Intelligent prediction method for shunting holes of aluminum alloy extrusion die

Abstract

The invention discloses an intelligent prediction method for diversion holes of an aluminum alloy extrusion die. The intelligent prediction method comprises the following steps of performing statistics and analysis on four-diversion-hole type extrusion die models and profile data; extracting effective information of the profile and the distribution hole graph, and performing data enhancement on the image by using cv2; establishing shunting hole data sets corresponding to the complete profile and the segmented profile respectively; training the data set by using the intelligent prediction model of the extrusion die shunting hole, and establishing the intelligent prediction model of the extrusion die shunting hole; and inputting the preprocessed picture into the trained extrusion die shunting hole intelligent prediction model to generate a shunting hole picture corresponding to the profile. The modeling method disclosed by the invention is simple to operate, the established prediction model can be used for rapidly designing the shunting hole structure of the aluminum profile extrusion die, and the die design process is shortened.

Description

Technical Field

[0001] The invention relates to the field of aluminum profile extrusion die design, and in particular to an intelligent prediction method for diversion holes in aluminum alloy extrusion dies. Background Art

[0002] Aluminum profiles are widely used in industries such as construction, transportation, electronics, and machinery due to their excellent performance and diverse applications. Extrusion is the primary production method for aluminum profiles, enabling the production of profiles in a variety of shapes and sizes. The manifold die is crucial in aluminum extrusion, particularly the four-hole manifold die, which is often used in the production of hollow parts. Its design evenly distributes the metal to the four cavities, ensuring consistent profile wall thickness and improving efficiency. The manifold holes ensure that the bar stock is evenly distributed across the various cavities during the extrusion process. The size, shape, and distribution of the manifold holes directly impact the stability of the extrusion process and the final quality of the product. Therefore, precise design of the manifold holes is crucial during the mold design phase.

[0003] Currently, simulation is often used for preliminary research to improve the accuracy and efficiency of extrusion die design. This allows for the identification of potential extrusion process issues before actual die production, thereby optimizing die design. However, while this approach improves design quality to a certain extent, its computational complexity and time-consuming nature slow down the overall design cycle.

[0004] An existing method for rapid optimization of diverter holes in hollow profile extrusion dies (CN117521430B) involves initially designing the hollow profile extrusion die, establishing the corresponding current field geometry model, setting the current simulation parameters and boundary conditions for the current field geometry model, setting the conductivity of the diverter hole region, performing current simulation calculations and result analysis within the current field geometry model, optimizing the diverter hole region, and finally performing flow analysis during the extrusion process and determining the working zone length. However, this method can only optimize diverter holes for one profile die.

[0005] An existing knowledge graph-based intelligent design method and system for extrusion dies (CN116167180A) includes associated information visualization, semantic retrieval and graph matching, a question-answering system, and intelligent recommendation. Combined with the knowledge graph, the system associates and organizes product information, die design information, die simulation information, die trial information, and die production information in the extrusion die design process and visualizes them. Through semantic retrieval and intelligent recommendation, the system effectively retrieves and reuses extrusion die design data, information, and knowledge. However, this method can only provide existing design solutions that are similar to new products. Summary of the Invention

[0006] To address the aforementioned technical issues, the present invention provides an intelligent prediction method for diverter holes in aluminum alloy extrusion dies. This method utilizes the correspondence between profiles and diverter holes to construct an intelligent prediction model for diverter hole structure. This model can rapidly generate diverter hole images for the diverter extrusion die required for the profile, thereby shortening the die design cycle.

[0007] The present invention is achieved through at least one of the following technical solutions.

[0008] An intelligent prediction method for diversion holes of an aluminum alloy extrusion die comprises the following steps:

[0009] (1) Preprocess the profile cross-section image to be predicted;

[0010] (2) The preprocessed image is input into the trained intelligent prediction model of the diversion hole of the extrusion die to generate an image of the diversion hole corresponding to the profile.

