Upsetting process and system for aircraft engines

Abstract

Disclosed is a top forging process and system for aircraft engines. First, a cylindrical billet is cut with a band saw, then peeled, flattened by 2 mm at both ends, and rounded to obtain a pre-treated billet. The pre-treated billet is then heated in a medium-temperature electric furnace to obtain a heated billet. The heated billet is then forged for the first time on a flat forging machine to obtain a primary forging billet. The first forged billet is then forged for the second time on a flat forging machine to obtain a secondary forging billet. The secondary forging billet is then formed on a flat forging machine to obtain a formed billet. The formed billet is then machined and polished to remove burrs from the die surface. The billet is then polished with a grinding wheel and then polished with a polishing machine to obtain a polished billet. Finally, the polished billet is inspected by coloring, and manufacturing is completed after passing the inspection. In this way, a qualified billet is obtained.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection, and more specifically, to an upset forging process and system for an aircraft engine. Background Art

[0002] Most wrought aluminum alloys have good forgeability and can be used to produce forgings of various shapes and types. Aluminum alloy forgings can be produced using various existing forging methods, including open die forging, upset forging, roll forging, and rolling. Aluminum alloys are primarily used in aircraft engines to manufacture intake casings, straightening blades, and clamps.

[0003] The traditional preparation method of medium-sized aluminum blades is the traditional top forging method. The billet making process does not allow burrs, burrs, etc. to appear. The top forging process is: cutting + chamfering + heating + gathering one + grinding to remove damage + heating + gathering two + grinding to remove damage + heating + forming.

[0004] This process has the following disadvantages:

[0005] 1. When using this process to make blanks, the punch part of each gathering step will have problems such as eccentricity, bending, and incompleteness. The final blank surface will be wrinkled or folded. The blank defects will be carried into the final blade, which may easily cause the blades to be scrapped in batches.

[0006] 2. When using this process to manufacture blanks, repeated heating may be required after each upsetting step before the next upsetting step can be carried out, resulting in a long production cycle and high production costs;

[0007] 3. The billet formed by this process is easy to be covered by surface cracks during grinding because the aluminum alloy is soft. Repeated coloring inspection and repeated grinding are required during grinding. The auxiliary process takes a long time, which seriously affects the product quality.

[0008] Therefore, an optimized top forging billet making scheme is expected. Summary of the Invention

[0009] In view of this, the present disclosure proposes an upset forging process and system for an aero-engine, which can produce products of qualified quality.

[0010] According to one aspect of the present disclosure, there is provided an upset forging process for an aircraft engine, comprising:

[0011] A cylindrical blank of 2A70 material was cut with a band saw, and the surface was peeled to 47±0.2, both end surfaces were flattened by 2 mm, and the corners were rounded by R5 to obtain a pre-processed blank, wherein the cylindrical blank had a diameter of 52 mm and a length of 716 mm;

[0012] Heating the pretreated blank in a medium-temperature electric furnace to obtain a heated blank, wherein the heating temperature is 460±° C. and the holding time is greater than 105 minutes;

[0013] The heated blank is forged for the first time by an 800-ton or 1250-ton flat forging machine to obtain a primary forging blank;

[0014] The first forged blank is forged for a second time using an 800-ton or 1250-ton flat forging machine to obtain a secondary forged blank;

[0015] The secondary forging blank is formed by an 800-ton or 1250-ton flat forging machine to obtain a formed blank;

[0016] The formed blank is subjected to mechanical processing and grinding to remove burrs on the door mold surface, and the burrs are ground off by a grinding wheel machine, and then the surface is polished by a polishing machine to obtain a polished blank; and

[0017] The polished blank is inspected by coloring to check whether its surface quality is qualified. After the inspection is qualified, the manufacturing is completed.

[0018] In a possible implementation, the polished blank is inspected by coloring to check whether its surface quality is qualified. After the inspection is qualified, the manufacturing is completed, including:

[0019] Collecting a surface colored image of the blank after coloring;

[0020] Performing image enhancement on the surface-shaded image using a CLAHE algorithm to obtain an enhanced surface-shaded image;

[0021] Passing the enhanced surface coloration image through a convolutional neural network model as a feature extractor to obtain a surface coloration feature map;

[0022] Performing non-negative matrix decomposition on a feature matrix along a channel dimension of the surface shading feature map to obtain a sparse surface shading feature map;

[0023] Passing the sparse surface coloring feature map through a channel attention module to obtain a classification feature map; and

[0024] The classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the surface quality of the blank meets a predetermined standard.

[0025] In one possible implementation, the enhanced surface coloration image is passed through a convolutional neural network model as a feature extractor to obtain a surface coloration feature map, including:

[0026] Each layer of the convolutional neural network model as a feature extractor performs the following on the input data in the forward pass of the layer:

[0027] Perform convolution on the input data to obtain a convolution feature map;

[0028] Performing mean pooling on the convolution feature map to obtain a pooled feature map; and

[0029] Performing nonlinear activation on the pooled feature map to obtain an activated feature map;

[0030] The output of the last layer of the convolutional neural network model serving as a feature extractor is the surface coloring feature map, and the input of the first layer of the convolutional neural network model serving as a feature extractor is the enhanced surface coloring image.

[0031] In one possible implementation, the sparse surface coloring feature map is passed through a channel attention module to obtain a classification feature map, including:

[0032] Performing explicit spatial encoding on the sparse surface shading feature map using the channel attention module to obtain a sparse surface shading correlation feature map;

[0033] Calculating the global mean of each feature matrix along the channel dimension of the sparse surface shading correlation feature map to obtain a channel feature vector;

[0034] Input the channel feature vector into the Sigmoid activation function to obtain a channel attention weighted feature vector;

[0035] Based on the autocovariance matrix of the channel attention weighted eigenvector, correcting the eigenvalues ​​of each position in the channel attention weighted eigenvector to obtain an optimized channel attention weighted eigenvector; and

[0036] Each feature matrix along the channel dimension of the sparse surface shading feature map is weighted using the eigenvalue of each position in the optimized channel attention weighted feature vector as a weight to obtain the classification feature map.

[0037] In a possible implementation, the classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the surface quality of the blank meets a predetermined standard, including:

[0038] Expanding the classification feature map into a classification feature vector according to row vectors or column vectors;

[0039] Performing full connection encoding on the classification feature vector using the fully connected layer of the classifier to obtain an encoded classification feature vector; and

[0040] The encoded classification feature vector is input into the Softmax classification function of the classifier to obtain the classification result.

