A turbine disk forging method based on deep model

Abstract

The present invention provides a turbine disk forging method based on a deep model, which relates to the field of image recognition. The turbine disk forging method based on a deep model of the present invention specifically includes: using a camera to collect infrared images and visible light images of a turbine disk during heat treatment, and respectively obtaining an infrared image set and a visible light image set of the turbine disk to be detected; inputting the infrared image set of the turbine disk to be detected into a regional detection model to obtain a spliced ​​infrared image set of the turbine disk to be identified, and constructing a visible light image set of the turbine disk to be identified based on the infrared image set of the turbine disk to be identified; inputting the visible light image set of the turbine disk to be identified into a smoothness level prediction model to obtain the smoothness level of each area of ​​the turbine disk; outputting heat treatment adjustment parameters according to the smoothness level of each area. Through the above steps, the present invention can finely control the heat treatment of different areas of the turbine disk, thereby improving the overall performance and reliability of the turbine disk.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to a turbine disk forging detection method based on a deep model, which is used for temperature detection and monitoring during the turbine disk forging process. Background Art

[0002] With the continuous advancement of science and technology, the research and development technology of large aircraft is constantly advancing. Among them, turbine discs, as key parts of aircraft, play a vital role in ensuring the flight safety of aircraft. The performance of turbine discs directly affects the reliability of the engine and the safety performance of the overall aircraft, so the accuracy and advancement of its manufacturing process are particularly important. In the existing technology, the forging technology of turbine discs mainly includes key steps such as heat treatment, which can meet the basic needs of turbine discs in general industrial production. However, when these technologies are applied to the aerospace industry, there are still some significant deficiencies. The requirements of the aerospace industry for turbine discs are much higher than general industrial standards. Not only do they require turbine discs to have higher quality grades and stronger performance, but they also need to be differentiated in their physical properties according to the different positions of the turbine discs in the engine and the different working environments they are subjected to. This demand requires that different areas of the turbine disc must be subjected to refined heat treatment during the forging process to ensure that each part can meet its specific performance requirements.

[0003] Existing forging technologies mostly use a unified heat treatment method. Although this method is highly efficient in traditional industrial production, it cannot meet the strict requirements for high performance and high reliability in aerospace applications. The unified heat treatment often ignores the special requirements of the turbine disk for material properties in different working areas. The central area requires better creep resistance, and the edge area pays more attention to fatigue resistance. The limitations of this technology make it impossible to achieve the optimal configuration of physical properties on the entire turbine disk, which may affect the performance and safety of the entire engine. Therefore, this deficiency in the existing technology has become a technical problem that needs to be solved urgently in the aerospace field.

[0004] In response to the above problems, the present invention proposes a turbine disk forging method based on a deep model. By utilizing machine vision technology to monitor and judge in real time the temperature and structural changes of different areas of the turbine disk during the heat treatment process, refined heat treatment control of different areas of the turbine disk can be achieved, thereby improving the overall performance and reliability of the turbine disk and ensuring that each area can reach its optimal performance state. Summary of the invention

[0005] The present invention provides a turbine disk forging method based on a depth model, the method specifically comprising the following steps:

[0006] S1: using a camera to collect infrared images and visible light images of the turbine disk during heat treatment, and obtaining an infrared image set and a visible light image set of the turbine disk to be inspected respectively;

[0007] S2: inputting the infrared image set of the turbine disk to be detected into the regional detection model to obtain a spliced ​​infrared image set of the turbine disk to be identified, and constructing a visible light image set of the turbine disk to be identified based on the infrared image set of the turbine disk to be identified;

[0008] S3: inputting the visible light image set of the turbine disk to be identified into a smoothness level prediction model to obtain the smoothness level of each area of ​​the turbine disk;

[0009] S4: Outputting heat treatment adjustment parameters according to the disk smoothness level of each area.