[0011] Furthermore, the pre-processing is to convert the profile cross-section image to be predicted into a unified format, and convert it into a unified size after trimming:

[0012] Image content: only contains the profile section outline and bar information, with a white background and a black profile section;

[0013] Image format: png format, grayscale image;

[0014] Image trimming: Determine the smallest circumscribed square of the image and perform cropping based on the boundary;

[0015] Image size: 640×640 pixels.

[0016] Furthermore, the extrusion die diversion hole intelligent prediction model is an image-to-image conversion model based on a conditional generative adversarial network, namely the Pix2pix model. The Pix2pix model includes a generator G and a discriminator D. The goal of the generator G is to generate an output as close to the real image as possible, while the goal of the discriminator D is to distinguish the generated image from the real image. The optimal solution G of the generator G is * It can be expressed as:

[0017]

[0018] in represents the adversarial loss function, represents the absolute value loss function, λ represents the weight parameter, Represents mathematical expectation, x represents the input image, y represents the target image corresponding to x, z represents the image generated by the generator, and ||.||1 represents the L1 norm.

[0019] Furthermore, the training of the intelligent prediction model for the diversion holes of the extrusion die includes the following steps:

[0020] S1. Collecting extrusion die model and profile data, including the profile and diversion hole diagram of the diversion extrusion die design drawing, wherein the number of diversion holes is four and the diversion holes are axisymmetric; the extrusion die model and profile data are in dwg format;

[0021] S2. Organize the extrusion die model and profile data, retain the effective information of the profile and diversion hole graphics, and perform data enhancement;

[0022] S3, establishing image set profiles and corresponding diversion hole data sets;

[0023] S4. Use the image set profile and the corresponding diversion hole data set to train the extrusion die diversion hole intelligent prediction model.

[0024] Furthermore, the extrusion die model and profile data are sorted out, including:

[0025] Delete redundant graphic information, including centerline, annotation and lower diversion hole;

[0026] Supplement bar cross-section information: The bar cross-section information is obtained by reading the record in the drawing detail list, and the bar cross-section is drawn with the center of the diversion hole as the center of the circle;

[0027] Filling profiles and diversion hole sections: Use the filling function of AutoCAD software to fill the profiles and diversion holes with solids, and set the transparency of the diversion holes;

[0028] The AutoCAD software outputs png format images; the images of profiles and bars are saved as the first image set A = {a1, a2, ..., a n}, a n represents the nth image in the first image set A; the images of the profiles, bars and diversion holes are saved as the second image set B = {b1, b2, ..., b n}, b n Represents the nth image in the second image set B; copy the first image set A and save it as the third image set C; copy the second image set B and save it as the fourth image set D;

[0029] Perform batch trimming on the images: trim the first image set A, the second image set B, the third image set C, and the fourth image set D using the cv2 library, and scale them to 640*640 pixels.

[0030] The data enhancement is as follows: performing data enhancement on the first image set A and the second image set B, and performing left-right flipping and rotation operations on the scaled images.

[0031] Furthermore, the image set profile and the corresponding diversion hole data set are established, including:

[0032] Establish a complete profile data set: the first image set A and the second image set B after data enhancement are called by the PIL library, according to the corresponding relationship, with b i On the left, a i Splice in the right way, where b i represents the i-th picture in the image set B, a i It represents the i-th image in the first image set A. The pixel size of the spliced image is 1280*640. The spliced image is used as the dataset of the complete profile corresponding to the four diversion holes;

[0033] Create a segmented profile dataset: By calling the PIL library, rotate the images of the third image set C and the fourth image set D 45 degrees clockwise, split the rotated images into four images of equal length and width in the form of four quadrants, and rotate the split images according to the following corresponding relationship:

[0034] First quadrant: no rotation;

[0035] Second quadrant: rotate 90 degrees clockwise;

[0036] The third quadrant: rotate 180 degrees clockwise;

[0037] Fourth quadrant: rotate 270 degrees clockwise;

[0038] The segmented and rotated image is called by the PIL library, and according to the corresponding relationship, d i On the left, c i Splice in the right way, where d i represents the i-th picture in the fourth image set D, c i It represents the i-th picture in the third image set C. The size of the spliced image is 640*320 pixels. The spliced image serves as the dataset of the partial profile corresponding to a single diversion hole after segmentation.