[0041] In a possible implementation, the method further includes a training step of training the convolutional neural network model serving as a feature extractor, the channel attention module, and the classifier.

[0042] In a possible implementation, the training step includes:

[0043] Acquiring training data, wherein the training data includes a colored training surface image of the blank after coloring, and a true value of whether the surface quality of the blank meets a predetermined standard;

[0044] Performing image enhancement on the training surface shading image using the CLAHE algorithm to obtain a training enhanced surface shading image;

[0045] Passing the training enhanced surface coloring image through the convolutional neural network model as a feature extractor to obtain a training surface coloring feature map;

[0046] performing non-negative matrix decomposition on a feature matrix along a channel dimension of the training surface shading feature map to obtain a training sparse surface shading feature map, and calculating a pseudo-cycle difference penalty factor based on the training sparse surface shading feature map;

[0047] Passing the training sparse surface coloring feature map through the channel attention module to obtain a training classification feature map;

[0048] Passing the training classification feature map through a classifier to obtain a classification loss function value; and

[0049] The convolutional neural network model as a feature extractor, the channel attention module and the classifier are trained using the weighted sum of the classification loss function value and the pseudo-cycle difference penalty factor as the loss function value.

[0050] In one possible implementation, performing non-negative matrix decomposition on a feature matrix along a channel dimension of the training surface shading feature map to obtain a training sparse surface shading feature map, and calculating a pseudo-cycle difference penalty factor based on the training sparse surface shading feature map includes:

[0051] Performing non-negative matrix decomposition on a feature matrix along a channel dimension of the training surface shading feature map to obtain a plurality of first sub-feature matrices and a plurality of second sub-feature matrices;

[0052] Expand each of the first sub-eigenmatrix and the second sub-eigenmatrix into a first sub-eigenvector and a second sub-eigenvector; and

[0053] The pseudo-cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector is calculated.

[0054] In a possible implementation, calculating the pseudo-cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector includes:

[0055] Calculate the pseudo-cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector using the following factor calculation formula;

[0056] The factor calculation formula is:

[0057]

[0058] Among them, V 1i is the first sub-eigenvector, V 2i is the second sub-eigenvector, D(V 1i ,V 2i ) is the distance matrix between the first sub-eigenvector and the second sub-eigenvector, ||·|| F represents the Frobenius norm of the matrix, L is the length of the eigenvector, d(V 1i ,V 2i ) is the distance between the first sub-eigenvector and the second sub-eigenvector, ||·||2 is the two-norm of the vector, log represents the logarithmic function with base 2, and ⊕ represent vector subtraction and vector addition respectively, and α and β are weighted hyperparameters, is the pseudo cycle difference penalty factor.

[0059] According to another aspect of the present disclosure, there is provided an upset forging system for an aircraft engine, comprising:

[0060] A pre-processing module is used to cut a cylindrical billet of 2A70 material using a band saw, and peel it to 47±0.2, flatten the two end surfaces by 2mm, and round the corners by R5 to obtain a pre-processed billet, wherein the cylindrical billet has a diameter of 52mm and a length of 716mm;

[0061] A medium-temperature heating module, used for heating the pretreated blank through a medium-temperature electric furnace to obtain a heated blank, wherein the heating temperature is 460±°C and the holding time is greater than 105 minutes;

[0062] A primary forging module, used for forging the heated blank for the first time through an 800-ton or 1250-ton flat forging machine to obtain a primary forging blank;

[0063] A secondary forging module, configured to subject the first forged blank to a second forging process using an 800-ton or 1250-ton flat forging machine to obtain a secondary forging blank;

[0064] A forming module, used for forming the secondary forging blank through an 800-ton or 1250-ton flat forging machine to obtain a formed blank;

[0065] A polishing module, for removing burrs from the door mold surface by machining and polishing the formed blank, polishing the burrs with a grinding wheel, and polishing the surface with a polishing machine to obtain a polished blank; and

[0066] The surface quality inspection module is used to inspect the polished blank by coloring to check whether its surface quality is qualified. After the inspection is qualified, the manufacturing is completed.

[0067] According to an embodiment of the present disclosure, a cylindrical blank is first cut with a band saw, and the skin is peeled, the two end faces are flattened by 2 mm, and the corners are rounded to obtain a pre-treated blank. Then, the pre-treated blank is heated in a medium-temperature electric furnace to obtain a heated blank. Then, the heated blank is forged for the first time by a flat forging machine to obtain a primary forging blank. Then, the first forged blank is forged for the second time by a flat forging machine to obtain a secondary forging blank. Then, the secondary forging blank is formed by a flat forging machine to obtain a formed blank. Then, the formed blank is machined and polished to remove burrs on the door die surface, and is polished with a grinding wheel and a polishing machine to obtain a polished blank. Finally, the polished blank is inspected by a coloring method, and the manufacturing is completed after passing the inspection. In this way, a qualified blank can be obtained.

[0068] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0070] Figure 1 A flow chart illustrating an upset forging process for an aircraft engine according to an embodiment of the present disclosure is shown.

[0071] Figure 2 A diagram illustrating an application scenario of sub-step S170 of the upset forging process of an aerospace engine according to an embodiment of the present disclosure is shown.

[0072] Figure 3 A flowchart illustrating sub-step S170 of the upset forging process of an aircraft engine according to an embodiment of the present disclosure is shown.

[0073] Figure 4 A schematic diagram illustrating the architecture of sub-step S170 of the upset forging process of an aircraft engine according to an embodiment of the present disclosure.

[0074] Figure 5 A flowchart illustrating sub-step S175 of the upset forging process of an aircraft engine according to an embodiment of the present disclosure is shown.

[0075] Figure 6 A flowchart illustrating sub-step S176 of the upset forging process of an aircraft engine according to an embodiment of the present disclosure is shown.

[0076] Figure 7 A flowchart illustrating a training step further included in the upset forging process of an aircraft engine according to an embodiment of the present disclosure.

[0077] Figure 8 A flowchart illustrating sub-step S240 of the upset forging process of an aircraft engine according to an embodiment of the present disclosure is shown.

[0078] Figure 9 A block diagram of an upset forging system for an aircraft engine according to an embodiment of the present disclosure is shown.

[0079] Figure 10 A diagram illustrating the upset forging process steps according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0080] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative efforts also fall within the scope of protection of this application.

[0081] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0082] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0083] Various exemplary embodiments, features, and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0084] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0085] In addition, numerous specific details are provided in the following detailed description to better illustrate the present disclosure. Those skilled in the art will appreciate that the present disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main points of the present disclosure.

