Positioning Method of Nuclear Fuel Rod Elements Based on Object Detection Network

Abstract

The present invention discloses a positioning method, a computing device, and a readable storage medium for nuclear fuel rod elements. The method includes: acquiring a first image of a nuclear fuel rod element; acquiring a second image of the nuclear fuel rod element after rotating a first predetermined angle; performing data enhancement processing and gradient image enhancement processing on the first image and the second image; inputting the processed images into a trained target detection network to obtain detection results of the first image and the second image, where the detection results include the types and positions of each spacer block on the nuclear fuel rod element; fusing the detection results of the first image and the second image, and determining the type, reference spacer block, and orientation of the nuclear fuel rod element according to the fused image; determining the adjustment angle and adjustment direction for the initial positioning of the nuclear fuel rod element. The technical solution of the present invention can achieve precise positioning of nuclear fuel rod elements.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent online retrieval, and particularly relates to a positioning method, a computing device, and a readable storage medium for nuclear fuel rod elements. Background Art

[0002] The safe and efficient production of nuclear fuel assemblies is an important guarantee for the safe utilization of nuclear energy. The fuel of a heavy water reactor nuclear power plant is a nuclear fuel rod bundle, and the nuclear fuel rod bundle is composed of 37 nuclear fuel rods welded by end plates. Since there are various types of nuclear fuel rods, before the end plate welding, it is necessary to insert the nuclear fuel rods into a prefabricated rod bundle fixture at a standard angle according to the type and front-back of the nuclear fuel, and then perform the end plate welding process. In the end plate welding process, the most core technical point is the rapid and accurate positioning of the nuclear fuel rods. Only by quickly completing the type judgment and accurate positioning of the nuclear fuel rods can the realization of the short-board welding automated production line be ensured.

[0003] At present, the type judgment and positioning of nuclear fuel rods are mainly manually operated. This method has a large manual labor intensity and low efficiency, greatly limiting the improvement of the production efficiency of nuclear fuel rods. At the same time, due to the particularity of nuclear fuel, long-term exposure of personnel to the nuclear fuel production line will cause potential nuclear radiation damage to personnel. Therefore, it is urgent to design a rapid and automatic nuclear fuel rod positioning algorithm, which is of great significance for promoting the localization of automated production equipment for nuclear fuel rod bundles.

[0004] In order to solve the above problems, there is an urgent need for a positioning method for nuclear fuel rod elements to improve the positioning accuracy of nuclear fuel rod elements. Summary of the Invention

[0005] To this end, the present invention provides a positioning method for nuclear fuel rod elements to solve or at least alleviate the problems existing above.

[0006] According to a first aspect of the present invention, there is provided a positioning method for nuclear fuel rod elements, including the steps of: obtaining a first image of a nuclear fuel rod element; obtaining a second image of the nuclear fuel rod element after rotating a first predetermined angle; performing data enhancement processing and gradient image enhancement processing on the first image and the second image; inputting the processed images into a trained target detection network to obtain detection results of the first image and the second image, the detection results including the types and positions of each spacer block on the nuclear fuel rod element, and the types of spacer blocks including thick spacer blocks, thin spacer blocks, and support pads; fusing the detection results of the first image and the second image, and determining the type, reference spacer block, and orientation of the nuclear fuel rod element according to the fused image; and determining the adjustment angle and adjustment direction for the initial positioning of the nuclear fuel rod element according to the determined type, reference spacer block, and orientation of the nuclear fuel rod element.

[0007] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the method further includes: obtaining a third image of the nuclear fuel rod element, where the pixels of the third image are higher than those of the first image and the second image; determining the regions of interest of each spacer block in the third image based on the fused image; extracting the contours of the regions of interest of each spacer block; correcting the center position of the nuclear fuel rod element according to the contours of the regions of interest of each spacer block; using the corrected center position as the unfolding center, unfolding the contour of the region of interest of the reference spacer block to obtain an unfolded diagram of the reference spacer block contour; calculating the actual angle corresponding to the center position of the unfolded diagram of the reference spacer block contour; and determining the precise adjustment angle and adjustment direction of the nuclear fuel rod element according to the actual angle.

[0008] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the target detection network is trained by the following method: obtaining training data including all types of nuclear fuel rod elements; screening the training data based on the angular distribution and quantity of various types of spacer blocks of the nuclear fuel rod elements in the training data, so that a predetermined number of dividing lines are formed in the mean and variance diagrams of the screened training data; performing gradient image enhancement processing and data enhancement processing on the screened training data; and inputting the processed training data into the target detection network for training to obtain a trained target detection network.

[0009] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the detection results of the first image and the second image are multiple images including the positioning frame regions, type markers, and confidence levels of each spacer block in the first image and the second image. The step of fusing the detection results of the first image and the second image includes: adjusting the orientations of the detection results of the first image and the second image to the same orientation, and determining whether the confidence level of each item in the detection results of the first image and the second image is less than a first predetermined value; if the confidence level of an item in the detection results of the first image and the second image is less than the first predetermined value, deleting the detection result of this item and obtaining the remaining detection results; superimposing the remaining detection results to obtain a superimposed image, where the superimposed image includes the detection results of the first image and the second image; and retaining one of the detection results of the first image and the second image in the superimposed image to obtain a fused image.

[0010] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the method further includes: determining the proportion of black in the positioning frame region of each item in the detection results of the first image and the second image; and if the proportion of black is less than a second predetermined value, deleting the detection result of this item.

[0011] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of retaining one of the detection results of the first image and the second image includes: for each spacer block in the nuclear fuel rod element, if the detection results of the first image and the second image both exist in the superimposed image, only retain the detection result of the second image to obtain a fused image.

[0012] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of determining the type of the nuclear fuel rod element includes: determining the type of the nuclear fuel rod element according to the number of spacer blocks of various types in the fused image.

[0013] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of determining the reference spacer block of the nuclear fuel rod element includes: if the type of the nuclear fuel rod element is the first predetermined type, respectively use each spacer block on the nuclear fuel rod element as an alternative reference block, obtain the first angular distribution between the alternative spacer block and each spacer block, calculate the distance between the first angular distribution and the standard angular distribution of the nuclear fuel rod element of the first predetermined type, and use the alternative reference block corresponding to the minimum distance as the reference block of the nuclear fuel rod element; otherwise, determine the reference block of the nuclear fuel rod element according to the type of the nuclear fuel rod element.

[0014] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of determining the orientation of the nuclear fuel rod element includes: if the type of the nuclear fuel rod element is the second predetermined type, obtain the second angular distribution between the reference block and each spacer block on the nuclear fuel rod element; calculate the distance between the second angular distribution and the standard angular distribution of the front and back of the nuclear fuel rod element of the second predetermined type respectively, and determine the orientation of the nuclear fuel rod element as the front or the back according to the calculation results.

[0015] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of determining the adjustment angle of the initial positioning of the nuclear fuel rod element includes: taking the center point of the nuclear fuel rod element as the origin of the coordinate system; determining the center point of the positioning frame area of the reference block, and determining the axis according to the center point and the origin; using the minimum angle of rotating the axis counterclockwise around the origin to the standard position as the adjustment angle of the initial positioning of the nuclear fuel rod element.

[0016] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of determining the region of interest of each spacer block in the third image includes: covering a mask circle on the third image with the mechanical calibration center of the nuclear fuel rod element as the center to obtain a fourth image; determining the prior positions of the centroids of each spacer block in the fourth image according to the fused image; determining the regions where the centroids of each spacer block are located according to the prior positions of the centroids of each spacer block; and in the fourth image, intercepting each spacer block with a predetermined intercept size corresponding to the region to obtain the region of interest of each spacer block in the fourth image.

[0017] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of extracting the contour of the region of interest of each spacer block includes: performing gradient calculation on the region of interest of each spacer block according to the gradient operator corresponding to the region where the centroid of each spacer block is located to obtain the gradient calculation effect diagram of each spacer block; and performing binarization and morphological processing on the gradient calculation effect diagram of each spacer block to obtain the contour of the region of interest of each spacer block.

[0018] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of correcting the center position of the nuclear fuel rod element includes: on the fourth image, drawing three rays from the mechanical calibration center of the nuclear fuel rod element to the thin spacer block contour to obtain three intersection points of the three rays and the thin spacer block contour; fitting the three intersection points to form a fitting circle, and taking the center of the fitting circle as the corrected center position of the nuclear fuel rod element.

[0019] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of determining the precise adjustment angle and adjustment direction of the nuclear fuel rod element includes: taking the difference between the standard angle and the actual angle of the reference spacer block as the precise adjustment angle of the nuclear fuel rod element; if the difference between the standard angle and the actual angle of the reference spacer block is greater than zero, the adjustment direction of the nuclear fuel rod element is clockwise; otherwise, the adjustment direction of the nuclear fuel rod element is counterclockwise.

[0020] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, it further includes: comparing the precise adjustment angle with a threshold angle; if the precise adjustment angle is less than the threshold angle, the adjustment of the nuclear fuel rod element is completed, and the precise adjustment angle and the adjustment direction are notified to the robot; otherwise, return to execute the step of collecting the third image of the nuclear fuel rod element.

[0021] Optionally, in the method for positioning a nuclear fuel rod element according to the present invention, the step of performing data enhancement processing on the first image and the second image includes: respectively rotating the first image and the second image by a second predetermined angle.

[0022] According to a second aspect of the present invention, there is provided a method for positioning a nuclear fuel rod element, comprising the steps of: determining an adjustment angle and an adjustment direction of a reference spacer block in the nuclear fuel rod element according to the method for positioning a nuclear fuel rod element as described above; and adjusting the position of the nuclear fuel rod element according to the adjustment angle and the adjustment direction of the reference spacer block so as to insert the nuclear fuel rod element into an end plate welding fixture.

[0023] According to a third aspect of the present invention, there is provided a computing device, comprising: at least one processor; and a memory storing program instructions, which, when read and executed by the processor, cause the computing device to execute the method for positioning a nuclear fuel rod element as described above.

[0024] According to a fourth aspect of the present invention, there is provided a readable storage medium storing program instructions, which, when read and executed by a computing device, cause the computing device to execute the method for positioning a nuclear fuel rod element as described above.

[0025] According to the technical solution of the present invention, there is provided a method for positioning a nuclear fuel rod element, which acquires a first image and a rotated second image of the nuclear fuel rod element, ensuring that images of each spacer block of the nuclear fuel rod element exist on the acquired images. Then, data enhancement processing and gradient image enhancement processing are performed on the first image and the second image, and through a target detection network, multiple pictures positioning the types and positions of each spacer block are obtained. Then, by fusing the detection results of the first image and the second image, the type and orientation of the nuclear fuel rod can be determined, and the reference spacer block of the nuclear fuel rod element can be determined. Finally, according to the type, orientation and reference spacer block of the nuclear fuel rod, the adjustment angle and adjustment direction of the nuclear fuel rod element are determined, realizing precise positioning of the nuclear fuel rod element.