[0039] Furthermore, the complete profile dataset and the segmented profile dataset were used to train the extrusion die diversion hole intelligent prediction model respectively, and two extrusion die diversion hole intelligent prediction models were obtained.

[0040] Furthermore, when training the intelligent prediction model for the diversion holes of the extrusion die, the complete profile dataset and the segmented profile dataset are divided into a training set, a test set, and a validation set in proportion.

[0041] Furthermore, the termination condition for the training of the intelligent prediction model for the diversion hole of the extrusion die is that when the loss curves of the discriminator D and the generator G tend to converge and the output probabilities of the discriminator D for the real image y and the generated image z are close to the threshold, the training stops and the model is saved as a pth file.

[0042] A computer device of the present invention includes: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, the intelligent prediction method for the diversion hole of the aluminum alloy extrusion die is implemented.

[0043] Place the prepared profile cross-section images in a designated folder. If batch prediction is required, place multiple images in the same folder. Input the preprocessed images into the model to generate images of the manifold holes corresponding to the profile. Save the generated manifold hole images to the designated output folder. Use the output images as a reference to design the shape, position, and size of the manifold holes.

[0044] Compared with the existing technology, the beneficial effects of the present invention are:

[0045] The method of the present invention utilizes the corresponding relationship between the profile and the diversion hole and adopts the Pix2pix model to quickly generate the diversion hole image of the diversion extrusion die required for the profile, thereby shortening the die design cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of the intelligent prediction model for the diversion holes of aluminum alloy extrusion dies based on the Pix2pix network;

[0047] Figure 2 This is a before-and-after comparison of the pictures after the trimming operation in the embodiment;

[0048] Figure 3 is a screenshot of the profile and diverter hole of the extrusion die design drawing in the embodiment;

[0049] Figure 4 This is a screenshot of the extrusion die design drawing after the profile and diverter hole are supplemented with the bar material information;

[0050] Figure 5 This is a screenshot of the profile and diversion hole in the embodiment after adding entity information;

[0051] Figure 6 This is the output diagram of the profile + bar, profile + bar + diversion hole in the embodiment;

[0052] Figure 7 is a graph after data enhancement in the embodiment;

[0053] Figure 8 This is the data set of the four diversion holes corresponding to the complete profile in the embodiment;

[0054] Figure 9 This is a diagram after segmentation in the embodiment;

[0055] Figure 10 This is the data set of a single diversion hole corresponding to some profiles in the embodiment

[0056] Figure 11 It is a model test case diagram in the embodiment. DETAILED DESCRIPTION

[0057] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be described in further detail below with reference to embodiments and accompanying drawings.

[0058] An intelligent prediction method for diversion holes in aluminum alloy extrusion dies, such as Figure 1 As shown, the following steps are included:

[0059] (1) Preprocess the profile cross-section image to be predicted;

[0060] (2) The preprocessed image is input into the trained intelligent prediction model of the diversion hole of the extrusion die to generate an image of the diversion hole corresponding to the profile.

[0061] As an embodiment, the preprocessing is to convert the profile cross-section image to be predicted into a unified format, and convert it into a unified size after trimming:

[0062] Image content: Contains only the profile section outline and bar information, with a white background and a black profile section.

[0063] Image format: png format, grayscale image;

[0064] Image trimming: Determine the smallest circumscribed square of the image and use its boundaries as a reference for cropping. Compare the image before and after trimming. Figure 2 As shown in (a) and (b);

[0065] Image size: 640×640 pixels;

[0066] The extrusion die diversion hole intelligent prediction model is an image-to-image conversion model based on a conditional generative adversarial network, or Pix2pix. The Pix2pix model consists of a generator (G) and a discriminator (D). The goal of the generator (G) is to produce outputs that are as close to the real image as possible, while the goal of the discriminator (D) is to distinguish the generated image from the real image. The generator typically uses a U-Net architecture, which effectively preserves image details while generating high-quality target images. The discriminator typically uses a PatchGAN architecture, which extracts features from the input image and determines authenticity.