[0086] This application provides a top forging process for an aero-engine, such as Figure 1 As shown, the specific steps include: S110, cutting a cylindrical billet of 2A70 material with a band saw, and peeling it to 47±0.2, flattening the two end surfaces by 2mm, and chamfering R5 to obtain a pre-treated billet, wherein the diameter of the cylindrical billet is 52mm and the length is 716mm; S120, heating the pre-treated billet in a medium-temperature electric furnace to obtain a heated billet, wherein the heating temperature is 460±℃ and the holding time is greater than 105 minutes; S130, forging the heated billet for the first time with an 800-ton or 1250-ton flat forging machine to obtain a primary forging billet; S1 40. The first forged blank is forged a second time by an 800-ton or 1250-ton flat forging machine to obtain a secondary forged blank; S150. The secondary forged blank is formed by an 800-ton or 1250-ton flat forging machine to obtain a formed blank; S160. The formed blank is subjected to machining and grinding to remove burrs on the door die surface, and the burrs are ground off by a grinding wheel machine, and then the surface is polished by a polishing machine to obtain a polished blank; and, S170, the polished blank is inspected by a coloring method to check whether its surface quality is qualified, and the manufacturing is completed after the inspection is qualified.

[0087] In step S170, traditional inspection methods typically involve visual inspection or simple tool inspection. This method has the following problems: the inspection results are easily affected by the operator's subjective factors, resulting in unstable inspection results, low inspection efficiency, and a large amount of time and manpower required for inspection.

[0088] In this regard, the technical concept of this application is to realize automated inspection of the surface quality of the blank after grinding based on deep learning and artificial intelligence technology to improve inspection efficiency and accuracy.

[0089] Specifically, given that manual visual inspection is often subjective and unstable, and difficult to meet the requirements of high-quality aircraft engines, the technical solution of this application first collects a surface coloring image of the colored blank to replace the human eye to observe the surface features of the colored blank. Typically, after dyeing with a dye, defects, cracks, holes, and other features on the blank surface are made more obvious, making them easier to detect. In a specific example of this application, a camera or digital camera is used to capture the surface of the colored blank to obtain the surface coloring image.

[0090] Then, the CLAHE algorithm is used to perform image enhancement on the surface colored image to obtain an enhanced surface colored image. The CLAHE algorithm is an adaptive histogram equalization algorithm that can effectively enhance the contrast and details of the image while avoiding the problem of over-enhancement of image noise in traditional histogram equalization algorithms. In a specific implementation, the CLAHE algorithm first divides the image into several small areas, then performs histogram equalization on the pixels in each small area, and finally splices the small areas together to form an enhanced image. In this way, the CLAHE algorithm can adaptively adjust the parameters of the histogram equalization according to the local features of the image, so that the contrast and details of the image are better enhanced. In this way, defects, cracks, holes and other features on the surface of the blank will be more prominent.

[0091] Considering the widespread application of convolutional neural networks in image processing, which can effectively extract image features, the technical solution of this application contemplates the use of convolutional neural networks for image feature extraction. Specifically, the enhanced surface coloring image is passed through a convolutional neural network model as a feature extractor to obtain a surface coloring feature map.

[0092] Specifically, a convolutional neural network is a deep learning model that extracts features and reduces dimensionality from images through multiple layers of convolution and pooling operations, ultimately outputting a high-dimensional feature map, the surface coloring feature map. This surface coloring feature map contains important characteristic information about the blank surface, facilitating subsequent surface quality inspection. In practical implementations, pre-trained convolutional neural network models such as VGG and ResNet can be used.

[0093] Then, the feature matrix along the channel dimension of the surface coloring feature map is subjected to non-negative matrix decomposition to obtain a sparse surface coloring feature map. Specifically, first, each feature matrix along the channel dimension of the surface coloring feature map is subjected to non-negative matrix decomposition, which can decompose the original data matrix into two non-negative matrices U and V. The decomposed matrix only contains non-negative factors, and U can well maintain the local structure of the data. Subsequently, the two non-negative matrices obtained by non-negative matrix decomposition of each feature matrix are arranged along the channel dimension to obtain the sparse surface coloring feature map. The advantage of doing so is that redundant information can be removed and the robustness and interpretability of the feature are improved.

[0094] Furthermore, the sparse surface coloring feature map is passed through a channel attention module to obtain a classification feature map. The channel attention module can adjust the weight of each channel of the sparse surface coloring feature map so that the network can pay more attention to important feature channels. In the surface quality detection task, different surface coloring feature channels may contribute differently to the classification task. Some channels may not contribute to the classification task or have a low contribution value. The channel attention module can enhance or suppress their contribution by applying different weights to each feature channel, thereby improving the performance of the classifier. In this way, the classification feature map can better reflect the quality of the blank surface and have higher discrimination ability and robustness.

[0095] The classification feature map is then passed through a classifier to obtain a classification result, which indicates whether the surface quality of the blank meets the predetermined standard. In other words, the classification feature map is input into the classifier, which classifies and determines the surface quality of the blank according to pre-trained classification rules, thereby obtaining a classification result indicating whether the surface quality of the blank meets the predetermined standard.

[0096] In the technical solution of the present application, by performing non-negative matrix decomposition on the feature matrix along the channel dimension of the surface coloring feature map, a representation that maintains the data hiding attribute is obtained while obtaining a first sub-feature matrix and a second sub-feature matrix with high sparsity. However, due to the uneven distribution of image semantics of the enhanced surface coloring image, after the local image-related feature semantics are extracted by the convolutional neural network model as a feature extractor, the feature semantic distribution of each feature matrix of the surface coloring feature map will also show a large unevenness, resulting in an unbalanced distribution of feature distributions between the first sub-feature matrix and the second sub-feature matrix obtained by decomposition, thereby affecting the expression effect of the sparse surface coloring feature map, and also affecting the accuracy of the classification result obtained by the classifier of the classification feature map.

[0097] Therefore, in addition to the classification loss function for the classification feature map, the applicant of this application further expands each of the first sub-feature matrix and the second sub-feature matrix into a first sub-feature vector and a second sub-feature vector, and then introduces the first sub-feature vector, for example, denoted as V 1i and the second sub-eigenvector, for example, denoted as V 2i The pseudo cycle difference penalty factor is used as the loss function, which is specifically expressed as:

[0098]

[0099] D(V 1i ,V 2i ) is the eigenvector V 1i and V 2i The distance matrix between them, that is, the eigenvalue of the (j, k)th position of the distance matrix is ​​the eigenvector V 1i The j-th eigenvalue v 1ij With the eigenvector V 2i The kth eigenvalue v 2ik The distance between them, ||·|| F represents the Frobenius norm of the matrix, L is the length of the eigenvector, d(V 1i ,V 2i ) is the eigenvector V 1i and V 2i , such as the Euclidean distance, ||·||2 is the two-norm of the vector, log represents the logarithm with base 2, and α and β are weighted hyperparameters.