[0010] At the same time, the present invention also provides a turbine disk forging system based on a deep model, the system comprising:

[0011] Acquisition equipment: using a camera to collect infrared images and visible light images of the turbine disk during heat treatment, and obtaining an infrared image set and a visible light image set of the turbine disk to be inspected respectively;

[0012] Region detection module: the region detection module inputs the infrared image set of the turbine disk to be detected into the region detection model to obtain a spliced ​​infrared image set of the turbine disk to be identified, and constructs a visible light image set of the turbine disk to be identified based on the infrared image set of the turbine disk to be identified;

[0013] Smoothness level prediction module: inputting the visible light image set of the turbine disk to be identified into the smoothness level prediction model to obtain the smoothness level of each area of ​​the turbine disk;

[0014] A heat treatment adjustment module is provided, wherein the heat treatment adjustment module outputs heat treatment adjustment parameters according to the disk smoothness level of each area.

[0015] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned deep model-based turbine disk forging method when executing the computer program.

[0016] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for forging a turbine disk based on a deep model is implemented.

[0017] Compared with the prior art, the present invention aims to solve the problem of being unable to finely control the heat treatment process of aviation turbine disks. The present invention first uses a regional detection model to detect the position of the same temperature zone in the turbine disk, and obtains the same spliced ​​images in the same turbine disk temperature zone by splicing. On this basis, the present invention predicts the smoothness level of the spliced ​​images to obtain the smoothness level of the turbine disks in different areas. Finally, based on the relationship between the smoothness of each area of ​​the turbine disk and the micro-grains, it is determined whether each area meets the heat treatment requirements, and then the fine heat treatment parameter adjustment is completed. Compared with the prior art, the present invention does not adopt the overall temperature control condition method, but adopts stage segmentation first, then smoothness level prediction, and finally local heating adjustment, so as to achieve fine adjustment of the different heat treatment requirements of the turbine disks in different areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flow chart of the turbine disk forging method based on the depth model of the present invention. DETAILED DESCRIPTION

[0020] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0021] The following describes the implementation methods of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.

[0022] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present application, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspect described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0023] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the examples can be practiced without these specific details.

[0024] The embodiment of this specification proposes a turbine disk forging method based on a depth model, and the method specifically includes the following steps:

[0025] S1: using a camera to collect infrared images and visible light images of the turbine disk during heat treatment, and obtaining an infrared image set and a visible light image set of the turbine disk to be inspected respectively;

[0026] S2: inputting the infrared image set of the turbine disk to be detected into the regional detection model to obtain a spliced ​​infrared image set of the turbine disk to be identified, and constructing a visible light image set of the turbine disk to be identified based on the infrared image set of the turbine disk to be identified;

[0027] S3: inputting the visible light image set of the turbine disk to be identified into a smoothness level prediction model to obtain the smoothness level of each area of ​​the turbine disk;

[0028] S4: Outputting heat treatment adjustment parameters according to the disk smoothness level of each area.

[0029] In order to effectively capture infrared and visible light images of the turbine disk in the heating furnace, the present invention designs a slide rail system, which enables the camera to cover multiple rows and columns according to a preset mode to form a complete image matrix or enable the camera to capture a complete turbine disk image when positioned in the center of the heating furnace. The system consists of two parts: a main slide rail for camera positioning and an auxiliary slide rail for evacuation. The main slide rail is made of high-temperature resistant Inconel alloy, has excellent mechanical strength and corrosion resistance, and can work in a high-temperature environment. The slide rail is designed as a modular structure, which is convenient for installation and adjustment in furnaces of different sizes.

[0030] The slide system includes an X-axis and a Y-axis, which control the lateral and longitudinal movement of the camera respectively, so as to cover every area of ​​the turbine disc. The X-axis and Y-axis are driven by servo motors, which can provide precise speed and position control, ensuring the smooth operation of the camera and capturing clear infrared images. Preferably, the servo motor drive model is Siemens SINAMICS S120.

[0031] After the camera has finished shooting, the auxiliary slide is activated. This is a separate linear track, also made of Inconel alloy, and equipped with an independent evacuation motor, preferably a NEMA 34 stepper motor. The evacuation track has an automatic trigger mechanism that starts immediately after the camera completes its mission, quickly and safely evacuating the camera from the hot area.