[0067] The optimal solution G of generator G * It can be expressed as:

[0068]

[0069]

[0070] in represents the adversarial loss function, represents the absolute value loss function, λ represents the weight parameter, represents mathematical expectation, x represents the input image, y represents the target image corresponding to x, z represents the image generated by the generator, and ||.||1 represents the L1 norm;

[0071] The training of the extrusion die diversion hole intelligent prediction model includes the following steps:

[0072] S1. Collect the extrusion die model and profile data. The extrusion die model and profile data include the profile and diversion hole diagram of the diversion extrusion die design drawing. The number of diversion holes is four, and the diversion holes are axisymmetric. The extrusion die model and profile data are in dwg format. The profile and diversion holes in the design drawing are as follows: Figure 3 As shown;

[0073] S2. Organize the extrusion die model and profile data, retain the effective information of the profile and diversion hole graphics, and perform data enhancement;

[0074] S3, establishing image set profiles and corresponding diversion hole data sets;

[0075] S4. Training the intelligent prediction model of the diversion hole of the extrusion die.

[0076] Arranging the extrusion die model and profile data in step S2 includes:

[0077] Delete redundant graphic information, including centerline, annotation and lower diversion hole;

[0078] Supplementary bar cross-section information: The bar cross-section information is obtained by reading the drawing detail list record, and the bar cross-section is drawn with the center of the diversion hole as the center of the circle, such as Figure 4 As shown;

[0079] Fill the profile and diversion hole section: Use the fill function of AutoCAD software to fill the profile and diversion hole with entities, and set the transparency of the diversion hole, such as Figure 5 As shown;

[0080] In step S2, effective information of the profile and diversion hole pattern is retained, and data enhancement is performed, including:

[0081] AutoCAD software outputs png format images, including images of profiles and bars, and saves them as image set A = {a1, a2, ..., a n}, a n represents the nth image in the first image set A, such as Figure 6 (a); the image of the profile + bar + diversion hole is saved as the second image set B = {b1, b2, ..., bn}, b n represents the nth image in the image set B, such as Figure 6 As shown in (b); copy the first image set A and save it as the third image set C; copy the second image set B and save it as the fourth image set D.

[0082] Batch trimming of images: trim the first image set A, the second image set B, the third image set C, and the fourth image set D using the cv2 library, and scale them to 640*640 pixels.

[0083] Data enhancement is to perform data enhancement on the first image set A and the second image set B, and perform left-right flipping and rotation operations on the scaled images, such as Figure 7 shown.

[0084] The image set profiles and the corresponding diversion hole datasets established in S3 include:

[0085] Establish a complete profile data set: the first image set A and the second image set B after data enhancement are called by the PIL library, according to the corresponding relationship, with b i On the left, a i Splice in the right way, where b i represents the i-th picture in the image set B, a i It represents the i-th picture in the image set A. The pixel size of the spliced image is 1280*640. The spliced image is a complete profile corresponding to the dataset of four diversion holes, such as Figure 8 shown.

[0086] Establish a segmented profile dataset: By calling the PIL library, the images of the third image set C and the fourth image set D are rotated 45 degrees clockwise, and the rotated images are divided into four images of equal length and width in the form of four quadrants. The divided images are rotated according to the following corresponding relationship, such as Figure 9 As shown:

[0087] First quadrant: no rotation;

[0088] Second quadrant: rotate 90 degrees clockwise;

[0089] The third quadrant: rotate 180 degrees clockwise;

[0090] Fourth quadrant: rotate 270 degrees clockwise;

[0091] The segmented and rotated image is called by the PIL library, and according to the corresponding relationship, d i On the left, c i Splice in the right way, where d i represents the i-th picture in the fourth image set D, c iIt represents the i-th picture in the third image set C. The size of the spliced image is 640*320 pixels. The spliced image is used as the dataset of the segmented partial profile corresponding to a single diversion hole, such as Figure 10 shown.