[0026] Further, taking the determined type, orientation and reference spacer block of the nuclear fuel rod element as prior information, it is possible to quickly locate the regions of interest of each spacer block, correct the center position of the nuclear fuel rod for different regions of interest, and then determine the precise adjustment angle and adjustment direction of the nuclear fuel rod element according to the contour development diagram of the region of interest of the reference spacer block, realizing the fast and precise positioning function of the nuclear fuel rod.

[0027] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other objects, features and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention are hereinafter specifically exemplified. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] To achieve the above and related purposes, certain illustrative aspects are described herein in connection with the following description and drawings, which indicate various ways in which the principles disclosed herein can be practiced, and all aspects and their equivalent aspects are intended to fall within the scope of the claimed subject matter. By reading the following detailed description in conjunction with the drawings, the above and other purposes, features, and advantages of the present disclosure will become more apparent. Throughout the present disclosure, the same reference numerals generally refer to the same components or elements.

[0029] Figure 1 A schematic diagram of a nuclear fuel rod element according to an embodiment of the present invention is shown;

[0030] Figure 2 A schematic diagram of the imaging result of a positioning device for a nuclear fuel rod element according to an embodiment of the present invention is shown;

[0031] Figure 3 A flowchart of a positioning method for a nuclear fuel rod element according to an embodiment of the present invention is shown;

[0032] Figure 4 A flowchart of a positioning method 400 for a nuclear fuel rod element according to another embodiment of the present invention is shown;

[0033] Figure 5 A schematic diagram of gradient image enhancement processing according to an embodiment of the present invention is shown;

[0034] Figure 6 A flowchart of a training method 600 for a target detection network according to an embodiment of the present invention is shown;

[0035] Figure 7 A schematic diagram of screening training data according to an embodiment of the present invention is shown;

[0036] Figure 8 A schematic diagram of the comparison of the improvement in the performance of the target detection network before and after preprocessing the training data according to an embodiment of the present invention is shown;

[0037] Figure 9 A schematic diagram of the exclusion process of non-isolated block regions is shown;

[0038] Figure 10 A schematic diagram of a method for determining the types and positions of isolation blocks according to an embodiment of the present invention is shown;

[0039] Figure 11 A schematic diagram of partial test results for determining the types, positions, and confidence levels of each isolation block on a nuclear fuel rod element according to an embodiment of the present invention is shown;

[0040] Figure 12Shows a schematic diagram of fusing the detection results of a first image and a second image according to an embodiment of the present invention;

[0041] Figure 13 Shows a schematic diagram of a method for determining the type of a nuclear fuel rod element and the adjustment angle of the initial positioning according to an embodiment of the present invention;

[0042] Figure 14 Shows a flowchart of a positioning method 1400 for a nuclear fuel rod element according to another embodiment of the present invention;

[0043] Figure 15 Shows a schematic diagram of the division of the isolation block core region according to an embodiment of the present invention;

[0044] Figure 16 Shows a schematic diagram of the gradient extraction process and effect according to an embodiment of the present invention;

[0045] Figure 17 Shows a schematic diagram of the offset of the center of a nuclear fuel rod according to an embodiment of the present invention;

[0046] Figure 18 Shows a schematic diagram of a method for correcting the center of a nuclear fuel rod element and determining the precise adjustment angle according to an embodiment of the present invention;

[0047] Figure 19 Shows a schematic diagram of the detection time according to an embodiment of the present invention; and

[0048] Figure 20 Shows a schematic diagram of a computing device 2000 according to an embodiment of the present invention. Detailed implementation manners

[0049] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be completely conveyed to those skilled in the art.

[0050] The insertion process of the heavy water nuclear fuel rod is a key link in the production process of the heavy water fuel rod bundle element. To achieve the rapid and correct insertion of the fuel rod, it is necessary to perform precise positioning operations on the fuel rod according to different types of identification before insertion. With the upgrade of the production line and the material management system, the existing fuel rod positioning equipment can no longer meet the speed matching requirements of other links in the production line in terms of efficiency, seriously affecting the production rhythm of the production line. Therefore, the development of a faster and more accurate nuclear fuel rod positioning method is imminent, which has important practical significance for improving the overall operation efficiency of the production line.

[0051] Machine vision is an important and rapidly developing branch of artificial intelligence. In mass repetitive industrial production processes, the use of machine vision detection methods can greatly improve production efficiency and automation. The basic tasks of machine vision are: classification, localization, detection, and segmentation. However, these four tasks are not completely independent. In many cases, there are intersections and hierarchical relationships.

[0052] In machine vision tasks, according to the selection and judgment strategy of features, image processing algorithms can generally be divided into three categories: classical algorithms, machine learning algorithms, and deep learning algorithms. Classical algorithms complete corresponding tasks by manually designing features and formulating corresponding judgment criteria.

[0053] Using the strategy of image unfolding, the types and distribution information of nuclear fuel rod spacer blocks are statistically analyzed, and through a multi-layer classification strategy specified by humans, an accurate classification of 100% of nuclear fuel types and their orientations is finally achieved. Machine learning classification methods do not require manual design of classification judgment strategies. Only the extraction and selection of features from data are needed, and the features are sent into the classifier to automatically complete the classification task. Common methods include decision tree classification, naive Bayesian classifier, classifier based on support vector machine (SVM), neural network method, k-nearest neighbor (kNN), etc. The biggest difference between deep learning methods and the above two methods is that it is an end-to-end process. Just input the image, and the final result can be obtained. There is no need to manually design features and processing strategies. All its processes are learned through the network, and at the same time, its running speed is very fast. Currently, deep learning has achieved very good or even better-than-human performance in text classification and image classification.

[0054] Research shows that in complex background environments, deep learning localization methods have advantages that traditional methods cannot match. A series of object detection networks such as Faster-RCNN, Yolo, and SSD have achieved very good performance in various practical applications. However, an issue that cannot be ignored is that it is very difficult for deep learning localization methods to achieve pixel-level accuracy. Although segmentation networks such as FCNs, SegNet, and U-Net can achieve pixel-level segmentation, their segmentation effects overly rely on the accuracy of manual labeling, and the labeling process is a very large task. At the same time, its learning process is a black-box problem, and the learned features do not have actual physical interpretation attributes. It is a high-dimensional feature representation, and it is very difficult to control its performance in practical applications. It is very difficult to achieve an ideal convergence effect when the samples are insufficient.

[0055] In summary, in order to improve the operating efficiency of the device, the present invention introduces the method of deep learning into the classification and initial positioning tasks of fuel rods. However, in order to ensure the final positioning accuracy, the present invention adopts a contour extraction algorithm to achieve pixel-level positioning accuracy.

[0056] For a specific machine vision application scenario, the implementation effect of the application not only depends on the algorithm, but is also limited by the design of the imaging system. A good imaging system can not only highlight the features of the object to be detected and simplify the difficulty of algorithm development, but also simplify the complexity of the system structure and improve the operating efficiency of the device. The present invention will elaborate on both the imaging system of nuclear fuel elements and the positioning method of nuclear fuel elements in detail.

[0057] The structure of the imaging system of nuclear fuel elements uses the method of telecentric lens plus backlight imaging, and designs a roller adjustment mechanism to achieve the positioning function of fuel rods. This open-angle adjustment mechanism can simplify the loading and unloading process of nuclear fuel rods and improve the imaging efficiency. The strategy of dual-station parallel operation is introduced to simultaneously achieve the functions of initial positioning and precise positioning of fuel rods.

[0058] The positioning method of nuclear fuel elements uses the deep learning target retrieval method to locate the type and position of the isolation blocks and count the results, so as to complete the type judgment and low-precision positioning function of fuel rods. The prior information can be used to quickly locate the ROI (region of interest) of the isolation blocks, and specific contour extraction operators are designed for different ROI regions to improve the contour positioning accuracy, and finally achieve the function of rapid and precise positioning of the isolation blocks.

[0059] Figure 1 Fig. shows a schematic diagram of a nuclear fuel rod element according to an embodiment of the present invention. As Figure 1 shown, the nuclear fuel rod (i.e., the nuclear fuel rod element) is 498 mm long and is divided into four types: #1, #2, #4, and #5. Among them, the #2 and #4 types of nuclear fuel rods have front and back sides. +2 and -2 respectively represent the front and back sides of the #2 type of nuclear fuel rod. Similarly, +4 and -4 respectively represent the front and back sides of the #4 type of nuclear fuel rod. The differences in the types of nuclear fuel rods are mainly reflected in the types, quantities, and distribution angles of the isolation blocks distributed along the circumference of the fuel rod center. There are three types of isolation blocks, including the support pad 110, the thin isolation block 120, and the thick isolation block 130. Among them, the support pad only exists in the #1 and #2 types of nuclear fuel rods. As Figure 1 shown, there is 1 support pad at each end of the #1 and #2 nuclear fuel rods, and its position is collinear with the middle support pad. The present invention will judge which type among the four types of nuclear fuel rod elements the nuclear fuel rod element belongs to and its front and back (i.e., orientation), and determine that the nuclear fuel rod element is adjusted to as Figure 1The adjustment angle and adjustment direction for the shown standard position need to be adjusted so that the nuclear fuel rod element can be adjusted according to the adjustment angle and adjustment direction and adjusted to the standard position. Subsequently, the robot will grip the nuclear fuel rod element with the adjusted position and insert the nuclear fuel rod element into a specific end plate welding fixture to complete the subsequent automatic end plate welding work.

[0060] To achieve the nuclear fuel rod positioning task, the present invention adopts the design concept of a two-station, and provides a positioning device for nuclear fuel rod elements, which splits the type identification and positioning tasks of the nuclear fuel rod elements. A nuclear fuel rod type identification station and a nuclear fuel rod precise positioning station are respectively designed on the positioning device for nuclear fuel rod elements. Among them, the nuclear fuel rod type identification station is used to determine the type and positive / negative of the nuclear fuel rod, and at the same time, perform rough positioning on the nuclear fuel rod; the nuclear fuel rod precise positioning station is used to perform precise positioning on the nuclear fuel rod, and the two stations are the same in structure. The only difference is that the image acquisition device of the nuclear fuel rod type identification station is a camera with a relatively low pixel count, for example: a 40W pixel camera, while the image acquisition device of the nuclear fuel rod precise positioning station is a camera with a relatively high pixel count, for example: a 1200W pixel camera. The reason for such a design is that considering that the type identification station has more computing processes and a large amount of calculation, but does not require a very high pixel accuracy, so a low-pixel camera is used for image acquisition, which can reduce the amount of calculation and save costs. For the nuclear fuel rod precise positioning station, it is necessary to achieve precise positioning of the position of the fuel rod, and it has a relatively high resolution requirement for the contour information of the fuel rod. A large-pixel camera is selected to ensure the image resolution and improve the final pose adjustment accuracy of the device.