[0032] The door opening and closing of the entire system is designed to be automated, using high-temperature resistant alloy materials and equipped with an electromagnetic lock control system to ensure that the door is tightly closed during camera operation to prevent heat loss, and automatically opens after the camera is evacuated. The movement and locking of the door are centrally managed by the same PLC control system, which is also responsible for coordinating the movement of the slide rails and the operation of the camera to ensure the safety of the entire shooting process.

[0033] In order to realize the acquisition of infrared images and visible light images of the turbine disk, the shell of the acquisition equipment needs to be insulated and protected. The infrared image and visible light image are acquired by an integrated imaging system that can integrate two different types of cameras under the same motion control framework.

[0034] The integrated imaging system consists of a high temperature and durable dual camera bracket, which is used to fix and protect the camera equipment. On this bracket, an infrared thermal imager and a high-resolution visible light camera are installed. Specifically, the infrared thermal imager is FLIR T1030sc, which is specially designed for high temperature industrial environments, and the visible light camera is Canon EOS 5DMark IV, which can capture high-quality images.

[0035] The two cameras are connected to the external control system via a high-temperature resistant coaxial cable. The cable material is a silicone sheath that is resistant to high temperatures and electromagnetic interference. The control system uses an industrial computer that can receive and process data transmitted from the camera in real time. In order to ensure accurate synchronization of image data, a time code synchronizer is also integrated to ensure that the image data of the two cameras can be accurately aligned for subsequent analysis and comparison.

[0036] The mobile control of the integrated imaging system adopts a precision servo motor system, specifically the Siemens SINAMICS V90, which can provide precise speed and position control.

[0037] The infrared thermal imager and visible light camera need to be insulated. The main materials for thermal insulation protection include aerogel insulation board and ceramic fiber insulation blanket. The aerogel insulation board is directly attached to the camera housing to form the first layer of protection barrier, and then covered with a layer of ceramic fiber insulation blanket to enhance the thermal insulation effect. A layer of silicon-based high-temperature resistant coating is applied on the outside to further improve the heat reflection and thermal insulation performance.

[0038] The protective shell is made of lightweight alloy materials, such as aluminum alloy or titanium alloy. During the assembly process, the aerogel plate and ceramic fiber blanket are cut and fixed according to the internal shape of the protective shell. The infrared thermal imager and visible light camera are then installed in the protective shell, and all insulation materials are evenly covered without any exposed parts. High-temperature resistant screws and fasteners are used to fix the protective shell components to ensure the stability and safety of the equipment in high temperature environments.

[0039] Specifically, an infrared thermal imager and a visible light camera are used to capture infrared images and visible light images of the turbine disk during heat treatment. Exemplarily, the shooting area is set to a specific area in the furnace where the turbine disk is placed. Before image acquisition, the sliding path of the integrated imaging system is designed using control software, and the integrated imaging system is set to start at one end of the heat treatment furnace and gradually slide from front to back to cover the entire area. The design of the camera sliding track needs to ensure that the camera can move smoothly, and the spacing of each movement is set to 0.5 meters, ensuring that the coverage of the camera has about 10% overlap, ensuring that the entire heat treatment area can be systematically covered.

[0040] During the acquisition process, the camera automatically takes a picture every 5 seconds to ensure that the heat treatment area of ​​each turbine disc is fully recorded. Each image will automatically record the location information to accurately mark the location where the image was taken. The thermal insulation protection measures ensure that the camera can work stably and permanently in high temperature environments.

[0041] After the shooting is completed, it will automatically exit the heating furnace through a preset mechanical structure, which is designed to withstand high temperatures and has automatic sliding and evacuation functions to protect the camera and ensure the safety of operation. The acquired images are sorted and sorted according to the sliding path of the camera to construct infrared image sets and visible light image sets of the turbine disk to be inspected. The infrared image of the i-th row and j-th column obtained in the heat treatment furnace is captured by the infrared camera equipment.

[0042] The infrared image set of the turbine disk to be detected is input into the regional detection model to obtain a spliced ​​infrared image set of the turbine disk to be identified. The regional detection model is used to detect images with similar features in the infrared image set of the turbine disk to be detected.