[0100] Here, considering the first sub-eigenvector V 1i and the second sub-eigenvector V 2i The unbalanced distribution between them will lead to abnormal gradient propagation in the model training process based on gradient descent back propagation, thereby forming a pseudo cycle of model parameter update. The pseudo cycle difference penalty factor introduces penalty factors for both spatial and numerical relationships of closely related numerical pairs used to express eigenvalues, so as to treat the pseudo cycle of model parameter update as a real cycle in the model training process of minimizing the loss function, so as to realize the first sub-eigenvector V by means of simulated activation of gradient propagation. 1i and the second sub-eigenvector V 2i The respective feature distributions are progressively coupled, thereby improving the distribution imbalance of the feature distribution between the first sub-feature matrix and the second sub-feature matrix, so as to enhance the expression effect of the sparse surface shading feature map, thereby enhancing the accuracy of the classification result obtained by the classifier of the classification feature map.

[0101] Figure 2The following diagram shows an application scenario of the upset forging process of an aerospace engine according to an embodiment of the present disclosure. Figure 2 As shown, in this application scenario, first, the colored blank (for example, Figure 2 N) shown in FIG (e.g., Figure 2 Then, the surface coloring image is input into a server that is equipped with an aerospace engine upset algorithm (e.g., Figure 2 S) as shown in , wherein the server is capable of processing the surface colored image using the upset forging algorithm of the aircraft engine to obtain a classification result indicating whether the surface quality of the blank meets a predetermined standard.

[0102] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0103] Figure 3 Flowchart showing the upset forging process of an aerospace engine according to an embodiment of the present disclosure. Figure 3 As shown, according to the top forging process of the aero-engine of the embodiment of the present application, the polished blank is inspected by a coloring method to check whether its surface quality is qualified, and the manufacturing is completed after the inspection is qualified, including the steps: S171, collecting the surface coloring image of the colored blank; S172, using the CLAHE algorithm to perform image enhancement on the surface coloring image to obtain an enhanced surface coloring image; S173, passing the enhanced surface coloring image through a convolutional neural network model as a feature extractor to obtain a surface coloring feature map; S174, performing non-negative matrix decomposition on the feature matrix along the channel dimension of the surface coloring feature map to obtain a sparse surface coloring feature map; S175, passing the sparse surface coloring feature map through a channel attention module to obtain a classification feature map; and, S176, passing the classification feature map through a classifier to obtain a classification result, and the classification result is used to indicate whether the surface quality of the blank meets the predetermined standard.

[0104] Figure 4 Schematic diagram of the structure of the upset forging process of the aircraft engine according to the embodiment of the present disclosure is shown. Figure 4As shown in the figure, in the network architecture, first, the surface coloring image of the colored blank is collected; then, the CLAHE algorithm is used to perform image enhancement on the surface coloring image to obtain an enhanced surface coloring image; then, the enhanced surface coloring image is passed through a convolutional neural network model as a feature extractor to obtain a surface coloring feature map; then, the feature matrix of the surface coloring feature map along the channel dimension is subjected to non-negative matrix decomposition to obtain a sparse surface coloring feature map; then, the sparse surface coloring feature map is passed through a channel attention module to obtain a classification feature map; finally, the classification feature map is passed through a classifier to obtain a classification result, which is used to indicate whether the surface quality of the blank meets the predetermined standard.

[0105] More specifically, in step S171, a colored surface image of the colored blank is collected. After being colored with a dye, defects, cracks, holes, and other features on the blank surface are made more obvious, making them easier to detect.

[0106] More specifically, in step S172, the surface-shaded image is enhanced using the CLAHE algorithm to obtain an enhanced surface-shaded image. The CLAHE algorithm is an adaptive histogram equalization algorithm that can effectively enhance image contrast and detail while avoiding the problem of over-enhancing image noise in traditional histogram equalization algorithms.

[0107] More specifically, in step S173, the enhanced surface coloring image is passed through a convolutional neural network model as a feature extractor to obtain a surface coloring feature map. Convolutional neural networks are widely used in the field of image processing and can effectively extract image features. Therefore, convolutional neural networks can be used to extract image features.

[0108] It should be understood that a convolutional neural network is a deep learning model that can extract features and reduce the dimensionality of an image through multiple layers of convolution and pooling operations, ultimately outputting a high-dimensional feature map, namely the surface coloring feature map. The surface coloring feature map contains important characteristic information about the blank surface, which facilitates subsequent surface quality inspection. In specific implementations, pre-trained convolutional neural network models such as VGG and ResNet can be used.

[0109] Accordingly, in one possible implementation, the enhanced surface coloring image is passed through a convolutional neural network model serving as a feature extractor to obtain a surface coloring feature map, comprising: performing the following steps on the input data in the forward pass of the layer: performing convolution processing on the input data to obtain a convolution feature map; performing mean pooling processing on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the convolutional neural network model serving as a feature extractor is the surface coloring feature map, and the input of the first layer of the convolutional neural network model serving as a feature extractor is the enhanced surface coloring image.

[0110] More specifically, in step S174, the feature matrix along the channel dimension of the surface coloring feature map is subjected to non-negative matrix decomposition to obtain a sparse surface coloring feature map. Specifically, first performing non-negative matrix decomposition on each feature matrix along the channel dimension of the surface coloring feature map can decompose the original data matrix into two non-negative matrices U and V. The decomposed matrix only contains non-negative factors, and U can well maintain the local structure of the data. Subsequently, the two non-negative matrices obtained by non-negative matrix decomposition of each feature matrix are arranged along the channel dimension to obtain the sparse surface coloring feature map. The advantage of doing so is that redundant information can be removed and the robustness and interpretability of the feature can be improved.

[0111] More specifically, in step S175, the sparse surface coloring feature map is passed through a channel attention module to obtain a classification feature map. The channel attention module can adjust the weight of each channel of the sparse surface coloring feature map so that the network can pay more attention to important feature channels. In the surface quality inspection task, different surface coloring feature channels may contribute differently to the classification task. Some channels may not contribute to the classification task or have a low contribution value. The channel attention module can enhance or suppress their contribution by applying different weights to each feature channel, thereby improving the performance of the classifier.

