Aerospace aluminum-vanadium alloy performance evaluation method based on deep learning

Abstract

The present application relates to the technical field of machine learning, in particular to an aluminum-vanadium alloy performance evaluation method based on deep learning for aerospace, comprising the following steps: obtaining a microscopic image to judge the structure boundary, comparing the gray difference direction before and after disturbance, screening the stable area combined with stress change, extracting the response path and marking the interference path, and generating the aluminum-vanadium alloy performance evaluation result.In the present application, the grain boundary graph is constructed by extracting the gray scale mutation, the disturbance propagation area is identified based on the consistency of the boundary point direction, the coupling position is extracted by superimposing the stress change graph, the corresponding relationship between the disturbance and the stress is established, the channel propagation structure is constructed by extracting the image response path, the interference path is classified and marked to generate the enhanced path set, the grain boundary recognition is more complete, the disturbance and stress matching is more accurate, the response path structure is more continuous, and the linkage analysis capability between the microstructure and the image data is effectively improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of machine learning, in particular to an aluminum-vanadium alloy performance evaluation method based on deep learning for aerospace. BACKGROUND

[0002] The technical field of machine learning relates to research and application of using computer systems to perform specific tasks by automatically learning and improving from data without explicit programming. The core includes algorithm model construction, training and testing data set, feature extraction and selection, model optimization and evaluation mechanism. The main technical system covers multiple branches such as supervised learning, unsupervised learning, semi-supervised learning and reinforcement learning. It is widely used in image recognition, speech recognition, natural language processing, material performance prediction and other fields. The development of this field promotes the integration of intelligent computing models and actual engineering problems. In material science, machine learning has become an important tool for assisting material design, property prediction and performance evaluation. The traditional aluminum-vanadium alloy performance evaluation method refers to the method of evaluating the key performance indicators such as microstructure, hardness, tensile strength and thermal stability of aluminum-vanadium alloy in the background of aerospace application, which usually relies on metallographic analysis, mechanical property test and heat treatment experiment, combined with X-ray diffraction, scanning electron microscope analysis and physical test data. Its process generally includes sample preparation, physical property detection, experimental data recording and manual analysis comparison. The evaluation method is limited by long experimental period, high cost and strong dependence on operators.

[0003] The traditional aluminum-vanadium alloy performance evaluation method relies on micrograph analysis and physical experimental data. The structure boundary recognition depends on static image observation, which is difficult to distinguish the continuous change of microstructure caused by disturbance. The boundary feature segmentation accuracy in image processing is limited, which cannot accurately present the closed boundary relationship between grains. The structure change before and after disturbance lacks effective connection means, and it is difficult to realize the directional extraction of difference area. The evaluation process lacks the dynamic path information linkage between image response, and the path screening does not form the response channel mapping between data, resulting in single analysis dimension of structure disturbance, and the spatial stress change cannot be accurately mapped, which limits the depth and reliability of the evaluation result. SUMMARY

[0004] In order to achieve the above purpose, the application adopts the following technical scheme. The aluminum-vanadium alloy performance evaluation method based on deep learning for aerospace includes the following steps:

[0005] S1: Obtain the micrograph of aluminum-vanadium alloy, judge the structure boundary by the gray difference of adjacent pixels, track the gray mutation points and number the continuous closed boundary, construct the two-dimensional boundary graph, and generate the grain boundary map;

[0006] S2: Based on the grain boundary map, locate the images before and after the disturbance, compare the consistency of the gray-scale difference direction between the boundary point and the adjacent point, construct the disturbance continuity surface, organize the boundary coincidence interference map, and generate the disturbance difference structure.

[0007] S3: Read the stress map corresponding to the location of the disturbance difference structure, filter the region where the disturbance and stress change are consistent, determine the change of the boundary point of the intersection block, analyze the spatial stability, and generate the stress coupling region.

[0008] S4: Locate the stress coupling region in the corresponding region of the feature map, extract the image response path and compare it with the neighboring response values ​​point by point, integrate paths with consistent differences, construct a channel extension map, and generate a response path configuration map.

[0009] S5: Use the response path configuration graph to mark paths, record interference paths and impose restrictions, filter out unmarked paths to form an enhanced mark set, summarize all path structures, and generate aluminum-vanadium alloy performance evaluation results.

[0010] As a further aspect of the present invention, the grain boundary map includes boundary position distribution, grain morphology characteristics, and boundary numbering information; the perturbation difference structure includes grayscale variation regions, directional consistency blocks, and boundary perturbation patterns; the stress coupling region includes stress response coordinates, perturbation stress superposition points, and stability variation intervals; the response path configuration map includes a set of response trajectory points, a group of difference-consistent paths, and a channel extension structure; and the aluminum-vanadium alloy performance evaluation results include perturbed path classification, path enhancement marker set, and channel performance mapping map.