[0061] When the positioning device for nuclear fuel rod elements is being detected, the two stations operate in parallel to complete the tasks of type identification and positioning simultaneously. The stepping mechanism feeds the material to the type identification station, and the nuclear fuel rod type identification station acquires the first picture (i.e., the first image); the servo motor drives the nuclear fuel rod to rotate along the axial direction of the workpiece by a first predetermined angle (for example: 180°) through a transmission belt to acquire the second picture (i.e., the second image), and then after processing the two images according to the positioning method of the nuclear fuel rod element of the present invention, calculates the type, positive / negative of the fuel rod, and the adjustment angle and adjustment direction of the rough positioning (i.e., initial positioning), and the servo motor rotates the nuclear fuel rod element to the initial positioning position according to the adjustment angle and adjustment direction of the rough positioning. Then, the stepping mechanism sends the nuclear fuel rod to the precise positioning station and simultaneously completes the feeding of the type identification station. The fuel rod precise positioning station will extract the contour of the reference isolation block to achieve the precise positioning function of the nuclear fuel rod. After the fuel rod is adjusted according to the precise adjustment angle and adjustment orientation, it will notify the robot to grip it to complete the final intubation work and insert the nuclear fuel rod element into the end plate welding fixture. Figure 2 A schematic diagram showing the imaging result of the positioning device for nuclear fuel rod elements according to an embodiment of the present invention is shown.Figure 2 Exemplary imaging results generated by the positioning device for nuclear fuel rod elements are given.

[0062] Figure 3 A flowchart showing the positioning method of nuclear fuel rod elements according to an embodiment of the present invention is shown. As Figure 3 shown, the positioning method of the nuclear fuel rod elements of the present invention includes two parts: initial positioning and precise positioning. In the initial positioning stage, the input is two images (256*256 Pixel) acquired at 0° (the first image) and 180° (the second image). After gradient image enhancement processing and data enhancement processing, each image respectively obtains 4 gradient images, a total of 8 images, which are input into a trained object detection network (for example: YOLOv2) to retrieve the types and positions of each isolation block on the nuclear fuel rod element. Then, the detection results are fused to determine the type and orientation of the nuclear fuel rod, and the reference isolation block on the nuclear fuel rod is found. The positioning angle γ, that is, the adjustment angle of the initial positioning, is solved with the center of the regression box of the object detection, so as to achieve rough positioning with low pixels. In the precise positioning stage, the input image is the third image with high pixels (for example: 3000*3000 Pixel). Using the fuel rod type information obtained in the initial positioning stage, the region of interest (ROI) of the isolation block can be quickly located. Only the contour of the reference isolation block is extracted and the contour is unfolded. By calculating the actual position and the target position of the isolation block, the precise positioning angle α, that is, the precise adjustment angle of the nuclear fuel rod element, can be calculated, and finally the precise positioning of the fuel rod is achieved.

[0063] To more clearly illustrate the content of the initial positioning and precise positioning involved in the positioning method of the nuclear fuel rod elements of the present invention, the following will be combined with Figure 4 to illustrate the content of the initial positioning involved in the positioning method of the nuclear fuel rod elements of the present invention. Figure 4 A flowchart showing a positioning method 400 of a nuclear fuel rod element according to another embodiment of the present invention is shown. The positioning method of the nuclear fuel element of the present invention is executed in a computing device, and the computing device can be the aforementioned positioning device for nuclear fuel rod elements. As Figure 4 shown, method 400 begins at step S402.

[0064] In step S402, a first image of the nuclear fuel rod element is acquired.

[0065] According to an embodiment of the present invention, the nuclear fuel rod element at the initial position is collected by an image acquisition device, and the collected image is used as the first image.

[0066] Subsequently, in step S404, a second image of the nuclear fuel rod element after rotating a first predetermined angle is acquired. The first predetermined angle is 180°.

[0067] Subsequently, in step S406, data augmentation processing is performed on the first image and the second image.

[0068] Specifically, after data augmentation is performed by rotating the first image and the second image by a second predetermined angle (the second predetermined angle is, for example, 90°, 180°, 270°) respectively, the first image, the image obtained by rotating the first image by 90°, the image obtained by rotating the first image by 180°, the image obtained by rotating the first image by 270°, as well as the second image, the image obtained by rotating the second image by 90°, the image obtained by rotating the second image by 180°, and the image obtained by rotating the second image by 270° can be obtained, for a total of eight images. Here, the direction of rotating the first image and the second image can be clockwise or counterclockwise.

[0069] Subsequently, in step S408, gradient image enhancement processing is performed on the images after data augmentation processing.

[0070] Since the images after image data augmentation processing are grayscale images, and the detection task only focuses on the isolation block regions, but there are a large number of black regions of the equipment support wheels and the fuel rod bodies in the images, and these regions have no value for the detection task. Therefore, in order to more accurately determine the types and positions of each isolation block through the object detection network without being interfered by other non-isolation block regions, a method of gradient image enhancement is adopted to preprocess the data.

[0071] According to an embodiment of the present invention, the image to be subjected to gradient image enhancement processing is referred to as the original image. For the original image f(x, y), sobel operators S x , S y in the x and y directions are used for filtering operations, and the gradient map is obtained through the following method

[0072]

[0073]

[0074]

[0075] Wherein:

[0076] Subsequently, the gradient amplitude image A(x, y) is calculated:

[0077]

[0078] The contrast of the gradient amplitude map A(x, y) is relatively low. In order to enhance the features of the regions of interest in the image and reduce the interference of noise and non-regions of interest in the image, gama transformation is used to obtain the enhanced image G(x, y).

[0079] G(x, y) = cA(x, y) γ

[0080] Where c is a constant coefficient with a value of 1.

[0081] Since the regions with higher grayscale in the amplitude image are the edges of the isolated blocks of interest, and the regions with lower grayscale are noise or regions of no interest, let γ = 3 to stretch the regions with higher grayscale levels in the image and compress the regions with lower grayscale levels simultaneously.

[0082] Finally, apply a mask template Mask(x, y) to the image to remove the interfering data located below the image, obtaining the final image I(x, y). Among them, since there are support wheels in the image, the isolated blocks in the lower region may be blocked, thus not having complete isolated block features and possibly causing uncontrollable misidentification situations. Therefore, the method of applying a mask is used to remove them.

[0083] Figure 5 Fig. shows a schematic diagram of gradient image enhancement processing according to an embodiment of the present invention. As Figure 5 shown, after calculating the amplitude image from the original image, a gamma transformation is used to obtain an enhanced image, and then a mask plate is applied to obtain the final image. As can be seen from Figure 5 , through gradient transformation, gamma enhancement, and applying a mask template, the data is greatly compressed and only the isolated block regions of interest are highlighted.

[0084] Subsequently, in step S410, the processed image is input into the trained object detection network to obtain the detection results of the first image and the second image. Among them, the detection results include the types and positions of the isolated blocks on the nuclear fuel rod element, and the types of the isolated blocks include thick spacers, thin spacers, and support pads.

[0085] According to an embodiment of the present invention, the detection results specifically include the bounding box regions Bboxes θ of each isolated block, where the bounding box region is used to mark the position of the isolated block, the type label Lables θ , where the type label is used to mark the type of the isolated block, and the confidence scores Scores θ .

[0086] The specific training method of the object detection network will be described below. Figure 6The flowchart of a training method 600 for an object detection network according to an embodiment of the present invention is shown. During the training process of the object detection network, first, the training data is labeled. To ensure the robustness of the retrieval network, it is necessary to screen and sample the training samples to overcome the problem of uneven training data. Among them, the training samples are grayscale images. To improve the feature extraction ability of the network, the contour features are extracted before putting them into the network for training, so as to accelerate the convergence speed of the network and improve the detection effect of the object detection network. Data augmentation is performed by rotating 90°, 180°, and 270° to increase the number of training samples and improve the generalization ability of the network. Finally, the object detection network is trained.

[0087] As Figure 6 shown, method 600 begins at step S610.

[0088] In step S610, training data including all types of nuclear fuel rod elements is obtained.

[0089] Specifically, the obtained training data includes images of four types of nuclear fuel rod elements, namely #1, #2, #4, and #5. And for nuclear fuel rod elements with different orientations, such as #2 and #4 of type #2 and #4, images of their front and back sides need to be obtained.

[0090] Subsequently, in step S620, based on the angular distribution and quantity of various types of spacer blocks of the nuclear fuel rod elements in the training data, the training data is screened so that a predetermined number of dividing lines are formed in the mean and variance diagrams of the screened training data. The predetermined number is 3.

[0091] Since the nuclear fuel rod elements included in the training data are divided into four types: #1, #2, #4, and #5, in a nuclear fuel rod bundle, the occurrence ratios of each type of nuclear fuel rod element are different, and the types and quantities of spacer blocks included in each fuel rod element are also different. Table 1 shows the occurrence ratios of nuclear fuel rod elements and their spacer blocks in a nuclear fuel rod bundle.

[0092] Table 1:

[0093] Type Support point Thin spacer block Thick spacer block Single rod bundle quantity Ratio #1 1 2 2 6 16.67％ #2 1 3 0 12 33.33％ #4 0 4 1 12 33.33％ #5 0 5 0 6 16.67％ Total single rod bundle 18 126 24 36 Ratio 10.71％ 75.00％ 14.29％

[0094] As shown in Table 1, when preparing the training data, to ensure the generalization ability of the object detection network, the training data needs to be balanced. The data balance is mainly reflected in two aspects:

[0095] First, since the initial pose of the nuclear fuel rod element during loading is random, the angular distribution of the three types of spacer blocks, namely the support pad, thin spacer block, and thick spacer block, in the training set needs to be balanced. Second, the quantity ratios of the three types of spacer blocks, namely the support pad, thin spacer block, and thick spacer block, in the training samples need to be reasonably configured.

[0096] To overcome the influence of the uneven angular distribution of the training set isolation blocks, the training set data is screened. Figure 7 A schematic diagram of screening training data according to an embodiment of the present invention is shown. As Figure 7 shown, in the mean and variance diagrams of the training data before and after screening. Due to the different heights of the isolation blocks, a dividing line will be formed when calculating the mean and variance (marked by the dotted lines in the figure, where the dotted line indicated by 710 is the dividing line of the thick isolation block, the dotted line indicated by 720 is the dividing line of the support pad, and the dotted line indicated by 730 is the dividing line of the thin isolation block). It can be seen from the enlarged view on the far right that when the training data is not screened, the isolation blocks in the mean and variance diagrams of the training data are still visible, and three clear dividing lines cannot be formed. After the training data is screened, three clear dividing lines are formed for different isolation blocks. After screening the training data, the distribution of the samples of the entire training data is more uniform, thus ensuring that the training data contains all angular distributions of the isolation blocks to improve the training effect of the target detection network.