[0112] Accordingly, in one possible implementation, Figure 5As shown, the sparse surface shading feature map is passed through a channel attention module to obtain a classification feature map, including: S1751, using the channel attention module to perform explicit spatial encoding on the sparse surface shading feature map to obtain a sparse surface shading association feature map; S1752, calculating the global mean of each feature matrix along the channel dimension of the sparse surface shading association feature map to obtain a channel feature vector; S1753, inputting the channel feature vector into a Sigmoid activation function to obtain a channel attention weighted feature vector; S1754, based on the autocovariance matrix of the channel attention weighted feature vector, correcting the eigenvalues ​​of each position in the channel attention weighted feature vector to obtain an optimized channel attention weighted feature vector; and, S1755, using the eigenvalues ​​of each position in the optimized channel attention weighted feature vector as weights to weight each feature matrix along the channel dimension of the sparse surface shading feature map to obtain the classification feature map.

[0113] More specifically, in step S176, the classification feature map is passed through a classifier to obtain a classification result, which is used to indicate whether the surface quality of the blank meets the predetermined standard. In other words, the classification feature map is input into the classifier, which classifies and determines the surface quality of the blank according to pre-trained classification rules, thereby obtaining a classification result indicating whether the surface quality of the blank meets the predetermined standard.

[0114] That is, in the technical solution of the present application, the labels of the classifier include whether the surface quality of the blank meets the predetermined standard (first label), and whether the surface quality of the blank does not meet the predetermined standard (second label), wherein the classifier determines which classification label the classification feature map belongs to by a soft maximum function. It is worth noting that the first label p1 and the second label p2 here do not contain artificially set concepts. In fact, during the training process, the computer model does not have the concept of "whether the surface quality of the blank meets the predetermined standard". It only has two classification labels and the probability of the output feature under these two classification labels, that is, the sum of p1 and p2 is one. Therefore, the classification result of whether the surface quality of the blank meets the predetermined standard is actually converted from the classification label to a binary classification class probability distribution that conforms to natural laws. In essence, what is used is the physical meaning of the natural probability distribution of the label, rather than the linguistic text meaning of "whether the surface quality of the blank meets the predetermined standard".

[0115] It should be understood that the role of a classifier is to use given categories and known training data to learn classification rules and classifiers, and then classify (or predict) unknown data. Logistic regression and SVM are commonly used to solve binary classification problems. For multi-class classification problems, logistic regression or SVM can also be used, but multiple binary classifications are required to form a multi-class classification. However, this is prone to errors and is inefficient. A commonly used multi-classification method is the Softmax classification function.

[0116] Accordingly, in one possible implementation, Figure 6 As shown, the classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the surface quality of the blank meets a predetermined standard, including: S1761, expanding the classification feature map into a classification feature vector according to a row vector or a column vector; S1762, using the fully connected layer of the classifier to perform full connection encoding on the classification feature vector to obtain an encoded classification feature vector; and, S1763, inputting the encoded classification feature vector into the Softmax classification function of the classifier to obtain the classification result.

[0117] Accordingly, in a possible implementation, the aero-engine upsetting process further includes a training step of training the convolutional neural network model as a feature extractor, the channel attention module, and the classifier. Figure 7 As shown, the training step includes: S210, obtaining training data, the training data including a training surface coloring image of the blank after coloring, and a true value of whether the surface quality of the blank meets the predetermined standard; S220, using the CLAHE algorithm to perform image enhancement on the training surface coloring image to obtain a training enhanced surface coloring image; S230, passing the training enhanced surface coloring image through the convolutional neural network model as a feature extractor to obtain a training surface coloring feature map; S240, performing non-negative matrix multiplication on the feature matrix along the channel dimension of the training surface coloring feature map. The invention relates to a method for training a convolutional neural network (CNN) extractor and a convolutional neural network (CNN) extractor. The method comprises the steps of: decomposing the convolutional neural network (CNN) matrix to obtain a training sparse surface shading feature map, and calculating a pseudo-cycle difference penalty factor based on the training sparse surface shading feature map; passing the training sparse surface shading feature map through the channel attention module to obtain a training classification feature map; passing the training classification feature map through a classifier to obtain a classification loss function value; and, S270, training the convolutional neural network model as a feature extractor, the channel attention module, and the classifier using a weighted sum of the classification loss function value and the pseudo-cycle difference penalty factor as a loss function value.

[0118] Accordingly, in one possible implementation, Figure 8 As shown, non-negative matrix decomposition is performed on the feature matrix along the channel dimension of the training surface shading feature map to obtain a training sparse surface shading feature map, and a pseudo-cycle difference penalty factor is calculated based on the training sparse surface shading feature map, including: S241, non-negative matrix decomposition is performed on the feature matrix along the channel dimension of the training surface shading feature map to obtain multiple first sub-feature matrices and multiple second sub-feature matrices; S242, each of the first sub-feature matrices and the second sub-feature matrices is expanded into a first sub-feature vector and a second sub-feature vector; and, S243, the pseudo-cycle difference penalty factor of the first sub-feature vector and the second sub-feature vector is calculated.

[0119] In the technical solution of the present application, when performing non-negative matrix decomposition on the feature matrix along the channel dimension of the training surface coloring feature map, a representation of the data hiding attribute is obtained while obtaining a first sub-feature matrix and a second sub-feature matrix with high sparsity. However, due to the uneven distribution of image semantics of the training enhanced surface coloring image, after the local image-related feature semantics are extracted by the convolutional neural network model as a feature extractor, the feature semantic distribution of each feature matrix of the training surface coloring feature map will also show a large unevenness, resulting in an unbalanced distribution of feature distributions between the first sub-feature matrix and the second sub-feature matrix obtained by decomposition, thereby affecting the expression effect of the training sparse surface coloring feature map, and also affecting the accuracy of the training classification result obtained by the classifier of the training classification feature map. Therefore, in addition to the classification loss function for the training classification feature map, the applicant of the present application further expands each of the first sub-feature matrix and the second sub-feature matrix into a first sub-feature vector and a second sub-feature vector, and then introduces a pseudo-cyclic difference penalty factor for the first sub-feature vector and the second sub-feature vector as a loss function.

