High-energy X-ray on-line monitoring device

Abstract

The invention discloses a high-energy X-ray online monitoring device, and aims to realize real-time monitoring and image quality optimization of a deposition process through multi-angle X-ray image acquisition and automatic parameter adjustment of a neural network. According to the method, high-energy X-ray equipment and high-resolution detectors are installed at the four corners of a laser head, and surface and internal structure changes of a deposition layer can be captured in real time from multiple visual angles. The system analyzes the image quality through the neural network, automatically adjusts the exposure time, light intensity and imaging angle of the X-ray equipment, and ensures that the obtained image has higher contrast ratio, signal-to-noise ratio and structural similarity. Finally, through three-dimensional reconstruction of multi-angle images, the system can present the geometrical morphology of the deposition layer in real time, and data support is provided for process optimization. The scheme also provides a double-branch convolutional neural network architecture combining ResNet and VGG, which is used for improving the quality evaluation of the X-ray image. According to the method, the quality control precision and efficiency of the laser metal deposition process are greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of additive manufacturing, and in particular relates to a high-energy X-ray online monitoring device for online monitoring and defect prediction of laser metal deposition. Background Art

[0002] During the laser metal deposition process, the rapid melting and solidification of the material can easily cause changes in the internal structure of the deposited layer, such as deformation and shrinkage. Existing optical monitoring technology mainly focuses on external morphology, and it is difficult to effectively monitor the internal structure during the deposition process, especially in deep areas. Traditional X-ray monitoring methods usually rely on imaging from a single angle, which leads to limited viewing angles and the inability to fully capture the entire picture of the deposition area. In actual production, this monitoring method is difficult to efficiently monitor the internal quality of laser metal deposition in real time and comprehensively, and it is impossible to dynamically adjust the imaging parameters based on real-time images to adapt to changing needs under different working conditions. Summary of the invention

[0003] The present invention provides a real-time monitoring device for laser metal deposition based on high-energy X-rays and high-resolution cameras, comprising: a laser head; four high-energy X-ray devices, namely a first high-energy X-ray device, a second high-energy X-ray device, a third high-energy X-ray device, and a fourth high-energy X-ray device, which are respectively installed at the four corners of the laser head; four high-resolution cameras, namely a first high-resolution camera, a second high-resolution camera, a third high-resolution camera, and a fourth high-resolution camera, each camera being coaxially installed with the corresponding X-ray device; a substrate being located directly below the laser head; and a central processing system for receiving and processing image data from the X-ray devices and cameras, and performing image fusion and analysis.

[0004] The high-energy X-ray device and camera dynamically select the working mode according to the printing trajectory of the laser head; when the laser head prints along the oblique trajectory, the two groups of high-energy X-ray devices and cameras in the diagonal position are activated; when the laser head prints along the horizontal or vertical trajectory, the two groups of high-energy X-ray devices and cameras parallel to the printing trajectory are activated. The present invention realizes the synchronous monitoring of the external morphology and internal structure of the deposited layer through multi-angle real-time monitoring and coaxial configuration of high-energy X-rays and cameras, which significantly improves the monitoring accuracy and data consistency. In addition, the real-time data acquisition method enhances the system's adaptability to dynamic changes in the deposition process, thereby effectively optimizing the quality control of laser metal deposition.

[0005] During the laser metal deposition process, the laser head prints layer by layer on the substrate along a preset trajectory. In order to monitor the printing process, this device activates different combinations of X-ray devices and cameras according to the printing position through a preset algorithm. The specific monitoring process is as follows:

[0006] When printing the molten pool, the high-energy X-ray device and the high-energy X-ray device, as well as the high-resolution camera and the high-resolution camera are activated. These two sets of equipment monitor the printing area from diagonal directions respectively, ensuring that the internal structure and surface morphology of the deposited layer can be accurately captured from different angles.

[0007] When printing the molten pool, high-energy X-ray devices and high-energy X-ray devices, as well as high-resolution cameras and high-resolution cameras are activated. These devices are located on both sides of the printing area and are responsible for obtaining the thickness of the deposited layer, the dynamics of the molten pool, and the external contour morphology to ensure data integrity.