[0097] As shown in Table 1 above, the proportion of thin isolation blocks appears the highest, and the proportions of thick isolation blocks and support pads appear relatively small. When training the target detection network, if the sample numbers of the training data of the support pad, thin isolation block, and thick isolation block are unbalanced, it will affect the detection results of the target detection network. In an ideal situation, the numbers of the three types of isolation blocks in the training data should be the same. However, in actual situations, the isolation blocks do not appear independently and there is interference between them, which causes trouble in precisely controlling the proportion of the isolation blocks in the samples. At the same time, the feature extraction capabilities of the same network for different detection objects are different, and it is not necessary for the training data sample numbers of different target retrieval objects to be exactly the same. Through experiments, it can be found that the network has a lower demand for the training data samples of thick isolation blocks and a higher demand for the training data samples of support pads. Finally, on the premise of ensuring the balanced angular distribution of the isolation blocks. The present invention selects 1382 training samples, which altogether contain 3223 isolation blocks. Among them, the number of support pads is 893 (accounting for 27.71% of the total training data), the number of thin isolation blocks is 1752 (accounting for 54.36% of the total training data), and the number of thick isolation blocks is 578 (accounting for 17.93% of the total training data).

[0098] Subsequently, in step S630, gradient image enhancement processing is performed on the screened training data.

[0099] Since the filtered training data are grayscale images, and only the isolation block areas are concerned during the training process, but there are a large number of black areas of the equipment support wheels and fuel rod bodies in the images, and these areas have no value for the detection tasks of the trained object detection network. Therefore, in order to more accurately determine the types and positions of each isolation block through the object detection network without being interfered by other non-isolation block areas, a method of gradient image enhancement is adopted to preprocess the data.

[0100] According to an embodiment of the present invention, the image to be subjected to gradient image enhancement processing is herein referred to as the original image. For the original image f(x, y), the sobel operators S x , S y in the x and y directions are used for filtering operations, and the gradient map is obtained by the following method

[0101]

[0102]

[0103]

[0104] wherein:

[0105] Subsequently, the gradient amplitude image A(x, y) is calculated:

[0106]

[0107] The contrast of the gradient amplitude map A(x, y) is relatively low. In order to enhance the features of the regions of interest in the image and reduce the interference of noise and non-regions of interest in the image, the gama transformation is used to obtain the enhanced image G(x, y).

[0108] G(x, y) = cA(x, y) γ

[0109] where c is a constant coefficient with a value of 1.

[0110] Since the regions with higher gray levels in the amplitude map are the edges of the isolation blocks of interest, and the parts with lower gray levels are noise or non-regions of interest, so let γ = 3, stretch the regions with higher gray levels in the image, and at the same time compress the parts with lower gray levels.

[0111] Finally, a mask template Mask(x,y) is applied to the image to remove the interfering data located below the image, resulting in the final image I(x,y). Among them, due to the presence of support wheels in the image, the isolation blocks in the lower region may be blocked, thus not having complete isolation block features, which may cause uncontrollable misidentification situations. Therefore, the method of applying a mask is used to remove them. Through gradient change and gamma enhancement, after applying the mask template, the data is greatly compressed, only highlighting the area of the isolation blocks of interest.

[0112] According to an embodiment of the present invention, in order to verify the influence of image preprocessing on the detection result, 1382 pieces of training data are used and tested using an object detection network based on YOLOv2. Among them, the feature extraction network in the object detection network selects MobileNet_v2, the feature extraction layer selects block_14_add, the number of anchor boxes is set to 7, the mini_batch_sizes is set to 16, and the learning rate is set to 0.001. The mean Average Precision (mAP), Average Precision (AP), precision, and recall metrics of the object detection network are statistically analyzed. Table II shows the comparison of the improvement in the performance of the object detection network after preprocessing the training data.

[0113] Table II:

[0114]

[0115] As can be seen from Table 2, for the target detection network without data preprocessing, the indicators of Precision, Recall, AP, and mAP after training are all very good, and the mAP reaches above 0.99. However, the ultimate task of the target detection network of the present invention is to classify and locate nuclear fuel rod elements, which increases the requirements for the robustness and generalization ability of the target detection network. In the research, it is found that the probability of misidentification or missed identification of the spacer block of the support pad type is the highest. The reason lies in the processing method of the support pad. On the nuclear fuel rod with a support pad, the number of support pads is three and they are arranged coaxially. When the image acquisition device collects images parallel to the axis of the fuel rod, due to the straightness of the fuel rod, the coaxially arranged support pads will affect the imaging quality of the support pads and introduce a large amount of noise interference, thus affecting the retrieval effect of the target detection network. The influence of noise on the support pad can be reduced through data preprocessing. After adding data preprocessing, the detection indicators for different spacer blocks have all improved, among which the improvement for the spacer block of the support pad type is the highest. As can be seen from Table 2, the indicators of the thin spacer block perform poorly. The main reason is that there is a lot of data blocked by the support wheel in the samples of the thin spacer block. When marking, some are marked and some are not. It cannot be guaranteed that the occluded thin spacer block can be detected during detection. However, since the present invention adopts the acquisition strategy of two images at 0° and 180° (i.e., the first image and the second image), the occluded spacer block in one image will definitely not be occluded in the other image. Therefore, the phenomenon of the low indicators of the thin spacer block is ignored here.

[0116] Figure 8 FIG. shows a schematic diagram comparing the improvement of the performance of the target detection network before and after preprocessing the training data according to an embodiment of the present invention. As Figure 8 From the comparison, it can be seen that both the target detection network trained with the unpreprocessed training data and the target detection network trained with the preprocessed training data have identified the thin spacer block. However, the target detection network trained with the unpreprocessed data missed the identification of the support pad. While the network trained after data preprocessing has identified the support pad. After data preprocessing, the target detection network only activates the circumferential region where the spacer block exists, unlike when no data preprocessing is performed, where the target detection network activates a large number of regions where there are no spacer blocks, which will increase the complexity of network training and further lead to misidentification and wrong identification of the retrieval network.

[0117] Subsequently, in step S640, data augmentation processing is performed on the image after gradient image enhancement processing.

[0118] It can be seen that the detection accuracy of the target detection network itself is very high. It is difficult to further improve the detection effect by modifying the hyperparameters in the network. If we want to improve the performance of the network, we need to make improvements from the detection strategy. Therefore, during the training process, the method of data rotation is used to achieve data augmentation. The gradient image-enhanced image is rotated by 90°, 180°, and 270° to quadruple the data. Its advantages are as follows: 1. Since the target retrieval algorithm based on YOLOv2 mainly focuses on the data features in the labeled area, after rotation, the features of the isolation blocks change drastically, which increases the richness of the training samples. 2. After the target detection network completes the training on the rotation-augmented data, it has the ability to detect four types of data samples at 0°, 90°, 180°, and 270°. During testing, the input 0° image is rotated to obtain four images at 0°, 90°, 180°, and 270° and input into the network for target retrieval. Finally, the results of the target retrieval are fused and summarized as the final detection result, which improves the robustness of the target detection network in a roundabout way.

[0119] Subsequently, in step S650, the processed training data is used as a training set and input into the target detection network for training to obtain a trained target detection network.

[0120] Specifically, the original preprocessed image, that is, the image after gradient image enhancement, is I 0° (x, y), and through rotation, I θ (x, y) is obtained, where θ = 0°, 90°, 180°, 270°. The rotated image, that is, the training data after data augmentation, is used as a training set and input into the target detection network for training to obtain a trained target detection network. Among them, the feature extraction network in the target detection network selects MobileNet_v2, the feature extraction layer selects block_14_add, the number of anchor boxes is set to 7, mini_batch_sizes is set to 16, and the learning rate is set to 0.001.

[0121] Subsequently, in step S412, the detection results of the first image and the second image are screened. The detection results of the first image and the second image are multiple images including the positioning box regions Bboxes θ 、type labels Lables θ and confidence scores Scores θ of each isolation block in the first image and the second image.

[0122] For the convenience of summarizing the detection results, the orientations of the detection results of the first image and the second image are adjusted to the same orientation. Specifically, the detection results are reversely rotated according to the rotation angle of the original image, and all the detection results are unified in the 0° image direction, which is the same as the direction of the first image, to obtain the rotated bounding box region Bboxes θ ′. The rotated type labels Lables θ and confidence scores Scores θ will not change in value.

[0123] Optionally, the detection results are filtered according to the confidence score. Traverse the detection results of all angular images, compare the confidence scores in the detection results of the first image and the second image with the confidence score threshold C (the first predetermined value). When the confidence score in the detection result is less than C, delete the detection result; otherwise, retain the detection result, as shown in the following formula:

[0124]

[0125] where C is the confidence score threshold with a value of 0.7, and i = 1, 2,..., N θ , N θ is the number of isolation blocks detected in the image corresponding to θ.

[0126] Optionally, non-isolation block regions are excluded. In actual work, there will be a small probability of mislocalization problems in the algorithm. To overcome the occurrence of such small probability events, a pre-judgment mechanism is set up. On the original image f θ (x, y) before preprocessing, intercept the region marked by Bboxes θ ′, which is the bounding box region of the isolation block. Use the binarization method to count the proportion of the black part in the bounding box region. When the proportion is greater than the threshold (i.e., the second predetermined value, which can be set to 0.5), it is considered correctly recognized; otherwise, it is regarded as not recognized successfully, and Bboxes θ ′, Lables θ , Scores θ are updated with the stored data. Figure 9 shows a schematic diagram of the process of excluding non-isolation block regions. As Figure 9 shown, where Figure 9 (a) shows the preprocessed image I 0° (x, y), Figure 9 (b) shows the detection results displayed on the original image f θ (x, y), including the bounding box region, type labels, and confidence scores, Figure 9 (c) shows the intercepted bounding box region of the identified isolation block, Figure 9(d) shows the value of the proportion of the black area counted after binarization in the positioning box area of the intercepted recognized isolation block. As Figure 9 (b) shows, there is abnormal reflection below the isolation block area of the misrecognized thick isolation block (indicated by thick in the figure), and there is a relatively large noise point above. This is because in Figure 9 (a) the preprocessed image shown, the interferences in these two places were not removed during the preprocessing, resulting in misrecognition by the algorithm. As Figure 9 (d) shows, through the statistics of the black area in the recognition area (positioning box area), it can be seen that for the misrecognized isolation block at the bottom, the proportion of the black part is very low. Therefore, through this method, the adverse effects of misrecognition can be effectively excluded.

