection is: taking the loss function graph of the training results as a reference, if after multiple iterations, the training loss tends to be stable and close to 0, it means that the YOLOv3 machine learning model has basically converged, and the YOLOv3 machine learning model formed by the corresponding hyperparameters will be adopted; if the training loss fluctuates greatly or is not close to 0, the hyperparameters will be adjusted and the YOLOv3 machine learning model will be retrained.
[0068] Specifically, when performing training calculations on the training set images, combined with Figure 5 The loss function graph of the training results, adjusting the values of the hyperparameters, ultimately achieves the recognition and calibration of the scale area of the metallographic image of the die-cast Al-Si alloy at different scales. In this embodiment, the hyperparameter settings for the YOLOv3 machine learning model are: number of iterations 1500, learning rate 0.0001, warmup period 1000, L2 regularization coefficient 0.0005, penalty threshold 0.5, confidence threshold 0.5, and non-maximum suppression overlap threshold 0.5.
[0069] S500, based on the YOLOv3 machine learning method and image processing method, automatically identifies, selects, and calculates the area and size of each α-Al grain in each die-cast Al-Si alloy image, and exports the calculation results of each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set.
[0070] S600: Obtain the trained YOLOv3 machine learning model.
[0071] The present invention also relates to a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the aforementioned method for automatically calculating the α-Al grain size in a die-cast Al-Si alloy.
[0072] The traditional image processing method is used to calculate the α-Al grain size in the die-cast Al-Si alloy, that is, Figure 3 As shown, the Image Pro software calculation toolbar is used to automatically count the bright α-Al areas in the figure, as shown in Figure 4 As shown in the figure, when the brightness area is automatically identified by the software, the scale position is also bright, so it is also identified as the α-Al area. In addition, the presence of scratches in the sample also has a certain impact on the identification of the α-Al area, resulting in Figure 4There are many unmarked dots in the distinguishing area of α-Al, which makes the calculated results different from the actual ones.
[0073] Compared with the calculations using Image Pro software, Figure 4 As shown in the results, in this embodiment, the YOLOv3 model is combined with traditional image processing methods to automatically calculate the α-Al grain size in the die-cast Al-Si alloy. The influence of scratches on the α-Al area is reduced through grayscale processing, binarization processing, noise reduction and other processes. The calculation results are shown in Figure 9 shown, than Figure 4 The calculation results are more accurate, and the YOLOv3 model combined with traditional image processing methods automatically calculates the α-Al grain size in die-cast Al-Si alloys, eliminating the need for manual calibration, image export, and data processing. This improves efficiency and makes the calculation of the α-Al grain size of die-cast Al-Si alloys faster and more convenient. With the help of the YOLOv3 machine learning model, the complex features of the grains are automatically learned, ensuring that the target area can be accurately located and segmented even in complex backgrounds. Based on the learned features, the minimum enclosing rectangle method is used to accurately calculate the size of α-Al, and a detailed statistical report is generated to support further data analysis and decision-making. This method not only improves measurement accuracy, but also greatly improves work efficiency.
Claims
1. A method for automatically calculating the α-Al grain size in a die-cast Al-Si alloy, characterized by: The following steps are involved: S1, initialization; S2. Loading a trained YOLOv3 machine learning model, wherein the YOLOv3 machine learning model can identify and calibrate the scale in the metallographic image of the die-cast Al-Si alloy, and can also identify and calculate the α-Al grain area and size in the metallographic image of the die-cast Al-Si alloy; S3, reading the metallographic image of the die-cast Al-Si alloy to be analyzed with a scale; S4. Based on the YOLOv3 machine learning model and image processing method, the scale area in the die-cast Al-Si alloy metallographic image is identified and calibrated, and the α-Al grain area in the die-cast Al-Si alloy metallographic image is identified and selected; S5. Calculate the size of each α-Al grain in the die-cast Al-Si alloy metallographic image based on the area of the α-Al grain region in the die-cast Al-Si alloy metallographic image and the calibration scale; S6. Derive the size results of each α-Al grain in the metallographic image of the die-cast Al-Si alloy.