[0008] When the melt pool is printed, the high-energy X-ray device and the high-energy X-ray device, as well as the high-resolution camera and the high-resolution camera are activated. The two combinations monitor the deposition process from different angles at the same time, ensuring that the geometry of the melt pool and the deposited layer can be accurately recorded.

[0009] Throughout the printing process, the system dynamically selects and activates the optimal combination of X-ray equipment and cameras through a self-developed intelligent algorithm to ensure that the printing process of each molten pool can always be monitored from the best angle. The algorithm automatically adjusts the imaging angle, exposure time and light intensity based on a multi-dimensional analysis of the laser head position, printing trajectory and real-time imaging requirements to obtain clear and high-quality images. Each time the equipment is working, the collected images are transmitted to the central processing unit in real time through a high-speed data transmission system. The central processing unit quickly analyzes the image data and generates feedback results. It can not only detect possible printing defects, but also adjust the current printing process parameters such as laser power or printing speed based on data feedback, thereby achieving efficient quality control and process optimization in a dynamic deposition environment.

[0010] Furthermore, this study proposed a dual-branch convolutional neural network architecture combining ResNet and VGG to improve the quality assessment of X-ray images. First, the system collects X-ray images of the laser metal deposition process from multiple angles in real time. Subsequently, deep features are extracted by ResNet and shallow features are extracted by VGG. The features of the two are spliced ​​in the fusion layer, and the quality score or classification result of the image is output. Based on these evaluation results, the system automatically adjusts the exposure time, light intensity and acquisition angle of the X-ray equipment to ensure continuous optimization of image quality. Throughout the process, the system continuously updates the equipment parameters through feedback control to adapt to the changing needs under different working conditions, and ultimately ensures high-quality real-time image acquisition.

[0011] Furthermore, this study reconstructs the three-dimensional morphology of the deposited layer by integrating the position and height information of the melt pool and adopting a subtractive modeling method. In this process, the system sets the cross section of the deposited layer based on the assumption of an arc shape and defines a spherical grab function. This function is used to gradually activate the cells in the area to be activated, thereby accurately mapping the actual shape of the melt pool.

[0012] In step 3, real-time data during the laser metal deposition process is obtained through multi-angle X-ray monitoring, and the three-dimensional morphology of the deposited layer is gradually reconstructed. Based on the image data captured from multiple angles, the system integrates and updates the geometric morphology of the deposited layer in real time. The reconstruction process adopts a subtractive modeling method, based on the position information and height data of the molten pool, and continuously adjusts the three-dimensional model of the deposited layer to ensure that the model can accurately reflect the dynamic changes of the actual deposition morphology. Through this method, the system can dynamically respond to structural changes during the deposition process, making the three-dimensional reconstruction more accurate. The cross-section of the deposited layer is designed to be an arc, and the spherical grab function is used to gradually activate the units in the area to be activated to dynamically update the three-dimensional geometric morphology of the deposited layer. This design allows the system to accurately adjust the reconstructed model after obtaining the new molten pool position and height data, thereby ensuring that the three-dimensional model is highly consistent with the actual deposition state. The spherical grab function realizes the refined processing of the dynamic update of the morphology, making the reconstruction process highly sensitive to subtle changes in the deposited layer, further improving the monitoring accuracy.

[0013] In step 3, the laser metal deposition process is monitored in real time by multi-angle X-rays, and the system obtains the dynamic height data of the deposited layer and the position information of the molten pool. Based on these real-time monitoring data, the idea of ​​subtractive modeling is adopted to gradually reconstruct the three-dimensional morphology of the deposited layer. In order to more accurately simulate the surface characteristics of the deposited layer naturally formed by the surface tension of the molten pool, it is assumed that the cross-section of the deposited layer is an arc, and a coordinate system with X as the scanning direction, Y as the height direction, and Z as the width direction of the deposited layer is established for the reconstruction process.