[0011] As a further aspect of the present invention, the disturbance continuity surface refers to a continuous region in the image before and after the disturbance where the direction of the grayscale difference change between the boundary point and its neighboring points remains consistent.

[0012] As a further aspect of the present invention, the change in the boundary point of the intersection block refers to the phenomenon in which the position and shape of the boundary point change in the overlapping block of the region where the disturbance and stress change are consistent.

[0013] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0014] S101: Acquire microscopic images of aluminum-vanadium alloy samples, calculate grayscale difference sequences by calling grayscale values ​​of adjacent pixels, determine whether the grayscale difference exceeds the threshold for structural abrupt change judgment, mark the coordinates of abrupt change pixels, and generate a set of grayscale abrupt change point coordinates.

[0015] S102: Call the gray-scale abrupt change point coordinate set, retrieve the path continuity of the abrupt change point, determine whether the Euclidean distance between adjacent points is less than the closed structure determination threshold, and generate a closed boundary path number matrix.

[0016] S103: Call the closed boundary path number matrix, aggregate the pixel coordinates of the same number and perform the boundary connection operation to establish the corresponding boundary polygon structure and generate the grain boundary map.

[0017] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0018] S201: Based on the position marked by the grain boundary map, read the image before and after the disturbance, extract the gray values ​​of overlapping pixels in the same area, calculate the gray value difference between the boundary point and the adjacent pixel, encode the direction of the difference, and generate a boundary difference direction data group.

[0019] S202: Call the boundary difference direction data group, perform angle difference calculation on the direction encoding between boundary points, determine the direction consistency mark, extract the connected set of pixels corresponding to the continuous and consistent region, and generate a set of perturbed continuous pixel patches.

[0020] S203: Call the set of continuous perturbation pixel patches, aggregate the grayscale differences of pixels before and after the perturbation in the corresponding perturbation region, organize the interference pattern of the boundary region, and generate the perturbation difference structure.

[0021] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0022] S301: Based on the location of the disturbance difference structure, read the stress change map recorded during the loading process, extract the gray value of the corresponding coordinate position of the structure point in the stress map, determine whether there is an overlap relationship between the location of the disturbance structure and the non-zero stress area in the stress map, mark all coordinate points of the overlapping area, and generate a coordinate set of the disturbance stress overlap area.

[0023] S302: Call the coordinate set of the overlapping area of ​​disturbance stress, extract the disturbance difference direction data and stress map gray change direction data in the corresponding area, perform the same direction judgment, filter the coordinate set of points with matching directions, and re-sort according to the structural point index to generate a sequence of points with consistent disturbance stress.

[0024] S303: Call the sequence of consistent disturbance stress points, divide it into continuously distributed sub-blocks, extract the number of boundary points of each sub-block, and calculate the ratio between the range of changes in the number and the area of ​​the region. If the ratio is lower than the spatial stability threshold, mark the block, aggregate all marked blocks and reconstruct the graphic outline to generate a stress coupling region.

[0025] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0026] S401: Based on the corresponding region in the feature map of all coordinate points in the stress coupling region, extract all channel values ​​in the image response matrix, identify the path structure elements corresponding to the coordinate points, establish spatial distribution connection relationships, and generate an image response path coordinate set.

[0027] S402: Call the image response path coordinate set, extract the values ​​of path points and adjacent regions in the image response matrix, perform point-by-point subtraction on the response values, determine whether the result is positive or negative, record the response direction, perform aggregation analysis on the direction sequence, and generate a set of path points with consistent response directions.

[0028] S403: Call the set of path points with consistent response direction, aggregate continuous point columns according to spatial connectivity, expand all aggregated structures into continuous graphic regions in the image channel, record the path boundary contours and mark the structure index, and generate a response path configuration map.

[0029] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0030] S501: Based on all marked path points in the response path configuration diagram, extract the path index information of interference identifiers, add restriction identifiers to the path points, filter out the path set with restriction identifiers, and classify the remaining path numbers uniformly to establish a path enhancement identifier set.

[0031] S502: Call the path enhancement mark set, aggregate the spatial positional relationships of all path points, verify the connection structure of each path, reconstruct the path channel structure morphology, and generate a response channel structure index set;

[0032] S503: Call the response channel structure index set, call the image response matrix data for each channel path, read the channel response value according to the path sequence and perform mean statistics, combine the path spatial distribution attributes and numerical characteristics, calculate the representative index of the path structure, and generate the performance evaluation result of aluminum-vanadium alloy.

[0033] As a further aspect of the present invention, the restriction mark refers to the marker used during the path extraction process to identify path points that are affected by interference and are not suitable for subsequent analysis.