[0043] S2: Input the infrared image set of the turbine disk to be detected into the regional detection model to obtain the spliced ​​infrared image set of the turbine disk to be identified:

[0044] S21: Input each turbine disk infrared image in the turbine disk infrared image set to be detected into the regional detection model, extract the vector to be compared corresponding to each turbine disk infrared image, and initialize the turbine disk infrared image whose row and column are both 1 as the current infrared image to be compared;

[0045] S22: Calculate the similarity between the current infrared image to be compared and the vectors to be compared of the four turbine disk infrared images above, below, left and right thereof. The infrared images that have been compared and the infrared images that do not exist do not need to be compared;

[0046] S23: Preset a similarity threshold, and classify corresponding turbine disk infrared images with similarity greater than the threshold into one category;

[0047] S24: taking each turbine disk infrared image in the turbine disk infrared image set to be detected as the current infrared image to be compared in order from left to right and from top to bottom, and repeatedly executing steps S22-S23 until all turbine disk infrared images have completed similarity comparison, and obtaining the clustering of each turbine disk infrared image in the turbine disk infrared image set to be detected;

[0048] S25: splicing the infrared images of turbine disks of the same category according to the clustering conditions of the infrared images of the turbine disks, and the infrared image set of the turbine disk to be identified is composed of the spliced ​​infrared images of the turbine disks.

[0049] The region detection model includes a first feature extraction branch and a second feature extraction branch;

[0050] The first feature extraction branch includes five feature extraction modules. The input of the first feature extraction module is the infrared image obtained by collecting the turbine disk area. , the five feature extraction modules obtain five intermediate feature maps respectively , each feature extraction module includes a multi-scale extraction module, a fusion module and a down-sampling module, and the feature extraction module is specifically calculated as follows:

[0051]

[0052] in, Represent the input and output of the feature extraction module respectively, Represents the multi-scale fusion feature map of the feature extraction module, represents the convolution calculation, represents the void ratio, , 3*3, 5*5, 7*7, 9*9 represent the size of the convolution kernel respectively, and MaxPooling represent average pooling and maximum pooling respectively, 2*2 is the pooling window size, Indicates channel fusion.

[0053] The input of the second feature extraction branch is the infrared image obtained by collecting the turbine disk area. , the output is , the second feature extraction branch is specifically calculated as:

[0054] ;

[0055] ;

[0056] ;

[0057] in Represents the input image The nth weight score of , is the total number of weights, Represents the loop variable, , , denote the nth linear transformation matrices of query, key and value respectively, Indicates the input image segmentation vectorization, D k represents the dimension of the key, T represents transpose, Represents the input image The fusion weight map, - is the n_last weight parameter matrix, represents the normalized calculation, It means that the convolution and maximum pooling calculations are performed three times in a loop. The convolution kernel is 3*3, the pooling window is 2*2, and the step size is 2.

[0058] The first feature extraction branch is obtained The second feature extraction branch obtains Fusion to get infrared image The vector to be compared is:

[0059]

[0060] in Infrared image The vector to be compared, Indicates the one-dimensional expansion of the image.

[0061] The training process of the region detection model also includes constructing a decoder module, which includes five feature restoration modules, each of which is composed of an upsampling module and a convolution module. During the training process, the input infrared image is extracted by the region detection model to obtain a vector to be compared, and the decoder module receives the vector to be compared to perform image restoration. A loss function is constructed based on the pixel error between the input infrared image and the restored image, and the region detection model and decoder are trained.

[0062] The turbine disc needs to work in an extremely high temperature and high pressure environment. Each area of ​​the turbine disc has different specifications in the physical properties of the material. These performance differences are mainly achieved through heat treatment during the forging process. During the forging process, the macroscopic physical properties and microscopic crystal structure of the turbine disc material are directly affected by the heat treatment temperature. Due to the different functional requirements of different areas, the optimal heat treatment temperature of each area is also different.