[0120] Accordingly, in one possible implementation, calculating the pseudo-cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector includes: calculating the pseudo-cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector using the following factor calculation formula; wherein the factor calculation formula is:

[0121]

[0122] Among them, V 1i is the first sub-eigenvector, V 2i is the second sub-eigenvector, D(V 1i ,V 2i ) is the distance matrix between the first sub-eigenvector and the second sub-eigenvector, ||·||F represents the Frobenius norm of the matrix, L is the length of the eigenvector, d(V 1i ,V 2i ) is the distance between the first sub-eigenvector and the second sub-eigenvector, ||·||2 is the two-norm of the vector, log represents the logarithmic function with base 2, and ⊕ represent vector subtraction and vector addition respectively, and α and β are weighted hyperparameters, is the pseudo cycle difference penalty factor.

[0123] Taking into account that the unbalanced distribution between the first sub-eigenvector and the second sub-eigenvector will cause gradient propagation anomalies in the model training process based on gradient descent back propagation, thereby forming a pseudo-cycle of model parameter updates, the pseudo-cycle difference penalty factor introduces penalty factors for both the spatial relationship and the numerical relationship of closely associated numerical pairs used to express eigenvalues, so as to regard the pseudo-cycle of model parameter updates as a real cycle in the model training process of minimizing the loss function, and realize the progressive coupling of the feature distributions of the first sub-eigenvector and the second sub-eigenvector through simulated activation of gradient propagation, thereby improving the distribution imbalance of the feature distribution between the first sub-feature matrix and the second sub-feature matrix, so as to improve the expression effect of the training sparse surface shading feature map, thereby improving the accuracy of the training classification results obtained by the classifier of the training classification feature map.

[0124] In summary, based on the top forging process of the aero-engine according to the embodiment of the present application, the surface coloring image of the blank after coloring is first collected, and then the CLAHE algorithm is used to perform image enhancement on the surface coloring image to obtain an enhanced surface coloring image. Then, the enhanced surface coloring image is passed through a convolutional neural network model as a feature extractor to obtain a surface coloring feature map. Then, the feature matrix of the surface coloring feature map along the channel dimension is subjected to non-negative matrix decomposition to obtain a sparse surface coloring feature map. Then, the sparse surface coloring feature map is passed through a channel attention module to obtain a classification feature map. Finally, the classification feature map is passed through a classifier to obtain a classification result indicating whether the surface quality of the blank meets a predetermined standard.

[0125] Figure 9 FIG. 1 is a block diagram of an upset forging system 100 for an aircraft engine according to an embodiment of the present disclosure. Figure 9As shown, the top forging system 100 of the aircraft engine according to the embodiment of the present application includes: a pretreatment module 110, which is used to cut a cylindrical billet of 2A70 material through a band saw, and peel it to 47±0.2, flatten the two end faces by 2mm, and fillet R5 to obtain a pretreated billet, wherein the diameter of the cylindrical billet is 52mm and the length is 716mm; a medium-temperature heating module 120, which is used to heat the pretreated billet through a medium-temperature electric furnace to obtain a heated billet, wherein the heating temperature is 460±℃ and the holding time is greater than 105 minutes; a primary forging module 130, which is used to forge the heated billet for the first time through an 800-ton or 1250-ton flat forging machine to obtain a primary forging making blanks; a secondary forging module 140, used for subjecting the first forged blank to a second forging through an 800-ton or 1250-ton flat forging machine to obtain a secondary forged blank; a forming module 150, used for subjecting the secondary forged blank to a forming through an 800-ton or 1250-ton flat forging machine to obtain a formed blank; a grinding module 160, used for subjecting the formed blank to a machining and grinding method to remove burrs on the door die surface, and to grinding burrs through a grinding wheel machine, and then to polishing the surface through a polishing machine to obtain a ground blank; and a surface quality inspection module 170, used for inspecting the ground blank by a coloring method to inspect whether its surface quality is qualified, and completing the manufacturing after passing the inspection.

[0126] In one possible implementation, in the above-mentioned aircraft engine top forging system 100, the surface quality inspection module 170 includes: an image acquisition unit for acquiring a surface coloring image of the blank after coloring; an image enhancement unit for performing image enhancement on the surface coloring image using the CLAHE algorithm to obtain an enhanced surface coloring image; a feature extraction unit for passing the enhanced surface coloring image through a convolutional neural network model as a feature extractor to obtain a surface coloring feature map; a non-negative matrix decomposition unit for performing non-negative matrix decomposition on the feature matrix along the channel dimension of the surface coloring feature map to obtain a sparse surface coloring feature map; a channel attention encoding unit for passing the sparse surface coloring feature map through a channel attention module to obtain a classification feature map; and a classification unit for passing the classification feature map through a classifier to obtain a classification result, and the classification result is used to indicate whether the surface quality of the blank meets a predetermined standard.

[0127] In one possible implementation, in the above-mentioned aircraft engine top forging system 100, the feature extraction unit is used to: each layer of the convolutional neural network model serving as a feature extractor performs the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a convolution feature map; mean pooling processing on the convolution feature map to obtain a pooled feature map; and nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the convolutional neural network model serving as a feature extractor is the surface shading feature map, and the input of the first layer of the convolutional neural network model serving as a feature extractor is the enhanced surface shading image.

[0128] In one possible implementation, in the above-mentioned aircraft engine top forging system 100, the channel attention encoding unit is used to: use the channel attention module to explicitly spatially encode the sparse surface shading feature map to obtain a sparse surface shading association feature map; calculate the global mean of each feature matrix along the channel dimension of the sparse surface shading association feature map to obtain a channel feature vector; input the channel feature vector into a Sigmoid activation function to obtain a channel attention weighted feature vector; based on the autocovariance matrix of the channel attention weighted feature vector, correct the eigenvalues ​​of each position in the channel attention weighted feature vector to obtain an optimized channel attention weighted feature vector; and, use the eigenvalues ​​of each position in the optimized channel attention weighted feature vector as weights to weight each feature matrix along the channel dimension of the sparse surface shading feature map to obtain the classification feature map.

[0129] In one possible implementation, in the above-mentioned aircraft engine upsetting system 100, the classification unit is used to: expand the classification feature map into a classification feature vector according to a row vector or a column vector; use the fully connected layer of the classifier to fully connect encode the classification feature vector to obtain an encoded classification feature vector; and input the encoded classification feature vector into the Softmax classification function of the classifier to obtain the classification result.