[0034] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0035] In this invention, grain boundary patterns are constructed by extracting grayscale abrupt changes, disturbance extension regions are identified based on the consistency of boundary point orientations, coupling positions are extracted from superimposed stress change maps, a correspondence between disturbances and stresses is established, image response paths are extracted to construct channel extension structures, interference paths are classified and labeled, and an enhanced path set is generated. This achieves more complete grain boundary identification, more accurate disturbance and stress matching, and more continuous response path structures, effectively improving the linkage analysis capability between microstructures and image data. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the steps of the present invention;

[0038] Figure 2 This is a detailed schematic diagram of S1 of the present invention;

[0039] Figure 3 This is a detailed schematic diagram of S2 of the present invention;

[0040] Figure 4 This is a detailed schematic diagram of S3 of the present invention;

[0041] Figure 5 This is a detailed schematic diagram of S4 of the present invention;

[0042] Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0043] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0044] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0045] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0046] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0047] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0048] Please see Figure 1 This invention provides a deep learning-based method for evaluating the performance of aluminum-vanadium alloys for aerospace applications, comprising the following steps:

[0049] S1: Acquire microscopic images of aluminum-vanadium alloy samples, use the gray-level difference between adjacent pixels to determine the structural change boundary, read the gray-level abrupt change point column to perform continuous tracking operation, mark the number of continuous closed boundary structure, construct a two-dimensional boundary graphic according to the number, and generate a grain boundary map.

[0050] S2: Read the images before and after the disturbance according to the position marked by the grain boundary map, call the gray value of the overlapping pixels in the same area, compare the gray value difference between the boundary point and the neighboring point, calculate the directional consistency of all adjacent differences, construct the disturbance continuity surface based on the boundary point direction combination, organize the boundary overlapping interference pattern, and generate the disturbance difference structure.

[0051] S3: Read the stress change map recorded during the loading process based on the set of structural locations with disturbance differences, perform overlap relationship judgment on the corresponding position of each structural point in the stress map, call the points in the overlap area where the disturbance and stress change match for parallel filtering, judge the spatial stability based on the range of change of the number of boundary points of the intersection block, and generate the stress coupling region.

[0052] S4: Based on the location feature map of all coordinate points in the stress coupling region, extract the corresponding image response path structure, perform point-by-point comparison of the image response values ​​of the region and the adjacent regions, record whether the direction of the numerical difference is consistent, and integrate the path points with consistent difference directions to construct the channel extension graph and generate the response path configuration map.

[0053] S5: Use all marked path points in the response path configuration graph, classify and record paths including interference markers and apply restriction markers, filter out unmarked paths to establish a path enhancement marker set, summarize all output path structures to call the response channel graph, and generate aluminum-vanadium alloy performance evaluation results.

[0054] The grain boundary map includes the distribution of boundary positions, grain morphology characteristics, and boundary numbering information. The perturbation difference structure includes grayscale variation regions, directional consistency blocks, and boundary perturbation patterns. The stress coupling region includes stress response coordinates, perturbation stress superposition points, and stability variation intervals. The response path configuration map includes the response trajectory point set, difference-consistent path group, and channel extension structure. The performance evaluation results of aluminum-vanadium alloys include perturbed path classification, path enhancement marker set, and channel performance mapping map.

[0055] Please see Figure 2 The specific steps of S1 are as follows:

[0056] S101: Acquire microscopic images of aluminum-vanadium alloy samples, calculate grayscale difference sequences by calling grayscale values ​​of adjacent pixels, determine whether the grayscale difference exceeds the threshold for structural abrupt change judgment, mark the coordinates of abrupt change pixels, and generate a set of grayscale abrupt change point coordinates.

[0057] First, the brightness and focal length of the metallurgical microscope's light source were adjusted to ensure a clear image with uniform brightness and no obvious reflective areas. Then, the sample surface was mechanically polished and chemically etched to reveal the grain structure. Images were acquired under the microscope. After image acquisition, an image processing program was used to convert the color image to a grayscale image. During the conversion, the RGB three-channel values ​​of each pixel were converted into a single grayscale value. Then, starting from the top left corner of the image, all pixels were traversed row by row, and the grayscale value of each pixel was read sequentially. The grayscale values ​​of the adjacent pixels to the right or below were selected, and the absolute value of the difference between the two was calculated as the grayscale difference. The sequence of grayscale differences between all pixels and their adjacent pixels was recorded, and then the sequence was statistically analyzed. The average and standard deviation of all differences in the column are used to set a structural mutation judgment threshold based on the uniformity of background brightness and the contrast of grain boundaries in the image. It is usually set as the sum of the average value and twice the standard deviation. If the gray value difference between a pixel and its neighboring pixels is greater than the threshold, the pixel is judged to be at the position of structural mutation. The coordinates of the mutation point are recorded to form an initial gray value mutation point set. For example, in an image, the average gray value difference is 5 and the standard deviation is 4, so the judgment threshold is set to 13. If the gray value difference between a point and its right neighbor is 15, it means that the threshold is exceeded and it is regarded as a mutation point. Its coordinates are recorded into the set. This process is repeated for the entire image to obtain a complete set of gray value mutation point coordinates.