[0063] The present invention proposes a method based on infrared imaging technology that can monitor and record the temperature distribution of the turbine disk during the forging process. Through an infrared camera, a thermal image of the surface of the turbine disk can be obtained, and these images are then stitched and analyzed to form a continuous temperature distribution map. In this way, areas on the turbine disk with the same or similar temperatures can be accurately identified. Using these data, the microstructure and macroscopic physical properties of the material in different temperature areas can be further predicted. Through comparative analysis and simulation, the present invention can predict the behavior of the material at a specific temperature. This prediction can guide whether the current heat treatment process has achieved the expected physical performance standards, or whether there are areas that are overheated or have not reached the ideal temperature. If it is found that the temperature of certain areas does not meet the requirements, the local heating temperature during the forging process can be adjusted immediately, such as by increasing or decreasing the heating time, adjusting the power output of the heater, or changing the position and layout of the heater.

[0064] Constructing a visible light image set of the turbine disk to be identified according to the infrared image set of the turbine disk to be identified;

[0065] Since both the infrared thermal imager and the visible light camera are fixed to the dual camera bracket of the integrated imaging system, it can be ensured that the areas captured by each captured infrared image and visible light image are completely consistent, that is, the mapping relationship between each infrared image and visible light image is one-to-one corresponding. Therefore, the present invention constructs a visible light image set of the turbine disk to be identified based on the infrared image set of the turbine disk to be identified obtained after splicing and the mapping relationship between the infrared image and the visible light image, and the visible light images in the visible light image set of the turbine disk to be identified correspond one-to-one to the spliced ​​infrared image of the turbine disk.

[0066] After obtaining the visible light image set of the turbine disk to be identified, determining the regional position information of each spliced ​​image in the visible light image set of the turbine disk to be identified;

[0067] The center point of the turbine disk region is set to (row / 2, col / 2), where row and col represent the number of rows and columns of the infrared image or visible light image acquired by the camera according to the sliding path;

[0068] When the distance from each visible light image in the stitched image to the center point is less than the first threshold, the position information of the stitched image is the center of the disk;

[0069] When the distance from each visible light image in the stitched image to the center point is greater than the second threshold, the position information of the stitched image is the leaf attachment area;

[0070] Except for the above two cases, the position information of the stitched image is the edge.

[0071] S3: inputting the visible light image set of the turbine disk to be identified into a smoothness level prediction model to obtain the turbine disk smoothness level of each area in the visible light image set of the turbine disk to be identified;

[0072] Each visible light image in the visible light image set of the turbine disk to be identified is preprocessed. The preprocessing first performs cropping and padding. The visible light image of the turbine disk to be identified after cropping and padding meets the input image size requirement of the smoothness grade prediction model; the cropped and padded turbine disk image to be identified is grayscale converted and normalized, and the normalized turbine disk image to be identified is input into the smoothness grade prediction model to obtain the corresponding smoothness grade prediction result. The recognition result includes 6 smoothness grades of the turbine disk surface, grades 1-6; macroscopically, each smoothness grade corresponds to the grain size during the heat treatment of the turbine disk, specifically, grade 1: >125 µm; grade 2: 100-125 µm; grade 3: 75-100 µm; grade 4: 50-75µm; grade 5: 35-50 µm; grade 6: <35 µm.

[0073] During the forging heat treatment process of aviation turbine disks, the physical properties required in various areas of the turbine disk vary due to their application environment, and these properties are directly related to the structural size of the microscopic grains. In the traditional production process, the detection of microscopic grains relies on precision instruments such as electron microscopes, which can provide accurate grain size and structural data. However, during the actual heat treatment process of the turbine disk, due to the limitations of high temperature environment and operating conditions, these devices cannot be directly used for real-time monitoring of the grain structure. Specifically, electron microscopes need to work in a high vacuum environment to prevent the electron beam from interacting with air molecules. However, in a forging furnace, it is technically very challenging to maintain a stable vacuum environment in a high temperature and humid furnace environment. This limitation makes the grain size assessment during the heat treatment process rely on the operator's experience and intuitive judgment, which not only increases the uncertainty of production, but also may lead to non-uniformity of product performance.