[0130] In a possible implementation, the above-mentioned aircraft engine upset forging system 100 further includes a module for training the convolutional neural network model serving as a feature extractor, the channel attention module, and the classifier. Wherein, the training module is used to: obtain training data, which includes a training surface coloring image of the blank after coloring, and a true value of whether the surface quality of the blank meets the predetermined standard; use the CLAHE algorithm to perform image enhancement on the training surface coloring image to obtain a training enhanced surface coloring image; pass the training enhanced surface coloring image through the convolutional neural network model as a feature extractor to obtain a training surface coloring feature map; perform non-negative matrix decomposition on the feature matrix along the channel dimension of the training surface coloring feature map to obtain a training sparse surface coloring feature map, and calculate a pseudo-cycle difference penalty factor based on the training sparse surface coloring feature map; pass the training sparse surface coloring feature map through the channel attention module to obtain a training classification feature map; pass the training classification feature map through a classifier to obtain a classification loss function value; and use the weighted sum of the classification loss function value and the pseudo-cycle difference penalty factor as the loss function value to train the convolutional neural network model as a feature extractor, the channel attention module and the classifier.

[0131] In one possible implementation, in the above-mentioned aircraft engine upsetting system 100, non-negative matrix decomposition is performed on the feature matrix along the channel dimension of the training surface shading feature map to obtain a training sparse surface shading feature map, and a pseudo-cycle difference penalty factor is calculated based on the training sparse surface shading feature map, including: performing non-negative matrix decomposition on the feature matrix along the channel dimension of the training surface shading feature map to obtain multiple first sub-feature matrices and multiple second sub-feature matrices; expanding each of the first sub-feature matrices and the second sub-feature matrices into a first sub-feature vector and a second sub-feature vector; and calculating the pseudo-cycle difference penalty factor of the first sub-feature vector and the second sub-feature vector.

[0132] In one possible implementation, in the above-mentioned aircraft engine upsetting system 100, calculating the pseudo cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector includes: calculating the pseudo cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector using the following factor calculation formula; wherein the factor calculation formula is:

[0133]

[0134] Among them, V 1i is the first sub-eigenvector, V 2i is the second sub-eigenvector, D(V 1i ,V2i ) is the distance matrix between the first sub-eigenvector and the second sub-eigenvector, ||·|| F represents the Frobenius norm of the matrix, L is the length of the eigenvector, d(V 1i ,V 2i ) is the distance between the first sub-eigenvector and the second sub-eigenvector, ||·||2 is the two-norm of the vector, log represents the logarithmic function with base 2, and ⊕ represent vector subtraction and vector addition respectively, and α and β are weighted hyperparameters, is the pseudo cycle difference penalty factor.

[0135] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the above-mentioned aircraft engine upset forging system 100 have been described in detail above. Figures 1 to 8 The upsetting process of the aircraft engine has been described in detail, and therefore, its repeated description will be omitted.

[0136] As described above, the aircraft engine upsetting system 100 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with an aircraft engine upsetting algorithm. In one possible implementation, the aircraft engine upsetting system 100 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the aircraft engine upsetting system 100 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the aircraft engine upsetting system 100 can also be one of the many hardware modules of the wireless terminal.

[0137] Alternatively, in another example, the aircraft engine's upsetting system 100 and the wireless terminal may also be separate devices, and the aircraft engine's upsetting system 100 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0138] Furthermore, in the upset forging process and system of the aircraft engine, in step S130, an 800-ton or 1250-ton flat forging machine is used for forging. Figure 10 In step S140, an 800-ton or 1250-ton flat forging machine is used for forging. Figure 10 In step S50, forging is performed using an 800-ton or 1250-ton flat forging machine. Figure 10 The molding process is carried out as shown in the figure.

[0139] According to the above embodiment of the present application, the beneficial effects include: using small-sized bar material for upsetting and forming (the aspect ratio of the deformation part is greater than 8), and using three steps for upsetting and forming, such as Figure 10 As shown, each step uses the gathering method with burrs to increase the resistance brought by the burrs to ensure that the two gathering processes are fully filled; relying on the third rule of top forging in the forging process, and using the principle of equal volume, such as Figure 10 As shown, V1≈V3≈V5, V2≈V4≈V6, the positions where the burrs and burrs of the blank are generated are controlled at the same parting surface position, so that the burrs and burrs are uniformly removed, and there is no need to add a burr removal process in the middle of the process, thereby ensuring the quality of the blank; it solves the problems of bending, twisting, and insufficient filling caused by the large aspect ratio of small-sized bars when making medium-sized aluminum alloy blades; it solves the problem that the blanks need to be polished or folded each time they are gathered during the making of medium-sized aluminum alloy blades; and it solves the problem of folding and cracking on the tenon surface and the blade body transition R caused by blank problems during the final forging of medium-sized aluminum alloy blades.

[0140] In an exemplary embodiment, a non-volatile computer-readable storage medium, such as a memory, including computer program instructions is also provided. The computer program instructions are executable by a processing component of an apparatus to perform the above method.

[0141] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0142] A computer-readable storage medium can be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0143] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0144] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, and conventional procedural programming languages ​​such as "C" language or similar programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., utilizing an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions. The electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0145] Various aspects of the present disclosure are described herein with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0146] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more blocks in the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0147] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0148] The flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and the part of the module, program segment or instruction contains one or more executable instructions for realizing the prescribed logical function. In some alternative implementations, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the prescribed function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0149] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. An upset forging process for an aircraft engine, characterized in that: include: A cylindrical blank of 2A70 material was cut with a band saw, and the surface was peeled to 47±0.2, both end surfaces were flattened by 2 mm, and the corners were rounded by R5 to obtain a pre-processed blank, wherein the cylindrical blank had a diameter of 52 mm and a length of 716 mm; Heating the pretreated blank in a medium-temperature electric furnace to obtain a heated blank, wherein the heating temperature is 460±° C. and the holding time is greater than 105 minutes; The heated blank is forged for the first time by an 800-ton or 1250-ton flat forging machine to obtain a primary forging blank; The primary forging blank is subjected to a second forging by an 800-ton or 1250-ton flat forging machine to obtain a secondary forging blank; The secondary forging blank is formed by an 800-ton or 1250-ton flat forging machine to obtain a formed blank; The formed blank is subjected to mechanical processing and grinding to remove burrs on the door mold surface, and the burrs are ground off by a grinding wheel machine, and then the surface is polished by a polishing machine to obtain a polished blank; and The polished blank is inspected by coloring to check whether its surface quality is qualified, and the manufacturing is completed after the inspection is qualified; The polished blank is inspected by coloring to check whether its surface quality is qualified. After the inspection is qualified, the manufacturing is completed, including: Collecting a surface colored image of the blank after coloring; Performing image enhancement on the surface-shaded image using a CLAHE algorithm to obtain an enhanced surface-shaded image; Passing the enhanced surface coloration image through a convolutional neural network model as a feature extractor to obtain a surface coloration feature map; Performing non-negative matrix decomposition on a feature matrix along a channel dimension of the surface shading feature map to obtain a sparse surface shading feature map; Passing the sparse surface coloring feature map through a channel attention module to obtain a classification feature map; and The classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the surface quality of the blank meets a predetermined standard.