[0014] While receiving real-time melt pool data, the system continuously updates the 3D model until the deposition process is completely completed. This continuous update process ensures that the reconstructed 3D morphology is highly matched with the geometric features of the actual deposited layer. The implementation of this method significantly improves the accuracy and practicality of the model by accurately simulating the dynamic changes of the deposited layer during the actual manufacturing process, making it an important tool for optimizing the printing process and identifying potential defects in advance. Compared with the prior art, the advantages of the present invention are as follows:

[0015] The present invention realizes multi-angle real-time monitoring by installing multiple high-energy X-ray devices and high-resolution cameras on the laser head. This design can fully cover the deposition area of ​​laser metal deposition, avoiding the imaging blind area problem that may be caused by traditional single-angle monitoring, and ensuring comprehensive monitoring of the surface morphology and internal structure of the deposited layer.

[0016] The present invention has an intelligent switching mechanism that can dynamically select different angles of X-ray equipment and camera combinations according to the printing trajectory of the laser head. During the diagonal, horizontal and vertical printing process, the system automatically activates the two sets of equipment at the best angles, effectively improving the monitoring efficiency and reducing the redundant use of equipment.

[0017] The present invention combines neural network technology to evaluate the quality of X-ray images in real time, automatically adjust exposure time, light intensity and imaging angle to ensure image contrast and clarity. This automated feedback mechanism can optimize equipment parameters according to actual working conditions, ensuring that the system can continuously provide high-quality image data.

[0018] The present invention can also perform real-time 3D reconstruction through multi-angle X-ray images to accurately reflect the geometric morphology and internal structure changes of the deposited layer. This real-time 3D reconstruction method provides reliable morphological data support for subsequent defect detection and process optimization, greatly improving the accuracy and quality control level of the laser metal deposition process. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a right oblique view of the online monitoring device of the present invention,

[0020] Figure 2 is a left oblique view of the online monitoring device of the present invention,

[0021] Figure 3 is a flow chart of the method for automatically adjusting high-energy X-ray parameters proposed by the present invention,

[0022] Figure 4 It is a flow chart of the high-energy X-ray three-dimensional reconstruction method proposed in the present invention.

[0023] In the figure: 100, laser head, 201, first high-resolution camera, 202, first high-energy X-ray device, 203, second high-resolution camera, 204, second high-energy X-ray device, 205, third high-resolution camera, 206, third high-energy X-ray device, 207, fourth high-resolution camera, 208, fourth high-energy X-ray device, 300, substrate, 301, first molten pool, 302, second molten pool, 303, third molten pool. DETAILED DESCRIPTION

[0024] In order to deepen the understanding of the present invention, the present embodiment is described in detail below with reference to the accompanying drawings.

[0025] Example 1: See Figure 1-Figure 4, a high-energy X-ray online monitoring device, comprising: a laser head 100; four high-energy X-ray devices, namely a first high-energy X-ray device 202, a second high-energy X-ray device 204, a third high-energy X-ray device 206, and a fourth high-energy X-ray device 208, respectively installed at the four corners of the laser head 100; four high-resolution cameras, namely a first high-resolution camera 201, a second high-resolution camera 203, a third high-resolution camera 205, and a fourth high-resolution camera 207, each camera is coaxially installed with the corresponding X-ray device; a substrate 300 is located directly below the laser head 100; and a central processing system for receiving and processing image data from the X-ray device and the camera, and performing image fusion and analysis. The layout of these devices ensures real-time monitoring of the printing area from multiple angles, and each X-ray device and camera combination captures the surface morphology and internal structure changes during the deposition process from different perspectives, avoiding visual blind spots in monitoring.

[0026] Among them, the high-energy X-ray equipment and camera dynamically select the working mode according to the printing trajectory of the laser head; when the laser head prints along the oblique trajectory, the two groups of high-energy X-ray equipment and cameras in the diagonal position are activated; when the laser head prints along the horizontal or vertical trajectory, the two groups of high-energy X-ray equipment and cameras parallel to the printing trajectory are activated.