[0074] The study found that during the production of turbine disks, the size and distribution of grains are crucial to their final physical properties. Larger grains mean better toughness, but lower hardness and strength; smaller grains mean higher hardness and strength, but lower toughness. Therefore, being able to accurately control and predict grain size is critical to ensuring the quality and performance of turbine disks. The smoothness and texture uniformity of the turbine disk surface are significantly correlated with grain size: rough surface, low smoothness and uneven texture are associated with larger grain size, while smooth and uniform surface indicates smaller grain size. For related research, see: Lu Yuhua, Wang Peng, Shen Xuejing, Fu Rui, Li Fulin, Li Dongling, Wang Haizhou. Characterization of the correlation between grain size and hardness of large-size deformed FGH96 superalloy turbine disks[J]. Physical and Chemical Testing—Physical Section, 2017, 53(8):544-547;

[0075] Effect of Grain Size on Low Cycle Fatigue Life in Compressor DiscSuperalloy GH4169 at 600 °;

[0076] and Materials Science and Engineering: An Introduction, William D. Callister Jr. and David G. Rethwisch.

[0077] Based on this discovery, the present invention has established a mapping relationship between surface smoothness and grain size through a large number of experiments, and can then use macroscopic optical detection technology to predict the grain size of a specific area of ​​the turbine disk after heat treatment. That is, the grain structure of the turbine disk can be predicted and controlled without the need for direct microstructural measurement. This method simplifies the traditional detection process and improves production efficiency and product quality. The heat treatment process can be monitored and adjusted to ensure that each turbine disk meets strict quality standards. This makes the performance of aviation turbine disks more stable and reliable.

[0078] The smoothness level prediction model includes a global branch and a local branch, and the local branch includes three backbone extraction modules and a feature enhancement module, wherein the three backbone extraction modules are respectively defined as:

[0079]

[0080]

[0081]

[0082]

[0083]

[0084] in is the pre-processed visible light image of the turbine disk to be identified, is the output of the first backbone extraction module, It means that the convolution-batch normalization-activation calculation is performed twice in a loop, where the first convolution step size is 1 and the second convolution step size is 2; represents pixel-by-pixel fusion, and They represent the intermediate output feature map and output feature map of the second backbone extraction module respectively, and They represent the intermediate output feature map and output feature map of the third backbone extraction module respectively.

[0085] The input image of the global branch The feature map that retains the original image information is obtained by downsampling and combined with the output feature map obtained by the local branch. Input them together into the feature enhancement module for feature map enhancement;

[0086] The feature enhancement module includes a cross-layer fusion module and an attention enhancement module, and the cross-layer fusion module is defined as:

[0087] ;

[0088] The attention enhancement module is defined as:

[0089] ;

[0090] in, represents the output features of the cross-layer fusion module, represents the output of the attention enhancement module;

[0091] The output features of the cross-layer fusion module and the attention enhancement module are input into the classification module to obtain the turbine disk smoothness level result of each area in the turbine disk visible light image set to be identified. The classification module is defined as:

[0092] ;

[0093] in Indicates that two fully connected layer calculations are performed. is the calculated probability matrix of turbine disk smoothness level results.

[0094] S4: Output heat treatment adjustment parameters according to the smoothness level of the turbine disk in each area.

[0095] The target grain size at the center of the disk is set to 125 µm, the target grain size at the blade attachment area is set to 75 µm, and the target grain size at the outer edge is set to 35 µm;

[0096] When the grain size interval corresponding to the smoothness level of the visible light image of the turbine disk to be identified is smaller than the target grain size of the corresponding position area, the preset temperature interval of the area where the visible light image of the turbine disk to be identified is located is increased; when the grain size interval corresponding to the smoothness level of the visible light image of the turbine disk to be identified is larger than the target grain size of the corresponding position area, the preset temperature interval is maintained;

[0097] The temperature range of the preset area in the center of the disk is 950°C-980°C; the temperature range of the preset area in the blade attachment area is 920°C-950°C; the temperature range of the preset area on the outer edge is 880°C-920°C.