2. The method for automatically calculating the α-Al grain size in the die-cast Al-Si alloy according to claim 1, characterized in that: α-Al grains include α-Al(ESCs) grains and (α-Al) Ⅱ grains.
3. The method for automatically calculating the α-Al grain size in the die-cast Al-Si alloy according to claim 2, characterized in that: In step S5, the area of the α-Al grain region selected by the frame is determined based on the area of the α-Al grain region and the (α-Al) Ⅱ Grain area, the α-Al (ESCs) grain area and (α-Al) in the metallographic diagram of die-cast Al-Si alloy Ⅱ The grain area is filled with different colors, and in step S6, the metallographic image of the die-cast Al-Si alloy after the color filling is simultaneously exported.
4. The method for automatically calculating the α-Al grain size in a die-cast Al-Si alloy according to any one of claims 1 to 3, characterized in that: Step S4 includes the following steps S4.1 to S4.4; S4.
1. Preprocess the metallographic image of the die-cast Al-Si alloy to be analyzed, convert it into single-precision floating-point numbers, and normalize it to meet the image size input requirements of the YOLOv3 machine learning model. S4.
2. Input the preprocessed die-cast Al-Si alloy metallographic image into the YOLOv3 machine learning model and perform target detection using the YOLOv3 machine learning model. The targets include the scale area and α-Al grain area in the die-cast Al-Si alloy metallographic image. Obtain the target's bounding box, confidence score, and category label. If a target is detected, draw the target's bounding box; otherwise, display the original image. S4.
3. Extracting a region of interest containing α-Al grains according to the detection target, and performing grayscale and binarization processing on the region of interest; S4.
4. Remove noise from the region of interest and mark connected regions; Step 5 The method includes the following steps S5.1 to S5.2; S5.
1. Calculate a scale factor based on the pixels in the scale area of the detection target; S5.
2. Calculate the area of each connected region based on the scale factor and determine the properties of the α-Al grains.
5. The method for automatically calculating the α-Al grain size in a die-cast Al-Si alloy according to any one of claims 1 to 3, characterized in that: The training method of the YOLOv3 machine learning model is as follows; S100, initializing the training environment; S200, obtaining a die-cast Al-Si alloy image training set, wherein each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set has a scale; S300, selecting a scale area in each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set; S400, based on the YOLOv3 machine learning method, trains the images in the die-cast Al-Si alloy image training set by setting hyperparameters, and then performs region recognition and calibration on the scale of each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set; S500, based on the YOLOv3 machine learning method and image processing method, automatically identifies, selects, and calculates the area and size of each α-Al grain in each die-cast Al-Si alloy image, and exports the calculation results of each die-cast Al-Si alloy image in the die-cast Al-Si alloy image training set. S600: Obtain the trained YOLOv3 machine learning model.
6. The method for automatically calculating the α-Al grain size in the die-cast Al-Si alloy according to claim 5, characterized in that: The hyperparameters include learning rate, number of iterations, regularization parameter, confidence threshold, and overlap threshold in non-maximum suppression.
7. The method for automatically calculating the α-Al grain size in the die-cast Al-Si alloy according to claim 6, characterized in that: The standard for selecting the hyperparameters is: taking the loss function graph of the training results as a reference, if after multiple iterations, the training loss tends to be stable and close to 0, it means that the YOLOv3 machine learning model has basically converged, and the YOLOv3 machine learning model formed by the corresponding hyperparameters will be adopted; if the training loss fluctuates greatly or is not close to 0, the hyperparameters will be adjusted and the YOLOv3 machine learning model will be retrained.
8. The method for automatically calculating the α-Al grain size in a die-cast Al-Si alloy according to any one of claims 1 to 3, characterized in that: Export the size results of each α-Al grain in the die-cast Al-Si alloy image in *.xlsx format.
9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the method for automatically calculating the α-Al grain size in the die-cast Al-Si alloy according to any one of claims 1 to 8 is implemented.