[0092] Step S4 trains the intelligent prediction model for the diverter holes of the extrusion die. There are two types of intelligent prediction models for the diverter holes of the extrusion die. One is trained with the complete profile data set as the data set, with the aim of directly obtaining images of the four diverter holes corresponding to the complete profile; the other is trained with the segmented profile data set as the data set, with the aim of first obtaining images of single diverter holes corresponding to partial profiles, and then splicing them to obtain images of the four diverter holes corresponding to the complete profile.

[0093] As an embodiment, when training the intelligent prediction model for the diversion holes of the extrusion die, the complete profile dataset and the segmented profile dataset are divided into a training set, a test set, and a validation set in a ratio of 0.80:0.10:0.10. Training set: used for the training process of the model, accounting for 80% of the total dataset. The training set contains a large number of image pairs, which are used to train the parameters of the generator and the discriminator. Test set: used to evaluate the performance of the model on unseen data, accounting for 10% of the total dataset. The test set is used to verify the generalization ability and prediction accuracy of the model. Validation set: used to adjust the hyperparameters of the model and monitor the training process of the model, accounting for 10% of the total dataset. The validation set is used to evaluate the performance of the model in real time during the training process to prevent overfitting.

[0094] The termination condition of training is that the loss curves of the discriminator D and the generator G tend to converge and the output probability of the discriminator D for the real image y and the generated image z are close to 0.5. The training stops and the model is saved as a pth file.

[0095] Place the prepared profile cross-section images in the specified folder. If batch prediction is required, place multiple images in the same folder. Input the pre-processed images into the model to generate images of the diversion holes corresponding to the profiles, and save the generated diversion hole images to the specified output folder. Figure 11 The figure shows a model test case in this embodiment. The shape, position and size of the diversion hole can be designed with reference to the output image.

[0096] Although the embodiments disclosed herein are as described above, the contents described herein are merely embodiments for facilitating understanding of the present invention and are not intended to limit the present invention. Any person skilled in the art may make any modifications and variations in the form and details of the embodiments without departing from the spirit and scope of the present invention. However, the scope of protection of the present invention shall remain subject to the scope defined by the appended claims.

Claims

1. An intelligent prediction method for diversion holes in aluminum alloy extrusion dies, characterized in that: The following steps are involved: (1) Preprocess the profile cross-section image to be predicted; (2) The preprocessed image is input into the trained intelligent prediction model of the diversion hole of the extrusion die to generate an image of the diversion hole corresponding to the profile.

2. The intelligent prediction method for diversion holes of aluminum alloy extrusion dies according to claim 1, characterized in that: The pre-processing is to convert the profile cross-section image to be predicted into a unified format, and convert it into a unified size after trimming: Image content: only contains the profile section outline and bar information, with a white background and a black profile section; Image format: png format, grayscale image; Image trimming: Determine the smallest circumscribed square of the image and perform cropping based on the boundary; Image size: 640×640 pixels.

3. The intelligent prediction method for diversion holes of aluminum alloy extrusion dies according to claim 1, characterized in that: The intelligent prediction model for the diversion hole of the extrusion die is an image-to-image conversion model based on the conditional generative adversarial network, namely the Pix2pix model. The Pix2pix model includes a generator G and a discriminator D. The goal of the generator G is to generate an output as close to the real image as possible, while the goal of the discriminator D is to distinguish the generated image from the real image. The optimal solution G of the generator G is * It can be expressed as: in represents the adversarial loss function, represents the absolute value loss function, λ represents the weight parameter, Represents mathematical expectation, x represents the input image, y represents the target image corresponding to x, z represents the image generated by the generator, and ||.||1 represents the L1 norm.

4. The intelligent prediction method for diversion holes of aluminum alloy extrusion dies according to claim 1, characterized in that: The training of the intelligent prediction model for the diversion holes of the extrusion die includes the following steps: S1. Collecting extrusion die model and profile data, including the profile and diversion hole diagram of the diversion extrusion die design drawing, wherein the number of diversion holes is four and the diversion holes are axisymmetric; the extrusion die model and profile data are in dwg format; S2. Organize the extrusion die model and profile data, retain the effective information of the profile and diversion hole graphics, and perform data enhancement; S3, establishing image set profiles and corresponding diversion hole data sets; S4. Use the image set profile and the corresponding diversion hole data set to train the extrusion die diversion hole intelligent prediction model.