[0098] The present invention provides a turbine disk forging system based on a depth model, the system comprising:

[0099] Acquisition equipment: using a camera to collect infrared images and visible light images of the turbine disk during heat treatment, and obtaining an infrared image set and a visible light image set of the turbine disk to be inspected respectively;

[0100] Region detection module: the region detection module inputs the infrared image set of the turbine disk to be detected into the region detection model to obtain a spliced ​​infrared image set of the turbine disk to be identified, and constructs a visible light image set of the turbine disk to be identified based on the infrared image set of the turbine disk to be identified;

[0101] Smoothness level prediction module: inputting the visible light image set of the turbine disk to be identified into the smoothness level prediction model to obtain the smoothness level of each area of ​​the turbine disk;

[0102] A heat treatment adjustment module is provided, wherein the heat treatment adjustment module outputs heat treatment adjustment parameters according to the disk smoothness level of each area.

[0103] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned deep model-based turbine disk forging method when executing the computer program.

[0104] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for forging a turbine disk based on a deep model is implemented.

[0105] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0106] In this specification, the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the embodiments described later, the description is relatively simple, and the relevant parts can be referred to the partial description of the previous embodiments.

[0107] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A turbine disk forging method based on a deep model, characterized in that: The method comprises the following steps: S1: using a camera to collect infrared images and visible light images of the turbine disk during heat treatment, and obtaining an infrared image set and a visible light image set of the turbine disk to be inspected respectively; S2: Inputting the infrared image set of the turbine disk to be detected into the regional detection model to obtain a spliced ​​infrared image set of the turbine disk to be identified, and constructing a visible light image set of the turbine disk to be identified according to the spliced ​​infrared image set of the turbine disk to be identified and the mapping relationship between the infrared image and the visible light image, wherein the visible light images in the constructed visible light image set of the turbine disk to be identified correspond one to one with the spliced ​​infrared image of the turbine disk; S3: inputting the visible light image set of the turbine disk to be identified into a smoothness level prediction model to obtain the smoothness level of each area of ​​the turbine disk; S4: Output heat treatment adjustment parameters according to the smoothness level of each area.

2. The method for forging a turbine disk based on a deep model according to claim 1, characterized in that: The infrared image set of the turbine disk to be detected is input into the regional detection model, and the spliced ​​infrared image set of the turbine disk to be identified includes: S21: Input each turbine disk infrared image in the turbine disk infrared image set to be detected into the regional detection model, extract the vector to be compared corresponding to each turbine disk infrared image, and initialize the turbine disk infrared image whose row and column are both 1 as the current infrared image to be compared; S22: Calculate the similarity between the current infrared image to be compared and the vectors to be compared of the four turbine disk infrared images above, below, left and right thereof. The infrared images that have been compared and the infrared images that do not exist do not need to be compared; S23: Preset a similarity threshold, and classify corresponding turbine disk infrared images with similarity greater than the threshold into one category; S24: taking each turbine disk infrared image in the turbine disk infrared image set to be detected as the current infrared image to be compared in order from left to right and from top to bottom, and repeatedly executing steps S22-S23 until all turbine disk infrared images have completed similarity comparison, and obtaining the clustering of each turbine disk infrared image in the turbine disk infrared image set to be detected; S25: splicing the infrared images of turbine disks of the same category according to the clustering conditions of the infrared images of the turbine disks, and the infrared image set of the turbine disk to be identified is composed of the spliced ​​infrared images of the turbine disks.

3. The method for forging a turbine disk based on a depth model according to claim 2, characterized in that: The region detection model includes a first feature extraction branch and a second feature extraction branch; The first feature extraction branch includes five feature extraction modules. The input of the first feature extraction module is the infrared image obtained by collecting the turbine disk area. The five feature extraction modules respectively obtain five intermediate feature images. Each feature extraction module includes a multi-scale extraction module, a fusion module and a downsampling module. The feature extraction module is specifically calculated as follows: in, and They represent the input and output of the feature extraction module respectively, Fmulti represents the multi-scale fusion feature map of the feature extraction module, Conv represents the convolution calculation, ρ represents the void ratio, 1*1, 3*3, 5*5, 7*7, 9*9 represent the convolution kernel size respectively, AvgPooling and MaxPooling represent the average pooling and maximum pooling respectively, 2*2 is the pooling window size, Indicates channel fusion.