[0127] After completing the screening of confidence and excluding non-isolation block areas, the final Bboxes θ ′, Lables θ and Scores θ results of the four input images at different angles are obtained. Construct a zero matrix J of uint8 type for statistics θ , and assign the value 65 to the positioning box area detected in Bboxes θ ′. The specific processing is carried out through the following formula:

[0128] J θ (x θ (i):x θ (i)+w θ (i)-1,y θ (i):y θ (i)+h θ (i)-1)=65

[0129] x θ (i)=Bbox θ ′(i,1)

[0130] y θ (i)=Bbox θ ′(i,2)

[0131] w θ (i)=Bbox θ ′(i,3)

[0132] h θ (i)=Bbox θ ′(i,4)

[0133] where i = 1, 2,..., N θ , N θis the number of isolated blocks detected in the image corresponding to the angle θ. θ = 0°, 90°, 180°, 270°.

[0134] Then, the obtained J at different angles θ is accumulated to obtain the image J. Figure 10 FIG. shows a schematic diagram of a method for determining the type and position of an isolated block according to an embodiment of the present invention. Figure 10 (a) shows the detection results at different angles and their corresponding statistical matrix J θ , Figure 10 (b) shows the accumulated image J, its binary image and the centroid. Figure 10 (c) shows the final screening result.

[0135] As Figure 10 shown in (b),

[0136] J is binarized (threshold 0.5). When the number of occurrences of the isolated block is less than 1, the recognition is incorrect and the detection result is ignored. Subsequently, as Figure 8 shown in (b), the centroid of the preselected isolated block region is statistically calculated and denoted as Centroid. The distance between the centroid of each Bboxes θ and Centroid is calculated by the following formula:

[0137]

[0138] where Centroid θ is the centroid position of the isolated block recognized in the image corresponding to the angle θ.

[0139] i = 1, 2,..., N θ , N θ is the number of isolated blocks detected in the image corresponding to θ.

[0140] j = 1, 2,..., N, where N is the number of centroids calculated after binarizing the image J.

[0141] Here, the threshold is set to 10 (empirical value). If d < 10, it is determined as a preselected isolated block of the same type; otherwise, no classification is performed. Finally, among the isolated blocks of the same type, the isolated block with the largest number of occurrences and the highest confidence is selected as the final recognition result. As Figure 10 shown in (c), compared with the detection result before screening ( Figure 10 (a)), the final screened detection result has a more accurate positioning and a higher confidence. For some misrecognition cases, effective exclusion can also be performed.

[0142] After the aforementioned data augmentation and result screening operations, the accuracy of judging the types, positions, and confidence levels of each isolation block on the nuclear fuel rod element has been significantly improved. To test the improvement effect of the target retrieval strategy, 60 image data were randomly selected for testing. Figure 11 FIG. Figure 11 shows a schematic diagram of partial test results for judging the types, positions, and confidence levels of each isolation block on the nuclear fuel rod element according to an embodiment of the present invention. Among them, Figure 11 (a)-(f) in FIG. Figure 11 respectively correspond to 6 test image samples, with a total of 5 columns of data. Among them, the first 4 columns respectively correspond to the detection results of the 0°, 90°, 180°, and 270° images, and the last column is the comprehensive detection result after adopting the aforementioned screening method. As Figure 11 shown, the thick isolation block was not recognized in the 0° image in (a); the thick isolation block was misrecognized as a thin isolation block in the 0° image in (b); the support pad was not recognized in the 270° image in (c), and there were many black areas around the isolation block in this image, with greater interference; the thin isolation block was wrongly recognized as a thick isolation block in the 90° image in (d); interference from non-existent isolation blocks was introduced in the recognition result in (e); the support pad was misrecognized in the 180° image in (f). From Figure 11 the comprehensive detection results in FIG. Figure 11 , it can be seen that after adopting the screening method of the present invention, a large number of misrecognition and wrong recognition problems can be overcome. By increasing the number of input samples of the network and screening the detection results, more accurate, more robust, and more generalized detection results can be obtained.

[0143] Subsequently, in step S414, the detection results of the first image and the second image after screening are fused.

[0144] According to an embodiment of the present invention, the retained detection results after screening are superimposed to obtain a superimposed image, where the superimposed image includes the detection results of the first image and the second image. In the superimposed image, one of the detection results of the first image and the second image is retained to obtain a fused image.

[0145] Specifically, to judge the type of the nuclear fuel rod element and perform the initial positioning of the nuclear fuel rod element, it is necessary to fuse the isolation block recognition results of the fuel rod images after 0° and 180° rotation, and count the type and angular distribution of the isolation blocks. Based on the prior theoretical distribution information of the isolation blocks, the fuel rod type and the reference isolation block position are determined, and finally the initial positioning angle of the fuel rod is calculated. Figure 12 FIG. Figure 12 shows a schematic diagram of fusing the detection results of the first image and the second image according to an embodiment of the present invention. As Figure 12As shown, during the actual detection result fusion process, when adding the detection result of the 0° acquired image (i.e., the first image) to the detection result of the 180° acquired image (i.e., the second image), there will be a situation of duplicate recognition. Here, screening is carried out for the situation of duplicate recognition. Since the image acquired at 180° corresponds to the starting position of the subsequent rotation after initial positioning, when there is an overlap in the recognition of the two images, only the detection result in the 180° acquired image is retained. That is to say, for each isolation block in the nuclear fuel rod element, if the detection results of the first image and the second image both exist in the superimposed image, only the detection result of the second image is retained. In this way, the complete detection results of each isolation block on the nuclear fuel rod element can be obtained, so as to count the quantity and angular distribution of various types of isolation blocks.

[0146] Subsequently, in step S416, according to the fused image, determine the type, reference isolation block, and orientation of the nuclear fuel rod element.

[0147] First, according to the quantity of various types of isolation blocks in the fused image, determine the type of the nuclear fuel rod element. Figure 13 The figure shows a schematic diagram of a method for determining the type of a nuclear fuel rod element and the adjustment angle of initial positioning according to an embodiment of the present invention. As Figure 13 shown, the determination process of the type of the nuclear fuel rod element and the adjustment angle of initial positioning is mainly divided into the following steps. The first step: According to the quantity of various types of isolation blocks, determine the type of the nuclear fuel rod element. As Figure 13 shown, the thick isolation block is represented by N1, the support pad is represented by N2, and the thin isolation pad is represented by N3. Specifically, if the quantity of thick isolation blocks in the fused image is 2, the quantity of support pads is 1, and the quantity of thin isolation blocks is 2, then determine that the type of the nuclear fuel rod element is #1. If the quantity of thick isolation blocks in the fused image is 0, the quantity of support pads is 1, and the quantity of thin isolation blocks is 3, then determine that the type of the nuclear fuel rod element is #2. If the quantity of thick isolation blocks in the fused image is 1, the quantity of support pads is 0, and the quantity of thin isolation blocks is 4, then determine that the type of the nuclear fuel rod element is #4. If the quantity of thick isolation blocks in the fused image is 0, the quantity of support pads is 0, and the quantity of thin isolation blocks is 5, then determine that the type of the nuclear fuel rod element is #5.

[0148] The second step: Select the reference block in the nuclear fuel rod element. After determining the reference block, use the reference block as the first isolation block to statistically calculate the angular distribution θ[θ 1 ,θ 2 ,...,θ N of each isolation block of the fuel rod counterclockwise, where N is the number of isolation blocks.

[0149] If the type of the nuclear fuel rod element is the first predetermined type (the first predetermined type is #5), then by means of the cyclic hypothesis method, each spacer block is successively assumed to be the reference block, and the distance between the angular distribution in the current situation and the angular distribution of the nuclear fuel rod element of type #5 under theory is calculated. The spacer block assumed to be the reference block in the case of the minimum distance is used as the finally determined reference spacer block. Specifically, each spacer block on the nuclear fuel rod element is respectively used as an alternative reference block, the first angular distribution between the alternative spacer block and each spacer block is obtained, the distance between the first angular distribution and the standard angular distribution of the nuclear fuel rod element of the first predetermined type is calculated, and the alternative reference block corresponding to the minimum distance is used as the reference block of the nuclear fuel rod element.

[0150] If the type of the nuclear fuel rod element is not the first predetermined type, that is, the type of the nuclear fuel rod element is #1, #2, or #4, then the reference block of the nuclear fuel rod element is determined according to the type of the nuclear fuel rod element. Specifically, if the nuclear fuel rod element is of type #1, the support pad on the nuclear fuel rod element is used as the reference block. If the nuclear fuel rod element is of type #2, the support pad on the nuclear fuel rod element is used as the reference block. If the nuclear fuel rod element is of type #4, the thick spacer block on the nuclear fuel rod element is used as the reference block.

[0151] The third step: Determine the orientation of the nuclear fuel rod element. According to the embodiments of the present invention, since there are front and back sides of the spacer blocks on the nuclear fuel rod elements of types #2 and #4, therefore, if the type of the nuclear fuel rod element is the second predetermined type (#2 or #4), then the second angular distribution between the reference block and each spacer block on the nuclear fuel rod element is obtained. The distance between the second angular distribution and the standard angular distribution on the front side and the standard angular distribution on the back side of the nuclear fuel rod element of the second predetermined type is calculated respectively. According to the calculation results, the orientation of the nuclear fuel rod element is determined to be the front side or the back side. Among them, the standard angular distribution on the front side and the standard angular distribution on the back side of the nuclear fuel rod element of the second predetermined type are the conventional values of the angular distribution on the front side and the angular distribution on the back side of the standard nuclear fuel rod element of this type. The distance calculation can be performed using the Euclidean distance. Determine the magnitude of the distance between the second angular distribution and the standard angular distribution on the front side and the distance between the second angular distribution and the standard angular distribution on the back side. If the distance between the second angular distribution and the standard angular distribution on the front side is the smallest, the orientation of the nuclear fuel rod element is the front side. If the distance between the second angular distribution and the standard angular distribution on the back side is the smallest, the orientation of the nuclear fuel rod element is the back side.

[0152] Subsequently, in step S418, according to the determined type, reference spacer block and orientation of the nuclear fuel rod element, the adjustment angle and adjustment direction for the initial positioning of the nuclear fuel rod element are determined.

[0153] Step 4: Obtain the angle by which the reference block rotates counterclockwise to the standard position in the coordinate system with the center as the origin. This angle is the rotation angle for the initial positioning of the nuclear fuel rod element.

[0154] According to an embodiment of the present invention, the center point of the nuclear fuel rod element is taken as the origin of the coordinate system, and the center point of the positioning frame area of the reference block is determined. The line connecting the center point and the origin is taken as the axis. The minimum angle by which the axis rotates counterclockwise around the origin to the standard position is taken as the adjustment angle for the initial positioning of the nuclear fuel rod element. Here, since it is a counterclockwise rotation, the adjustment direction is defaulted to the counterclockwise direction. Among them, the standard position can be the standard placement positions of various types of fuel rods as shown in Figure 1 the figure.