5. The intelligent prediction method for diversion holes of aluminum alloy extrusion dies according to claim 4, characterized in that: Organize the extrusion die model and profile data including: Delete redundant graphic information, including centerline, annotation and lower diversion hole; Supplement bar cross-section information: The bar cross-section information is obtained by reading the detailed record in the drawing, and the bar cross-section is drawn with the center of the diversion hole as the center of the circle; Filling profiles and diversion hole sections: Use the filling function of AutoCAD software to fill the profiles and diversion holes with solids, and set the transparency of the diversion holes; The AutoCAD software outputs png format images; the images of profiles and bars are saved as the first image set A = {a1, a2, ..., a n }, a n represents the nth image in the first image set A; the images of the profiles, bars and diversion holes are saved as the second image set B = {b1, b2, ..., b n }, b n Represents the nth image in the second image set B; copy the first image set A and save it as the third image set C; copy the second image set B and save it as the fourth image set D; Perform batch trimming on the images: trim the first image set A, the second image set B, the third image set C, and the fourth image set D using the cv2 library, and scale them to 640*640 pixels. The data enhancement is as follows: performing data enhancement on the first image set A and the second image set B, and performing left-right flipping and rotation operations on the scaled images.

6. The intelligent prediction method for diversion holes of aluminum alloy extrusion dies according to claim 5, characterized in that: Create image set profiles with corresponding diversion hole datasets, including: Establish a complete profile data set: the first image set A and the second image set B after data enhancement are called by the PIL library, according to the corresponding relationship, with b i On the left, a i Splice in the right way, where b i represents the i-th picture in the image set B, a i It represents the i-th image in the first image set A. The pixel size of the spliced image is 1280*640. The spliced image is used as the dataset of the complete profile corresponding to the four diversion holes; Create a segmented profile dataset: By calling the PIL library, rotate the images of the third image set C and the fourth image set D 45 degrees clockwise, split the rotated images into four images of equal length and width in the form of four quadrants, and rotate the split images according to the following corresponding relationship: First quadrant: no rotation; Second quadrant: rotate 90 degrees clockwise; The third quadrant: rotate 180 degrees clockwise; Fourth quadrant: rotate 270 degrees clockwise; The segmented and rotated image is called by the PIL library, and according to the corresponding relationship, d i On the left, c i Splice in the right way, where d i represents the i-th picture in the fourth image set D, c i It represents the i-th picture in the third image set C. The size of the spliced image is 640*320 pixels. The spliced image serves as the dataset of the partial profile corresponding to a single diversion hole after segmentation.

7. The intelligent prediction method for diversion holes of aluminum alloy extrusion dies according to claim 6, characterized in that: The complete profile dataset and the segmented profile dataset were used to train the intelligent prediction model for the extrusion die diverter holes, respectively, and two intelligent prediction models for the extrusion die diverter holes were obtained.

8. The intelligent prediction method for diversion holes of aluminum alloy extrusion dies according to claim 7, characterized in that: When training the intelligent prediction model for the diversion holes of extrusion dies, the complete profile dataset and the segmented profile dataset are divided into training set, test set, and validation set in proportion.

9. The intelligent prediction method for diversion holes of aluminum alloy extrusion dies according to claim 4, characterized in that: The termination condition for the training of the intelligent prediction model for the diversion hole of the extrusion die is that when the loss curves of the discriminator D and the generator G tend to converge and the output probabilities of the discriminator D for the real image y and the generated image z are close to the threshold, the training stops and the model is saved as a pth file.

10. A computer device, characterized in that: The invention comprises: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, an intelligent prediction method for diversion holes of an aluminum alloy extrusion die as described in any one of claims 1 to 9 is implemented.

Images