[0027] In order to monitor the printing process, the device activates different combinations of X-ray devices and cameras according to the printing position. The specific monitoring process is as follows:

[0028] When printing the second molten pool (302), the second high-energy X-ray device (204) and the fourth high-energy X-ray device (208), as well as the second high-resolution camera (203) and the fourth high-resolution camera (207) are activated. These two sets of devices monitor the printing area from diagonal directions respectively to ensure that the internal structure and surface morphology of the deposited layer can be accurately captured from different angles.

[0029] When printing the first molten pool (301), the first high-energy X-ray device (202) and the fourth high-energy X-ray device (208), as well as the first high-resolution camera (201) and the fourth high-resolution camera (207) are activated. These devices are located on both sides of the printing area and are responsible for obtaining the thickness of the deposited layer, the dynamics of the molten pool and the external contour morphology to ensure data integrity;

[0030] When printing the third molten pool (303), the second high-energy X-ray device (204) and the third high-energy X-ray device (206), as well as the second high-resolution camera (203) and the third high-resolution camera (205) are activated. The two combinations simultaneously monitor the deposition process from different angles to ensure that the geometry of the molten pool and the deposited layer can be accurately recorded.

[0031] During the laser metal deposition process, the device intelligently activates different X-ray and camera combinations according to the movement trajectory of the laser head to ensure the best monitoring effect at different stages of printing. When the laser head prints along an oblique line, horizontally or vertically, the system selects the corresponding two sets of X-ray devices and cameras to work according to the current printing trajectory, thereby ensuring that no matter how the printing direction changes, the key information of the deposited layer can be captured in real time. The X-ray device provides an image of the internal structure of the deposited layer, and the camera simultaneously records the external morphology data. This information forms a complete image data set during the printing process.

[0032] By installing high-energy X-ray equipment and high-resolution cameras at the four corners of the laser head, the system can capture the surface and internal structure of the molten pool during the laser metal deposition process from multiple angles in real time. This configuration enables the equipment to fully cover the molten pool area and provide detailed images of key structural changes and melting dynamics. The collected image data is transmitted to the central processing system. The system uses advanced image processing algorithms to analyze the data in real time and automatically adjust the exposure time, light intensity and imaging angle of the X-ray equipment to optimize the image quality. This process not only improves the clarity and contrast of the image, but also provides high-quality input data for subsequent 3D reconstruction. Finally, the system uses these optimized images for accurate 3D reconstruction to simulate the 3D geometry of the deposited layer in detail to support further analysis and manufacturing process optimization.

[0033] The scheme significantly enhances the monitoring accuracy of the molten pool dynamics by implementing multi-angle real-time monitoring. Traditional technologies generally rely on a single perspective for monitoring and fail to fully capture the detailed changes in the internal structure of the molten pool. The automatic image analysis and parameter adjustment functions of the central processing system greatly improve operational efficiency, reduce dependence on professional operators, and ensure the consistency and accuracy of data acquisition. In addition, the application of three-dimensional reconstruction technology provides in-depth information beyond traditional two-dimensional images, further refining the quality control of the deposited layer and the subsequent analysis of material properties. These innovative measures have significantly optimized the practical application of laser metal deposition technology and brought significant technical advantages.

[0034] Example 2: See Figure 1-Figure 4 , a high energy X-ray online monitoring device, a method of using, comprising the following steps:

[0035] Step 1: Use high-energy X-ray equipment and cameras to capture the surface and internal structure of the molten pool during laser metal deposition in real time;

[0036] Step 2: The collected image data is transmitted to the central processing system, which uses advanced image processing algorithms to analyze the data in real time and automatically adjust the exposure time, light intensity and imaging angle of the X-ray equipment to optimize image quality.

[0037] Step 3: Reconstruct the geometric morphology of the sedimentary layer in three dimensions based on the multi-angle image data.