[0155] The above describes the rough positioning of the nuclear fuel rod by the positioning method of the nuclear fuel rod element according to the present invention. However, since the image size processed during the initial positioning is small and the pixel accuracy is low, and the positioning of the isolation block is only a regional-level segmentation and cannot reach the pixel-level segmentation, the initial positioning accuracy cannot guarantee the subsequent intubation accuracy of the robot. For this reason, the present invention also provides another positioning method for the nuclear fuel rod element. After the nuclear fuel rod completes the initial positioning, it is transported to the fuel rod pose precise positioning station, and image acquisition work is first carried out. Then, the area of interest of the isolation block is selected, and the contour extraction algorithm is used to extract the contour. To overcome the offset of the fuel rod center caused by the straightness of the fuel rod, the center of the acquired image is accurately solved to improve the pose positioning accuracy of the fuel rod. Subsequently, the obtained contour map is unfolded counterclockwise according to the new center, the angle corresponding to the center position of the unfolded contour map of the reference isolation block is calculated, and the precise positioning adjustment angle α is determined. The calculated angle α is compared with the preset threshold angle β. When α < β, the fuel rod pose precise positioning task is completed, a signal is sent to the robot, and waiting for the robot to grab. When α ≥ β, the pose is adjusted first according to the adjustment angle α, and the acquisition, calculation, and judgment processes are continued until the calculated adjustment angle α is less than the preset threshold angle β. The positioning method of the nuclear fuel rod element of the present invention greatly speeds up the positioning speed and accuracy of the nuclear fuel rod, realizes the automatic positioning of the nuclear fuel rod, solves the bottleneck problem of the end plate welding process in the nuclear fuel rod bundle production line, and greatly improves the production efficiency of the enterprise.

[0156] Figure 14 shows a flowchart of a positioning method 1400 for a nuclear fuel rod element according to another embodiment of the present invention. As shown in Figure 14 the figure, method 1400 starts at step S1410.

[0157] In step S1410, a third image of the nuclear fuel rod element is acquired, where the pixels of the third image are higher than those of the first image and the second image.

[0158] According to an embodiment of the present invention, image acquisition is performed on the nuclear fuel rod element. The image acquisition device at the precise positioning station of the nuclear fuel rod is a camera with a relatively high number of pixels, for example, 12 million pixels, such that the pixels of the third image are higher than those of the first image and the second image. The reason for such a design is that considering that there are many calculation processes and a large amount of calculation in the type recognition station, but high pixel accuracy is not required. Therefore, a camera with low pixels is used for image acquisition, which can reduce the amount of calculation and save costs. For the precise positioning station of the nuclear fuel rod, precise positioning of the position of the fuel rod is required, and it has a relatively high resolution requirement for the contour information of the fuel rod. A camera with a large number of pixels is selected to ensure the image resolution and improve the final pose adjustment accuracy of the device.

[0159] Subsequently, in step S1420, based on the type and orientation of the nuclear fuel rod element, the regions of interest of each spacer block in the third image are determined.

[0160] According to an embodiment of the present invention, first, with the mechanical calibration center of the nuclear fuel rod element as the center, a mask circle is covered on the third image to obtain a fourth image. Among them, the center of the covered mask circle coincides with the mechanical calibration center of the nuclear fuel rod element, and the radius of the mask circle can be 13.1 mm, and the size of the radius of the mask circle can also be appropriately adjusted as needed.

[0161] Then, using the fused image obtained by the aforementioned method 400 as prior information, the prior positions of the centroids of each spacer block in the fourth image are determined. Since the fused image includes the positioning frame regions, type markings, and confidence levels of each spacer block on the nuclear fuel rod element, the center points of the positioning frame regions of each spacer block on the fused image can be used as the prior positions of the centroids of each spacer block in the fourth image.

[0162] Subsequently, the regions where the centroids of each spacer block are located are determined according to the prior positions of the centroids of each spacer block. Specifically, according to the prior positions of the centroids of the spacer blocks, the spacer blocks are divided into two regions: with the mechanical calibration center of the nuclear fuel rod element as the center, assuming a coordinate system, in the coordinate system, the positive half-axis of the x-axis is 0 degrees, the positive half-axis of the y-axis is 90 degrees, the negative half-axis of the x-axis is 180 degrees, and the negative half-axis of the y-axis is -90 degrees. One is the 30-degree region enclosed by the rays along the center to 15 degrees and -15 degrees, the 30-degree region enclosed by the rays along the center to 75 degrees and 105 degrees, and the 30-degree region enclosed by the rays along 165 degrees and -165 degrees, corresponding to Figure 15 the black region in. The other region is the 60-degree region enclosed by the rays along the center to 15 degrees and 75 degrees, and the 60-degree region enclosed by the rays along the center to 105 degrees and 165 degrees, corresponding to Figure 15 the gray region in. Figure 15 shows a schematic diagram of the division of the centroid regions of the spacer blocks according to an embodiment of the present invention. AsFigure 15 As shown, the spacer blocks are divided into two regions according to the position of the spacer block core: a black region and a gray region. Different gradient contour extraction operators are used for the spacer blocks in different regions. Among them, for the black region, 0° and 90° gradient operators S 0 , S 90 are used; for the gray region, ±45° gradient operators S -45 , S 45 are used. For the spacer block ROI (region of interest) whose core falls in the vertical black region, the intercepted size is 500*1000; for the horizontal black region, the intercepted size of the ROI is 1000*500; for the gray region, the intercepted size of the ROI is 1000*1000, and the core of the intercepted ROI region is the prior position of the spacer block core. The intercepted regions of the fuel rods of the same type and positive / negative are fixed and the same.

[0163] Among them,

[0164]

[0165]

[0166]

[0167] Finally, in the fourth image, the predetermined intercepted sizes corresponding to the regions are used to intercept each spacer block, and the regions of interest of each spacer block in the fourth image are obtained.

[0168] According to the embodiments of the present invention, if the core of the spacer block on the nuclear fuel rod element is located in the 30-degree region enclosed by the rays made along the center in the directions of 15 degrees and -15 degrees ( Figure 15 the region corresponding to 1550 in Figure 15 ), or in the 30-degree region enclosed by the rays made along the center in the directions of 165 degrees and -165 degrees ( Figure 15 the region corresponding to 1510 in Figure 15 ), then the intercepted size of the region of interest of the intercepted spacer block is 1000*500. If the core of the spacer block on the nuclear fuel rod element is located in the 30-degree region enclosed by the rays made along the center in the directions of 75 degrees and 105 degrees ( Figure 15 the region corresponding to 1530 in

[0169] Subsequently, in step S1430, the contours of the regions of interest of each isolation block are extracted.

[0170] According to an embodiment of the present invention, the gradients of the regions of interest of each isolation block are obtained according to the gradient operators corresponding to the regions where the centroids of each isolation block are located, and the gradient calculation effect diagrams of each isolation block are obtained. The gradient calculation effect diagrams of each isolation block are binarized and morphologically processed to obtain the contours of the regions of interest of each isolation block. Figure 16 The schematic diagram shows the gradient extraction process and effect according to an embodiment of the present invention. Figure 16 (a) is the original image, Figure 16 (b) is the image after covering with a masked circle, Figure 16 (c) is the image of the region of interest of the intercepted isolation block, Figure 16 (d) is the effect diagram of calculating the gradient according to the present invention, Figure 16 (e) is the effect diagram of calculating the gradient of the prior art. As Figure 16 shown, Figure 16 Taking the nuclear fuel rod of #-4 type as an example, the process and effect of gradient extraction are presented. First, the original image (i.e., the third image) is covered with a masked circle centered on the mechanically calibrated center of the circle. According to the centroid position of the isolation block, 4 ROI regions are extracted: ①(1000*1000), ②(500*1000), ③ and ④(1000*500). Then, according to the position of the centroid of the isolation block, the S-45 and S45 gradient operators are used to calculate the gradient of region ①, and the S0 and S90 gradient operators are used to calculate the gradient of regions ②, ③, and ④. From Figure 16 the comparison effect of (d) and (e), it can be seen that using the prior knowledge of the isolation block position, the designed targeted isolation block gradient operator has a better effect on calculating the contour gradient than the existing algorithm without prior information. The method in this paper has a more significant gradient response to the isolation block contour and is insensitive to the contours outside the isolation block. After calculating the contour gradient, the method of binarization and morphological processing can be used to obtain the final single-pixel contour boundary.

[0171] For the existing contour extraction algorithms, due to the lack of prior information on the distribution of isolation blocks, only multi-scale and multi-directional structural elements can be used to calculate the gradient to balance the effect of calculating the contours of isolation blocks at various angles. Since the positioning method of the nuclear fuel rod element of the present invention has completed the type judgment and initial positioning of the fuel rod at station 1, when accurately positioning the pose of the fuel rod, the type and initial pose of the collected image are known prior information. Using the prior information, the region of interest of the isolation block can be quickly located. By selecting the region of interest, the calculation amount of contour extraction can be reduced, and at the same time, the design of the contour extraction structural element can be simplified.

[0172] Subsequently, in step S1440, the center position of the nuclear fuel rod element is corrected according to the contour of the region of interest of each spacer block.

[0173] Figure 17 FIG. shows a schematic diagram of the offset of the center of the nuclear fuel rod according to an embodiment of the present invention. As Figure 17 shown, when the straightness of the nuclear fuel rod element is large, there will be a large deviation between the actual center O' and the mechanically calibrated center O. At this time, if the accurate adjustment angle α calculated according to the mechanically calibrated center O is used, there will be a large deviation from the actual accurate adjustment angle α'. Therefore, in order to ensure the accuracy of the fuel rod pose adjustment, it is necessary to re-solve the center of the nuclear fuel rod.

[0174] According to an embodiment of the present invention, on the fourth image, three rays are made from the mechanically calibrated center of the nuclear fuel rod element to the contour of the thin spacer block on the nuclear fuel rod element, and three intersection points of the three rays and the contour of the thin spacer block are obtained. The three intersection points are fitted to form a fitted circle, and the center of the fitted circle is used as the corrected center position of the nuclear fuel rod element.

[0175] Subsequently, in step S1450, with the corrected center position as the unfolding center, the contour of the region of interest of the reference spacer block is unfolded to obtain the unfolded diagram of the reference spacer block contour.

[0176] According to an embodiment of the present invention, with the corrected center position as the unfolding center, multiple rays are made to the contour of the reference spacer block in the region of interest, and the distances D and angles θ between each point on the contour of the reference spacer block and the corrected center position O' are statistically calculated in the counterclockwise direction to obtain the unfolded diagram of the reference spacer block contour.

[0177] Subsequently, in step S1460, the actual angle corresponding to the center position of the unfolded diagram of the reference spacer block contour is calculated.

[0178] According to an embodiment of the present invention, a straight line is drawn in the obtained unfolded diagram of the reference spacer block contour such that the straight line has two intersection points with the unfolded diagram. The straight line is, for example, D = 1200. The abscissas of these two intersection points are θ 1 and θ 2 , and the actual angle of the center position of the reference spacer block is the average of the abscissas of the two intersection points (θ 1 +θ 2 ) / 2.