4. The method for forging a turbine disk based on a deep model according to claim 3, characterized in that: The input of the second feature extraction branch is the infrared image obtained by collecting the turbine disk area. The output is The second feature extraction branch is specifically calculated as: in Represents the input image The nth weight score, 1≤n≤n_last, n_last is the total number of weights, n_temp represents the loop variable, Represents the nth linear transformation matrix of query, key and value respectively, pe() represents the input image segmentation vectorization, D k represents the dimension of the key, T represents transpose, Represents the input image The fusion weight map, T1-T n_last is the n_last weight parameter matrix, Normalize represents the normalization calculation, Conv_Maxpooling 3 It means that convolution and maximum pooling calculations are performed three times in a loop. In the convolution calculation, the convolution kernel is 3*3, the pooling window is 2*2, and the step size is 2.

5. The method for forging a turbine disk based on a deep model according to claim 4, characterized in that: The first feature extraction branch is obtained The second feature extraction branch obtains Fusion to get infrared image The vector to be compared is: Where V i，j Infrared image The vector to be compared is , and Flatten represents the one-dimensional expansion of the image.

6. The method for forging a turbine disk based on a deep model according to claim 1, characterized in that: The visible light image set of the turbine disk to be identified is input into the smoothness level prediction model to obtain the smoothness level of the turbine disk in each area of ​​the visible light image set of the turbine disk to be identified; the identification result includes 6 smoothness levels of the turbine disk surface, levels 1-6; each smoothness level corresponds to the grain size during the heat treatment process of the turbine disk.

7. The method for forging a turbine disk based on a depth model according to claim 6, characterized in that: The smoothness level prediction model includes a global branch and a local branch, and the local branch includes three backbone extraction modules and a feature enhancement module, wherein the three backbone extraction modules are respectively defined as: in is the pre-processed visible light image of the turbine disk to be identified, is the output of the first backbone extraction module, Sigmoid_BN_Conv 2 It means that the convolution-batch normalization-activation calculation is performed twice in a loop, where the first convolution step size is 1 and the second convolution step size is 2; represents pixel-by-pixel fusion, and They represent the intermediate output feature map and output feature map of the second backbone extraction module respectively, and represent the intermediate output feature map and the output feature map of the third backbone extraction module respectively; The input image of the global branch The feature map that retains the original image information is obtained by downsampling and combined with the output feature map obtained by the local branch. Input them together into the feature enhancement module for feature map enhancement; The feature enhancement module includes a cross-layer fusion module and an attention enhancement module, and the cross-layer fusion module is defined as: The attention enhancement module is defined as: Among them, F G represents the output features of the cross-layer fusion module, F A represents the output of the attention enhancement module; the output features of the cross-layer fusion module and the attention enhancement module are input into the classification module to obtain the turbine disk smoothness level result of each area in the turbine disk visible light image set to be identified. The classification module is defined as: Prob=Softmax(g 2 (AvgPooling(F G ☉F A ))); where g 2 It indicates that two fully connected layer calculations are performed, and Prob is the calculated probability matrix of the turbine disk smoothness level result.

8. A turbine disk forging system based on a deep model, used to execute a turbine disk forging method based on a deep model as claimed in claim 1, characterized in that: The system includes: Acquisition equipment: using a camera to collect infrared images and visible light images of the turbine disk during heat treatment, and obtaining an infrared image set and a visible light image set of the turbine disk to be inspected respectively; Region detection module: the region detection module inputs the infrared image set of the turbine disk to be detected into the region detection model to obtain a spliced ​​infrared image set of the turbine disk to be identified, and constructs a visible light image set of the turbine disk to be identified according to the spliced ​​infrared image set of the turbine disk to be identified and the mapping relationship between the infrared image and the visible light image, wherein the visible light images in the constructed visible light image set of the turbine disk to be identified correspond one to one with the spliced ​​infrared image of the turbine disk; Smoothness level prediction module: inputting the visible light image set of the turbine disk to be identified into the smoothness level prediction model to obtain the smoothness level of each area of ​​the turbine disk; A heat treatment adjustment module is provided, wherein the heat treatment adjustment module outputs heat treatment adjustment parameters according to the smoothness level of each area.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for forging a turbine disk based on a deep model as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements a turbine disk forging method based on a deep model as claimed in any one of claims 1 to 7.

Images