2. The upset forging process of an aircraft engine according to claim 1, characterized in that: Passing the enhanced surface coloration image through a convolutional neural network model as a feature extractor to obtain a surface coloration feature map, comprising: Each layer of the convolutional neural network model as a feature extractor performs the following on the input data in the forward pass of the layer: Perform convolution on the input data to obtain a convolution feature map; Performing mean pooling on the convolution feature map to obtain a pooled feature map; and Performing nonlinear activation on the pooled feature map to obtain an activated feature map; The output of the last layer of the convolutional neural network model serving as a feature extractor is the surface coloring feature map, and the input of the first layer of the convolutional neural network model serving as a feature extractor is the enhanced surface coloring image.

3. The upset forging process of an aircraft engine according to claim 2, characterized in that: Passing the sparse surface coloring feature map through a channel attention module to obtain a classification feature map, including: Performing explicit spatial encoding on the sparse surface shading feature map using the channel attention module to obtain a sparse surface shading correlation feature map; Calculating the global mean of each feature matrix along the channel dimension of the sparse surface shading correlation feature map to obtain a channel feature vector; Input the channel feature vector into the Sigmoid activation function to obtain a channel attention weighted feature vector; Based on the autocovariance matrix of the channel attention weighted eigenvector, correcting the eigenvalues ​​of each position in the channel attention weighted eigenvector to obtain an optimized channel attention weighted eigenvector; and Each feature matrix along the channel dimension of the sparse surface shading feature map is weighted using the eigenvalue of each position in the optimized channel attention weighted feature vector as a weight to obtain the classification feature map.

4. The upset forging process for an aircraft engine according to claim 3, characterized in that: The classification feature map is passed through a classifier to obtain a classification result, which is used to indicate whether the surface quality of the blank meets a predetermined standard, including: Expanding the classification feature map into a classification feature vector according to row vectors or column vectors; Performing full connection encoding on the classification feature vector using the fully connected layer of the classifier to obtain an encoded classification feature vector; and The encoded classification feature vector is input into the Softmax classification function of the classifier to obtain the classification result.

5. The upset forging process for an aircraft engine according to claim 4, characterized in that: It also includes a training step for training the convolutional neural network model as a feature extractor, the channel attention module and the classifier.

6. The upset forging process for an aircraft engine according to claim 5, characterized in that: The training step comprises: Acquiring training data, wherein the training data includes a colored training surface image of the blank after coloring, and a true value of whether the surface quality of the blank meets a predetermined standard; Performing image enhancement on the training surface shading image using the CLAHE algorithm to obtain a training enhanced surface shading image; Passing the training enhanced surface coloring image through the convolutional neural network model as a feature extractor to obtain a training surface coloring feature map; performing non-negative matrix decomposition on a feature matrix along a channel dimension of the training surface shading feature map to obtain a training sparse surface shading feature map, and calculating a pseudo-cycle difference penalty factor based on the training sparse surface shading feature map; Passing the training sparse surface coloring feature map through the channel attention module to obtain a training classification feature map; Passing the training classification feature map through a classifier to obtain a classification loss function value; and The convolutional neural network model as a feature extractor, the channel attention module and the classifier are trained using the weighted sum of the classification loss function value and the pseudo-cycle difference penalty factor as the loss function value.

7. The upset forging process for an aircraft engine according to claim 6, characterized in that: Performing non-negative matrix decomposition on a feature matrix along a channel dimension of the training surface shading feature map to obtain a training sparse surface shading feature map, and calculating a pseudo-cycle difference penalty factor based on the training sparse surface shading feature map, including: Performing non-negative matrix decomposition on a feature matrix along a channel dimension of the training surface shading feature map to obtain a plurality of first sub-feature matrices and a plurality of second sub-feature matrices; Expand each of the first sub-eigenmatrix and the second sub-eigenmatrix into a first sub-eigenvector and a second sub-eigenvector; and The pseudo-cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector is calculated.

8. The upset forging process for an aircraft engine according to claim 7, characterized in that: Calculating the pseudo-cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector includes: Calculate the pseudo-cycle difference penalty factor of the first sub-eigenvector and the second sub-eigenvector using the following factor calculation formula; The factor calculation formula is: Among them, V 1i is the first sub-eigenvector, V 2i is the second sub-eigenvector, D(V 1i ,V 2i ) is the distance matrix between the first sub-eigenvector and the second sub-eigenvector, ||·|| F represents the Frobenius norm of the matrix, L is the length of the eigenvector, d(V 1i ,V 2i ) is the distance between the first sub-eigenvector and the second sub-eigenvector, ||·||2 is the two-norm of the vector, log represents the logarithmic function with base 2, and ⊕ represent vector subtraction and vector addition respectively, and α and β are weighted hyperparameters, is the pseudo cycle difference penalty factor.

9. An aircraft engine upsetting system, using the aircraft engine upsetting process according to any one of claims 1 to 8, characterized in that: include: A pre-processing module is used to cut a cylindrical billet of 2A70 material using a band saw, and peel it to 47±0.2, flatten the two end surfaces by 2mm, and round the corners by R5 to obtain a pre-processed billet, wherein the cylindrical billet has a diameter of 52mm and a length of 716mm; A medium-temperature heating module, used for heating the pretreated blank through a medium-temperature electric furnace to obtain a heated blank, wherein the heating temperature is 460±°C and the holding time is greater than 105 minutes; A primary forging module, used for forging the heated blank for the first time through an 800-ton or 1250-ton flat forging machine to obtain a primary forging blank; A secondary forging module is used to forge the primary forging blank for a second time using an 800-ton or 1250-ton flat forging machine to obtain a secondary forging blank; A forming module, used for forming the secondary forging blank through an 800-ton or 1250-ton flat forging machine to obtain a formed blank; A polishing module, for removing burrs from the door mold surface by machining and polishing the formed blank, polishing the burrs with a grinding wheel, and polishing the surface with a polishing machine to obtain a polished blank; and The surface quality inspection module is used to inspect the polished blank by coloring to check whether its surface quality is qualified. After the inspection is qualified, the manufacturing is completed.

Images