[0179] Subsequently, in step S1470, according to the actual angle, the accurate adjustment angle and adjustment direction of the nuclear fuel rod element are determined.

[0180] According to an embodiment of the present invention, the difference between the standard angle and the actual angle of the reference isolation block is used as the precise adjustment angle of the nuclear fuel rod element. If the difference between the standard angle and the actual angle of the reference isolation block is greater than zero, the adjustment direction of the nuclear fuel rod element is clockwise. Otherwise, the adjustment direction of the fuel rod element is counterclockwise.

[0181] Optionally, the precise adjustment angle is compared with a threshold angle. If the precise adjustment angle is less than the threshold angle, the adjustment of the nuclear fuel rod element is completed, and the precise adjustment angle and the adjustment direction are notified to the robot. Otherwise, return to the step of acquiring the third image of the nuclear fuel rod element. The threshold angle can be set by those skilled in the art, and the present invention does not make specific limitations thereon. According to an embodiment of the present invention, the position of the nuclear fuel rod element is adjusted according to the adjustment angle and adjustment direction of the reference isolation block so as to insert the nuclear fuel rod element into the end plate welding fixture.

[0182] Figure 18 A schematic diagram showing a method for correcting the center of a nuclear fuel rod element and determining a precise adjustment angle according to an embodiment of the present invention is shown. Among them, Figure 18 (a) Schematic diagram for correcting the center of a nuclear fuel rod element, Figure 18 (b) Schematic diagram of the unfolded profile of the reference block, Figure 18 (c) Schematic diagram for determining the precise adjustment angle. As Figure 18 (a) shows,

[0183] Starting from the mechanically calibrated center O of the circle, three rays with fixed angles are made. The angles of the rays are angles predetermined according to the type of nuclear fuel rod. The ray angles vary according to the type of nuclear fuel rod, but the ray angles of the same type of fuel rod with the same front and back are fixed. Three intersection points of the rays and the profile of the thin isolation block are obtained, and a fitted circle (the outer dotted line in Figure 18 (a)) and the re-calibrated actual center O' of the circle are obtained according to the intersection points. As Figure 18 (b) shows, with the actual center O' as the unfolding center of the circle, the profile of the reference isolation block is unfolded. The distances D and angles θ between each point on the profile of the reference isolation block and O' are statistically calculated in the counterclockwise direction to obtain the final unfolded diagram. As Figure 18 (c) shows, the abscissa of the unfolded diagram of the profile of the reference isolation block is the angle θ, and the ordinate is the distance D. By drawing a straight line D = 1200, two intersection points can be obtained. The abscissas of the intersection points are θ 1 , θ 2 . Taking the average value can obtain the actual angle θ r of the center position of the reference isolation block = (θ 1 + θ 2 ) / 2. The standard position of the reference isolation block is θ 0 , then the angle α that the nuclear fuel rod needs to be adjusted = θ 0 - θr When α > 0, the adjustment direction is clockwise; when α < 0, the adjustment direction is counterclockwise.

[0184] To verify the accuracy of type recognition and the positioning accuracy of the positioning method for the nuclear fuel rod element of the present invention, 5 groups of nuclear fuel rod bundles were selected for fuel rod type recognition and positioning accuracy tests. A total of 180 fuel rods were selected, among which there were 30 fuel rods each of #1, #2, #-2, #4, #-4, and #5. Table 3 below statistics the nuclear fuel type recognition, positive and negative judgment, initial positioning, and precise positioning accuracy.

[0185] Table 3:

[0186]

[0187]

[0188] From the experimental results statistically shown in Table 3, it can be seen that the accuracy rates of type recognition and positive and negative judgment of the positioning method for the nuclear fuel rod element of the present invention are 100%. The average positioning accuracy statistically obtained by using the deep learning target retrieval algorithm in the initial positioning stage is 0.617°. The detection accuracies of category, positive and negative, and initial positioning ensure the reliability of prior knowledge when the fuel rod is transported to the precise positioning station of the fuel rod, thereby improving the positioning reliability of the precise positioning station. When the present invention performs precise positioning, a camera with a higher resolution is adopted, and a contour extraction algorithm designed in combination with prior knowledge is used. The final pose positioning accuracy of the fuel rod is 0.035°, and the positioning accuracy is significantly improved. In terms of detection efficiency, a double-station design layout is adopted. For a single fuel rod, its average detection time is the maximum value of the detection times of the type recognition and judgment initial positioning station and the precise positioning station, which is 4.234 s / rod, and the detection efficiency of the equipment is significantly improved.

[0189] The positioning accuracy of the entire system includes two aspects: initial positioning accuracy and precise positioning accuracy. The positioning accuracy is mainly affected by the corresponding positioning algorithm and the accuracy of the positioning adjustment mechanism. Through experiments, the rotation accuracy of the pose adjustment mechanism was measured to be 0.01°, so the positioning accuracy of the entire system mainly depends on the detection accuracy of the positioning algorithm. Among them, the initial positioning accuracy is mainly determined by the positioning accuracy of the isolation block target retrieval algorithm. The precise positioning accuracy is determined by the contour unfolding algorithm. The positioning of the nuclear fuel rod is solved based on the position of the reference isolation block. The reference isolation blocks for #1 and #2 fuel rods are support pads, the reference block for #4 fuel rod is a thick isolation block, and the reference block for #5 fuel rod is determined by a thin isolation block. According to the different reference blocks, the mean and variance of the positioning deviations in the initial positioning stage and the precise positioning stage were statistically analyzed, as shown in Table 4 below:

[0190] Table 4:

[0191]

[0192] As can be seen from Table 4, in the initial positioning stage, the positioning accuracy of the fuel rod with the support pad as the reference block is the worst, and the average absolute error is 0.876°. The positioning errors with the thick spacer block and the thin spacer block as the reference benchmarks are close. After analysis, the main reason is that the positioning angle in the initial positioning mainly depends on the positioning accuracy of the spacer block target detection network. For the support pad, due to its straightness and arrangement characteristics, there is noise in its image, which affects the positioning accuracy of the target detection network. While the thick spacer block and the thin spacer block have clear images and less noise interference, so their initial positioning accuracies are similar and stable. In the precise positioning stage, the positioning accuracy of the support pad is better than that of the thick spacer block and the thin spacer block. In the process of contour extraction, since a contour extraction operator adaptive to the ROI is designed, the noise of the support pad can be overcome. At the same time, because the contour geometry of the support pad is clearer than that of the thick spacer block and the thin spacer block in imaging, its extraction effect is also better, so its precise positioning accuracy is higher. Based on the data in the above table, the average positioning accuracy in the initial positioning stage of the equipment is 0.78°, and its accuracy can ensure that the prior information provided meets the requirements of the precise positioning stage. In the precise positioning stage, the precise positioning accuracy reaches 0.035°, meeting the requirements of precise positioning.

[0193] The positioning method of the nuclear fuel rod element of the present invention adopts double-station parallel detection, so the detection time of the system for a single nuclear fuel rod is the maximum value of the detection times of the two stations of the initial positioning station and the precise positioning station. The present invention also conducts statistical analysis on the detection time in the experiment. Figure 19 shows a schematic diagram of the detection time according to an embodiment of the present invention. As Figure 19As shown in the figure, the detection time of the preliminary positioning station consists of five parts: feeding, image acquisition, 180-degree rotation of the fuel rod, preliminary positioning algorithm, and pose adjustment. Among them, feeding, image acquisition, and 180-degree rotation of the fuel rod are fixed times and will not change with the change of the detection object. The detection time is the average value of test statistics. The pose adjustment time is related to the feeding position of the fuel rod, and the time fluctuation of this part is relatively large. The value shown here is the average value of the test results of 180 fuel rods. The average detection time of the preliminary positioning station is 4.234 s / rod. The detection time of the precise positioning station consists of four parts: feeding, image acquisition, precise positioning algorithm, and positioning adjustment. The interval shown as t1 in the figure is the detection process of one fine-tuning cycle: image acquisition, algorithm calculation, positioning adjustment, and algorithm calculation and accuracy verification again. If the positioning angle is less than the preset value β, the adjustment work ends; otherwise, continue to adjust. Therefore, the total fine-tuning time Ft = N × t1, where N is the number of fine-tuning cycles. After testing, on average, each fuel rod needs to be adjusted N = 1.81 times. Therefore, the average detection time of the precise positioning station is 4.072 s / rod. Considering the double-station synchronous feeding and parallel operation, the overall detection efficiency of the equipment is the detection beat of the preliminary positioning station, which is 4.234 s / rod.

[0194] The present invention proposes a new positioning method for nuclear fuel rod components, which realizes accurate type identification and precise positioning of nuclear fuel rod components, divides the type identification and positioning tasks, and constructs a type identification, preliminary positioning station, and precise positioning station. Through the fast target retrieval ability of the target detection network, the identification and positioning of the fuel rod spacer blocks are completed, and the prior information provided by the target detection network is used to extract the region of interest; for different regions of interest, corresponding contour extraction operators are determined to quickly and accurately extract the spacer block contours, and finally the precise positioning function is realized. The experimental results show that the accuracy of fuel rod type and positive / negative identification of the system is 100%, the precise positioning accuracy reaches 0.035°, and the detection beat reaches 4.234 s / rod.

[0195] According to the technical solution of the present invention, a positioning method for nuclear fuel rod components is provided. The first image and the rotated second image of the nuclear fuel rod components are obtained to ensure that the images of each spacer block of the nuclear fuel rod component exist on the obtained images. Then, the first image and the second image are subjected to data enhancement processing and gradient image enhancement processing. Through the target detection network, multiple pictures with the types and positions of each spacer block located are obtained. Then, by fusing the detection results of the first image and the second image, the type and orientation of the nuclear fuel rod can be determined, and the reference spacer block of the nuclear fuel rod component can be determined. Finally, according to the type, orientation, and reference spacer block of the nuclear fuel rod, the adjustment angle and adjustment direction of the nuclear fuel rod component are determined to realize the precise positioning of the nuclear fuel rod component.

[0196] Further, taking the determined type, orientation, and reference isolation block of the nuclear fuel rod element as prior information, it is possible to quickly locate the regions of interest of each isolation block, correct the center position of the nuclear fuel rod for different regions of interest, and then determine the precise adjustment angle and adjustment direction of the nuclear fuel rod element according to the contour development diagram of the region of interest of the reference isolation block, thereby realizing the fast and precise positioning function of the nuclear fuel rod.

[0197] Figure 20 FIG. 4 shows a schematic diagram of a computing device 2000 according to an embodiment of the present invention.

[0198] It should be noted that Figure 20 the shown computing device 2000 is only an example. In practice, the computing device for implementing the positioning method of the nuclear fuel rod element of the present invention can be a device of any model, and its hardware configuration can be the same as Figure 20 that of the shown computing device 2000 or different. In practice, the computing device for implementing the positioning method of the nuclear fuel rod element of the present invention can add or subtract hardware components of Figure 20 the shown computing device 2000. The present invention places no restrictions on the specific hardware configuration of the computing device.