[0038] After each X-ray image and camera image acquisition is completed, the data will be sent to the central processing unit through a high-speed data transmission system. The processing system performs data fusion based on the images collected at different angles to ensure high contrast and high definition of multi-angle X-ray images, so that the system can accurately monitor subtle changes in the surface and internal structure of the deposited layer. Through these high-quality image data, subsequent 3D reconstruction and data analysis will be based on complete and accurate image information to ensure fine control and optimization of the laser metal deposition process. Among them, step 2 is as follows:

[0039] The system uses a convolutional neural network architecture that combines ResNet and VGG to perform multi-level deep feature analysis on X-ray images. The system first uses a convolutional neural network to score the quality of the X-ray images collected in real time, and outputs scoring results covering image signal-to-noise ratio, contrast, and detail resolution. Based on these scoring results, the system intelligently adjusts the exposure time, light intensity, and imaging angle of the X-ray equipment to optimize the imaging parameters in real time, thereby ensuring that the acquired images have high clarity and contrast.

[0040] Furthermore, after completing the initial parameter adjustment, the system re-collects images through an intelligent feedback mechanism and re-scores them to verify the improvement of the image quality after adjustment. If the scoring results show that the image quality still does not meet the preset standards, the system will further iterate and adjust the parameters until the score meets the set requirements. This loop feedback mechanism realizes the automatic optimization and dynamic adjustment of image quality, ensuring that high-quality images are always maintained during the imaging process. The advanced neural network technology introduced in this technical solution significantly improves the operating efficiency and image quality of the X-ray imaging system, and the automated parameter adjustment and continuous quality improvement process are innovations that have not been achieved in existing technologies.

[0041] Among them, the neural network must be trained with a special training data set before application to ensure the accuracy of analysis and automatic adjustment. The training data set covers X-ray images collected under various working conditions, and each image is carefully labeled according to key quality indicators such as signal-to-noise ratio and contrast. This training process equips the neural network with the ability to adapt to various imaging conditions, ensuring that it can accurately adjust the equipment settings according to the monitored image quality in actual application.

[0042] The neural network is used to automatically adjust parameters. The specific implementation plan is: a convolutional neural network combining ResNet and VGG is selected as the core architecture of image analysis. The neural network needs to be trained before practical application. The training data set includes X-ray images collected under different working conditions. These images are annotated according to indicators such as signal-to-noise ratio (SNR), contrast, and structural similarity (SSIM). Through training, the convolutional neural network can effectively extract features from the image and output corresponding scores based on these features, ensuring that the system can accurately evaluate the image quality in actual operation.

[0043] After training is completed, the system enters the working state and processes the real-time collected X-ray images. After feature extraction by the convolutional neural network, the image features are input into three independent scoring networks for evaluating SNR, contrast and SSIM respectively. The score output by each scoring network is in the range of 0 to 1, which is used to represent the performance of the image on this quality indicator. The scoring mechanism is implemented through the relationship between the features and labels learned by the convolutional neural network during the training process, ensuring that the score can accurately reflect the image quality.

[0044] Based on the output of the scoring network, the system automatically adjusts the parameters of the X-ray equipment. If the SNR score is low, the system will increase the signal strength of the image by increasing the exposure time. The adjustment of the exposure time follows the following formula:

[0045] T=T min +(1-S SNR )×(T max -T min )

[0046] Among them, T max and T min Respectively represent the minimum and maximum exposure time, S SNR Indicates the signal-to-noise ratio score of the image. When the signal-to-noise ratio score is low, the system will automatically increase the exposure time to improve the signal strength of the image.

[0047] If the contrast score is low, the system will reduce the light intensity to improve the clarity of image details. The light intensity adjustment formula is:

[0048] I=I min +(1-S Contrast )×(I max -I min )

[0049] Among them, I max and I min are the minimum and maximum light intensity, S Contrast Indicates the contrast score of the image. When the contrast score is low, the system automatically increases the light intensity to improve the contrast of the image.

[0050] If the SSIM score is low, the system will adjust the imaging angle to optimize the capture of structural information in the image. The angle adjustment follows the following formula:

[0051] θ=θ ideal +(1-S SSIM )×Δθ

[0052] Among them, θ ideal is the ideal imaging angle, Δθ is the step size of angle adjustment, S SSIM SSIM is used to score the structural similarity. When the SSIM score is low, the system will adjust the angle accordingly to improve the structural clarity of the image.

[0053] Each time the equipment parameters are adjusted, the system re-collects the X-ray image and performs feature extraction and scoring again through the convolutional neural network. If the score does not meet the set standard, the system will continue to adjust the parameters to form a closed-loop feedback control. This feedback mechanism continuously optimizes the equipment's exposure time, light intensity, and angle until the image quality score reaches the predetermined standard.

[0054] When all scores meet the high standards set, that is:

[0055] S SNR ≥S SNR,threshold

[0056] S Contrast ≥S Contrast,threshold

[0057] S SSIM ≥S SSIM，threshold

[0058] Among them, S SNR,threshold is the scoring threshold of the signal-to-noise ratio, S Contrast,threshold is the contrast scoring threshold, S SSIM，threshold Structural similarity scoring threshold

[0059] The system will stop further adjusting the device parameters, and the exposure time, light intensity and imaging angle will remain unchanged to ensure that the device parameters match the image quality. At this point, the operating state of the device is stable, and the monitoring process can continue to provide high-quality images, avoiding unnecessary adjustments.

[0060] In step 3, the three-dimensional reconstruction of the deposited layer morphology based on the real-time monitoring data is as follows: the system obtains the dynamic height data of the deposited layer and the position information of the molten pool by real-time monitoring of the laser metal deposition process through multi-angle X-rays. Based on these real-time monitoring data, the idea of ​​subtractive modeling is adopted to gradually reconstruct the three-dimensional morphology of the deposited layer. In order to more accurately simulate the surface characteristics of the deposited layer naturally formed by the surface tension of the molten pool, it is assumed that the cross-section of the deposited layer is an arc, and a coordinate system with X as the scanning direction, Y as the height direction, and Z as the width direction of the deposited layer is established for the reconstruction process.

[0061] In this step, first, a basic geometric model is constructed, which includes the substrate and the area to be activated. The substrate and the area to be activated are designed to fully cover the entire expected deposited layer, ensuring that the geometry of the deposited layer can be accurately reproduced. The size of the area to be activated is slightly larger than the predicted size of the actual deposited layer in order to more accurately capture the dynamic changes during the deposition process.

[0062] The system receives the coordinates of the center of the molten pool M(x i ,y i ,z i ) and molten pool height information H i With this information, define the spherical grab function GetElement(G i ,w), where G i is the center of the spherical grasping area, and w is the width of the sedimentary layer, which is a fixed value in this paper. Calculate the coordinates of the center of the spherical grasping area G i (x i ,y i ,z i ), where x i ,z i is the current molten pool center coordinate, y i =H i -w / 2. In G i The cells in the computational domain are captured as a set Δ to accurately capture the current geometry of the melt pool.

[0063] Furthermore, the intersection of the set Δ and the area to be activated B is calculated to form the current molten pool unit set A to be activated. i , and activate these units to output the three-dimensional morphology of the current molten pool. As the deposition process continues, the system continues to receive new information about the position and height of the molten pool, and continuously updates the area to be activated B=BA based on the new information. i state until the deposition is completed.

[0064] It should be noted that the above embodiments are not intended to limit the protection scope of the present invention, and equivalent changes or substitutions made on the basis of the above technical solutions all fall within the protection scope of the claims of the present invention.

Claims

1. A high-energy X-ray online monitoring device, characterized in that ,include: A laser head (100); four high-energy X-ray devices, namely a first high-energy X-ray device (202), a second high-energy X-ray device (204), a third high-energy X-ray device (206), and a fourth high-energy X-ray device (208), respectively installed at the four corners of the laser head (100); four high-resolution cameras, namely a first high-resolution camera (201), a second high-resolution camera (203), a third high-resolution camera (205), and a fourth high-resolution camera (207), each camera being coaxially installed with a corresponding X-ray device; a base plate (300) located directly below the laser head (100); and a central processing system for receiving and processing image data from the X-ray devices and cameras, and performing image fusion and analysis.