[0199] As Figure 20 shown, in the basic configuration 2002, the computing device 2000 typically includes a system memory 2006 and one or more processors 2004. A memory bus 2008 can be used for communication between the processor 2004 and the system memory 2006.

[0200] Depending on the desired configuration, the processor 2004 can be any type of processing, including but not limited to: microprocessor (μP), microcontroller (μC), digital signal processor (DSP), or any combination thereof. The processor 2004 can include one or more levels of cache such as a level 1 cache 2010 and a level 2 cache 2012, a processor core 2014, and registers 2016. An example of a processor core 2014 can include an arithmetic logic unit (ALU), a floating point unit (FPU), a digital signal processing core (DSP core), or any combination thereof. An example of a memory controller 2018 can be used with the processor 2004, or in some implementations, the memory controller 2018 can be an internal part of the processor 2004.

[0201] Depending on the desired configuration, system memory 2006 can be any type of memory, including but not limited to: volatile memory (such as RAM), non-volatile memory (such as ROM, flash memory, etc.) or any combination thereof. System memory 2006 can include an operating system 2020, one or more programs 2022, and program data 2024. In some embodiments, program 2022 can be arranged to execute instructions on the operating system by one or more processors 2004 using program data 2024.

[0202] Computing device 2000 may also include an interface bus 2040 that facilitates communication from various interface devices (e.g., output device 2042, peripheral interface 2044, and communication device 2046) to the basic configuration 2002 via a bus / interface controller 2030. Example output devices 2042 include a graphics processing unit 2048 and an audio processing unit 2050. They can be configured to facilitate communication with various external devices such as a display or speakers via one or more A / V ports 2052. Example peripheral interfaces 2044 may include a serial interface controller 2054 and a parallel interface controller 2056, which can be configured to facilitate communication with external devices such as input devices (e.g., keyboard, mouse, pen, voice input device, touch input device) or other peripherals (e.g., printer, scanner, etc.) via one or more I / O ports 2058. Example communication device 2046 may include a network controller 2060, which can be arranged to facilitate communication with one or more other computing devices 2062 via one or more communication ports 2064 through a network communication link.

[0203] The network communication link can be an example of a communication medium. A communication medium can generally embody computer-readable instructions, data structures, program modules in a modulated data signal such as a carrier wave or other transmission mechanism, and can include any information delivery medium. A "modulated data signal" can be a signal in which one or more of its data sets or its changes can encode information in the signal. As a non-limiting example, the communication medium can include wired media such as a wired network or a dedicated line network, and various wireless media such as sound, radio frequency (RF), microwave, infrared (IR), or other wireless media. The term computer-readable medium as used herein can include both storage media and communication media.

[0204] In computing device 2000 according to the present invention, program data 2024 includes instructions for performing the method of positioning a nuclear fuel rod element of the present invention, such that the computing device determines the adjustment angle and adjustment direction of the nuclear fuel rod element by executing the method of the present invention.

Claims

1. A positioning method for nuclear fuel rod elements, comprising the steps of: Obtaining a first image of the nuclear fuel rod element; Obtaining a second image of the nuclear fuel rod element after rotating a first predetermined angle; Performing data enhancement processing and gradient image enhancement processing on the first image and the second image; Inputting the processed images into a trained object detection network to obtain detection results of the first image and the second image, the detection results including the types and positions of each spacer block on the nuclear fuel rod element, and the types of the spacer blocks including thick spacer blocks, thin spacer blocks, and support pads; Fusing the detection results of the first image and the second image, and determining the type, reference spacer block, and orientation of the nuclear fuel rod element according to the fused image; Determining the adjustment angle and adjustment direction for the initial positioning of the nuclear fuel rod element according to the determined type, reference spacer block, and orientation of the nuclear fuel rod element; Obtaining a third image of the nuclear fuel rod element, the pixel of the third image being higher than that of the first image and the second image; Determining the regions of interest of each spacer block in the third image based on the fused image; Extracting the contours of the regions of interest of each spacer block; Correcting the center position of the nuclear fuel rod element according to the contours of the regions of interest of each spacer block; Taking the corrected center position as the unfolding center, unfolding the contour of the region of interest of the reference spacer block to obtain an unfolded contour diagram of the reference spacer block; Calculating the actual angle corresponding to the center position of the unfolded contour diagram of the reference spacer block; Determining the precise adjustment angle and adjustment direction of the nuclear fuel rod element according to the actual angle.

2. The method according to claim 1, wherein, The object detection network is trained by the following method: Obtaining training data including nuclear fuel rod elements of all types; Based on the angle distribution and quantity of each type of spacer block of the nuclear fuel rod element in the training data, screening the training data so that a predetermined number of dividing lines are formed in the mean value and variance diagram of the screened training data; Performing gradient image enhancement processing and data enhancement processing on the screened training data; Inputting the processed training data into the object detection network for training to obtain a trained object detection network.

3. The method according to any one of claims 1 or 2, wherein, The detection results of the first image and the second image are multiple images including the positioning frame regions, type markings, and confidence levels of each spacer block in the first image and the second image. The step of fusing the detection results of the first image and the second image includes: Adjusting the orientations of the detection results of the first image and the second image to the same orientation, and determining whether the confidence level of each item in the detection results of the first image and the second image is less than a first predetermined value; If the confidence level of an item in the detection results of the first image and the second image is less than the first predetermined value, deleting the detection result of this item and obtaining the remaining detection results; Superimpose the retained detection results to obtain a superimposed image, where the superimposed image includes the detection results of the first image and the second image; In the superimposed image, retain one of the detection results of the first image and the second image to obtain a fused image.

4. The method according to claim 3, further comprising: Determine the proportion of black in the positioning frame area of each item of the detection results of the first image and the second image; If the proportion of black is less than a second predetermined value, delete the detection result of this item.

5. The method according to claim 4, wherein, The step of retaining one of the detection results of the first image and the second image includes: For each isolation block in the nuclear fuel rod element, if the detection results of the first image and the second image exist simultaneously in the superimposed image, only retain the detection result of the second image to obtain a fused image.

6. The method according to claim 3, wherein, The step of determining the type of the nuclear fuel rod element includes: Determine the type of the nuclear fuel rod element according to the number of isolation blocks of various types in the fused image.

7. The method according to claim 6, wherein, The step of determining the reference isolation block of the nuclear fuel rod element includes: If the type of the nuclear fuel rod element is a first predetermined type, respectively use each isolation block on the nuclear fuel rod element as an alternative reference block, obtain the first angular distribution between the alternative isolation block and each isolation block, calculate the distance between the first angular distribution and the standard angular distribution of the nuclear fuel rod element of the first predetermined type, and use the alternative reference block corresponding to the minimum distance as the reference block of the nuclear fuel rod element; Otherwise, determine the reference block of the nuclear fuel rod element according to the type of the nuclear fuel rod element.

8. The method according to claim 6, wherein, The step of determining the orientation of the nuclear fuel rod element includes: If the type of the nuclear fuel rod element is a second predetermined type, obtain the second angular distribution between the reference block and each isolation block on the nuclear fuel rod element; Calculate the distance between the second angular distribution and the standard angular distribution of the front and the standard angular distribution of the back of the nuclear fuel rod element of the second predetermined type respectively, and determine the orientation of the nuclear fuel rod element as the front or the back according to the calculation result.

9. The method according to claim 1, wherein, The step of determining the adjustment angle of the initial positioning of the nuclear fuel rod element includes: Take the center point of the nuclear fuel rod element as the origin of the coordinate system; Determine the center point of the positioning frame area of the reference block, and determine the axis according to the center point and the origin; Take the minimum angle of rotating the axis counterclockwise around the origin to the standard position as the adjustment angle of the initial positioning of the nuclear fuel rod element.

10. The method according to claim 1, wherein, The step of determining the region of interest of each isolation block in the third image includes: With the mechanical calibration center of the nuclear fuel rod element as the center, cover a mask circle on the third image to obtain a fourth image; Determine the prior positions of the centroids of the isolation blocks in the fourth image according to the fused image; Determine the regions where the centroids of the isolation blocks are located according to the prior positions of the centroids of the isolation blocks; In the fourth image, intercept each isolation block with the predetermined interception size corresponding to the region to obtain the regions of interest of each isolation block in the fourth image.

11. The method according to claim 10, wherein, the step of extracting the contours of the regions of interest of the isolation blocks includes: Perform gradient calculation on the regions of interest of the isolation blocks according to the gradient operator corresponding to the regions where the centroids of the isolation blocks are located to obtain the gradient calculation effect diagrams of the isolation blocks; Perform binarization and morphological processing on the gradient calculation effect diagrams of the isolation blocks to obtain the contours of the regions of interest of the isolation blocks.

12. The method according to claim 11, wherein, the step of correcting the center position of the nuclear fuel rod element includes: On the fourth image, draw three rays from the mechanical calibration center of the nuclear fuel rod element to the thin isolation block contour to obtain three intersection points of the three rays and the thin isolation block contour; Fit the three intersection points to form a fitted circle, and use the center of the fitted circle as the corrected center position of the nuclear fuel rod element.

13. The method according to claim 1, wherein, the step of determining the precise adjustment angle and adjustment direction of the nuclear fuel rod element includes: Take the difference between the standard angle and the actual angle of the reference isolation block as the precise adjustment angle of the nuclear fuel rod element; If the difference between the standard angle and the actual angle of the reference isolation block is greater than zero, the adjustment direction of the nuclear fuel rod element is clockwise; Otherwise, the adjustment direction of the nuclear fuel rod element is counterclockwise.

14. The method according to claim 1, further includes: Compare the precise adjustment angle with a threshold angle; If the precise adjustment angle is less than the threshold angle, the adjustment of the nuclear fuel rod element is completed, and notify the robot of the precise adjustment angle and the adjustment direction; Otherwise, return to execute the step of collecting the third image of the nuclear fuel rod element.

15. The method according to claim 1, wherein, the step of performing data enhancement processing on the first image and the second image includes: Rotate the first image and the second image by a second predetermined angle respectively.

16. A positioning method for a nuclear fuel rod element, including the steps of: Determine the adjustment angle and adjustment direction of the reference isolation block in the nuclear fuel rod element according to the method according to any one of claims 1 to 15; Adjust the position of the nuclear fuel rod element according to the adjustment angle and adjustment direction of the reference isolation block so as to insert the nuclear fuel rod element into the end plate welding fixture.

17. A computing device, including: At least one processor; and A memory storing program instructions; When the program instructions are read and executed by the processor, the computing device is caused to execute the method according to any one of claims 1-16.

18. A readable storage medium storing program instructions, which, when read and executed by a computing device, cause the computing device to execute the method according to any one of claims 1-16.

Images