2. The high-energy X-ray online monitoring device according to claim 1, characterized in that: The high-energy X-ray device and the camera dynamically select the working mode according to the printing trajectory of the laser head; when the laser head prints along the oblique trajectory, the two sets of high-energy X-ray devices and cameras at the diagonal position are activated; when the laser head prints along the horizontal or vertical trajectory, the two sets of high-energy X-ray devices and cameras parallel to the printing trajectory are activated.

3. The high energy X-ray online monitoring device according to claim 2, characterized in that: The device activates different X-ray devices and camera combinations according to the printing position. The specific monitoring process is as follows: When printing the second molten pool (302), the second high-energy X-ray device (204) and the fourth high-energy X-ray device (208), as well as the second high-resolution camera (203) and the fourth high-resolution camera (207) are activated. These two sets of devices monitor the printing area from diagonal directions respectively to ensure that the internal structure and surface morphology of the deposited layer can be accurately captured from different angles. When printing the first molten pool (301), the first high-energy X-ray device (202) and the fourth high-energy X-ray device (208), as well as the first high-resolution camera (201) and the fourth high-resolution camera (207) are activated. These devices are located on both sides of the printing area and are responsible for obtaining the thickness of the deposited layer, the dynamics of the molten pool and the external contour morphology to ensure data integrity; When printing the third molten pool (303), the second high-energy X-ray device (204) and the third high-energy X-ray device (206), as well as the second high-resolution camera (203) and the third high-resolution camera (205) are activated. The two combinations simultaneously monitor the deposition process from different angles to ensure that the geometry of the molten pool and the deposited layer can be accurately recorded.

4. The high energy X-ray online monitoring device according to claim 2 or 3, characterized in that: The method of use comprises the following steps: Step 1: Use high-energy X-ray equipment and cameras to capture the surface and internal structure of the molten pool during laser metal deposition in real time; Step 2: The collected image data is transmitted to the central processing system, which uses advanced image processing algorithms to analyze the data in real time and automatically adjust the exposure time, light intensity and imaging angle of the X-ray equipment to optimize image quality. Step 3: Reconstruct the geometric morphology of the sedimentary layer in three dimensions based on the multi-angle image data.

5. The high energy X-ray online monitoring device according to claim 4, characterized in that: Step 2 is as follows: The convolutional neural network architecture combining ResNet and VGG is used to perform multi-level deep feature analysis on X-ray images. First, the quality of the X-ray images collected in real time is scored through the convolutional neural network, and the scoring results covering the image signal-to-noise ratio, contrast and detail resolution are output. Based on these scoring results, the system intelligently adjusts the exposure time, light intensity and imaging angle of the X-ray equipment to optimize the imaging parameters in real time, thereby ensuring that the acquired images have high clarity and contrast. After completing the initial parameter adjustment, the system re-collects images through an intelligent feedback mechanism and scores them again to verify the improvement of image quality after adjustment. If the scoring results show that the image quality still does not meet the preset standards, the system will further iterate and adjust the parameters until the score meets the set requirements. Among them, the neural network must be trained with a special training data set before application to ensure the accuracy of analysis and automatic adjustment. The training data set covers X-ray images collected under various working conditions, and each image is detailedly labeled according to key quality indicators such as signal-to-noise ratio and contrast.

6. The high energy X-ray online monitoring device according to claim 4, characterized in that: In step 2, after the neural network is trained, the system enters the working state and processes the real-time collected X-ray images. After the feature extraction of the convolutional neural network, the image features are input into three independent scoring networks for evaluating SNR, contrast and SSIM respectively. The score output by each scoring network is in the range of 0 to 1, which is used to represent the performance of the image in this quality indicator. The scoring mechanism is realized through the relationship between the features and labels learned by the convolutional neural network during the training process to ensure that the score can accurately reflect the image quality. According to the output results of the scoring network, the system automatically adjusts the parameters of the X-ray equipment. If the SNR score is low, the system will increase the signal strength of the image by increasing the exposure time. The adjustment of the exposure time follows the following formula: T=T min +(1-S SNR )×(T max -T min ) Among them, T max and T min Respectively represent the minimum and maximum exposure time, S SNR Indicates the signal-to-noise ratio score of the image. When the signal-to-noise ratio score is less than 0.95, the system will automatically increase the exposure time to improve the signal strength of the image. If the contrast score is less than 0.95, the light intensity will be reduced to improve the detail clarity of the image. The light intensity adjustment formula is: I=I min +(1-S Contrast )×(I max -I min ) Among them, I max and I min are the minimum and maximum light intensity, S Contrast Indicates the contrast score of the image. When the contrast score is low, the system will automatically increase the light intensity to improve the contrast of the image. If the SSIM score is less than 0.95, the system will adjust the imaging angle to optimize the capture of structural information in the image. The angle adjustment follows the following formula: θ=θ ideal +(1-S SSIM )×Δθ Among them, θ ideal is the ideal imaging angle, Δθ is the step size of angle adjustment, S SSIM SSIM is a structural similarity score. When the SSIM score is less than 0.95, the system will adjust the angle accordingly to improve the structural clarity of the image. Each time the equipment parameters are adjusted, the system re-collects the X-ray image and performs feature extraction and scoring again through the convolutional neural network. If the score does not meet the set standard, the system will continue to adjust the parameters to form a closed-loop feedback control. This feedback mechanism continuously optimizes the equipment's exposure time, light intensity and angle until the image quality score reaches the predetermined standard. When all scores meet the high standards set, that is: S SNR ≥0.95 S Contrast ≥0.95 S SSIM ≥0.95 The system will stop further adjusting the device parameters, and the exposure time, light intensity and imaging angle remain unchanged to ensure that the device parameters match the image quality. At this time, the operating status of the equipment is stable, and the monitoring process can continue to provide high-quality images, avoiding unnecessary adjustments.

7. The high energy X-ray online monitoring device according to claim 4, characterized in that: In step 3, real-time data of the laser metal deposition process is obtained through multi-angle X-ray monitoring, and the three-dimensional morphology of the deposited layer is gradually reconstructed. The system integrates and updates the geometric morphology of the deposited layer in real time based on the image data captured from multiple angles. The reconstruction process adopts a subtractive modeling method, based on the position information and height data of the molten pool, and continuously adjusts the three-dimensional model of the deposited layer. The cross section of the deposited layer is designed to be an arc shape, and a spherical grabbing function is used to gradually activate the cells in the area to be activated, so as to dynamically update the three-dimensional geometric morphology of the deposited layer.

8. The high energy X-ray online monitoring device according to claim 2, characterized in that: In step 3, the cross section of the deposited layer is assumed to be an arc, and a coordinate system is established with X as the scanning direction, Y as the height direction, and Z as the width direction of the deposited layer for the reconstruction process. First, a basic geometric model consisting of the substrate and the area to be activated is constructed. The substrate and the area to be activated are designed to fully cover the entire expected deposited layer to ensure that the geometric morphology of the deposited layer can be accurately reproduced. The size of the area to be activated is slightly larger than the predicted size of the actual deposited layer in order to more accurately capture the dynamic changes in the deposition process. The system receives the coordinates of the center of the molten pool M(x i ,y i ,z i ) and molten pool height information H i , through this information, define the spherical grab function GetElement(G i ,w), where G i is the center of the spherical grasping area, w is the width of the sedimentary layer, and this paper takes a fixed value to calculate the coordinates of the center of the spherical grasping area G i (x i ,y i ,z i ), where x i ,z i is the current molten pool center coordinate, y i =H i -w / 2, in G i is the center of the sphere, w / 2 is the radius, and the cells in the computational domain are captured and formed into a set Δ to accurately capture the current geometry of the melt pool. Calculate the intersection of the set Δ and the area to be activated B to form the current molten pool unit set A to be activated i , and activate these units to output the three-dimensional morphology of the current molten pool. As the deposition process continues, the system continues to receive new information about the position and height of the molten pool, and continuously updates the area to be activated B=BA based on the new information. i state until the deposition is completed.

Images