[0058] S102: Call the grayscale abrupt change point coordinate set, retrieve the path continuity of the abrupt change point, determine whether the Euclidean distance between adjacent points is less than the closed structure determination threshold, and generate a closed boundary path number matrix.

[0059] Starting from the first mutation point, the coordinates of all points are read sequentially. For each mutation point, the coordinates of its eight surrounding pixels in eight directions are extracted. The system then searches for other mutation points in the set that are close to it, analyzing the coordinate differences between these points one by one. For any two adjacent mutation points, the difference between their row and column coordinates is calculated. The geometric relationship formed by these two differences is used to determine whether they are continuous path points. If the difference is between three and five pixels, they are considered continuous path points, and these two points are grouped under the same number. Each time a continuous path is formed, a new number is assigned, and the search continues from this numbered path to find connected points. The next mutation point is identified, and the above operation is repeated until the current path can no longer be extended outward. At this point, the path is considered to have ended. If the distance between the starting point and the ending point of the path does not exceed the threshold of three pixels, the path is considered to form a closed structure. Otherwise, it is judged as a non-closed path and is no longer recorded. After traversing and analyzing all mutation points in the entire image, a set of closed paths with multiple different numbers is obtained. For example, if a path consists of 15 mutation points and the coordinates of the starting point and the ending point are less than two pixels apart, it is determined to form a closed boundary and the path is numbered as path 1. The final result generates a closed boundary path number matrix, in which each mutation point corresponds to its path number in the matrix.

[0060] S103: Call the closed boundary path number matrix, aggregate the pixel coordinates of the same number and perform the boundary connection operation to establish the corresponding boundary polygon structure and generate the grain boundary map;

[0061] The coordinates of all mutation points corresponding to each number are read. Points with the same number are considered a group of paths. These points are then connected in the order they appear in the image to construct a path point sequence. During this process, the connection order is chosen based on minimizing the coordinate changes between adjacent points, avoiding path jumps caused by crossing non-adjacent areas. The connection operation is performed sequentially, from the first point to the last. If the path is closed, the endpoint is connected back to the starting point to form a complete closed polygon. After performing this connection operation on each closed path, a series of boundary polygons are obtained. Each polygon represents a grain boundary region. To ensure the accuracy of the boundary map, paths with fewer than a preset threshold (e.g., fewer than 10 points) are not included in the polygon construction because they cannot accurately represent the grain boundary contour. After all paths are constructed, all boundary polygons are superimposed and drawn onto an image layer. For example, the abrupt change point in path number 3 starts from image coordinates (100, 200), passes through a series of points, and returns to (100, 200), forming a closed path, and is finally drawn as a complete boundary region in the map. After all paths are processed, the complete grain boundary map of the aluminum-vanadium alloy sample is obtained.

[0062] Please see Figure 3 The specific steps of S2 are as follows:

[0063] S201: Based on the position marked by the grain boundary map, read the image before and after the disturbance, extract the gray values ​​of overlapping pixels in the same area, calculate the gray value difference between the boundary point and the adjacent pixel, encode the direction of the difference, and generate a boundary difference direction data group.

[0064] First, the perturbation analysis area is determined. Based on the polygonal structure of each boundary recorded in the grain boundary map, the pixel coordinates on each boundary path are extracted and used as the base area for analysis points. Then, the images before and after perturbation are read to ensure consistency in resolution, grayscale depth, size, and pixel arrangement between the image pairs. Position alignment is performed to avoid image offset causing errors. After image registration, the pixel areas covered by the aforementioned boundary paths are extracted from the two images, and the grayscale values ​​corresponding to all boundary points within these areas are obtained. The grayscale values ​​of each boundary point at the two time points are then mapped one-to-one and categorized by boundary path number. Points under the same number form a set of comparison data. Then, for each boundary point in the image before perturbation, multiple adjacent pixels in multiple directions are selected as references, and these adjacent pixels are extracted. The grayscale values ​​are calculated and subtracted from the grayscale values ​​of the current boundary points one by one to obtain grayscale differences in multiple directions. The differences are judged: if the difference is positive, the current direction is considered to be an enhancing trend; if the difference is negative, it is considered to be a weakening trend. The trend results are recorded in eight directions in sequence, and the trends are encoded in binary to form a fixed-length direction encoding sequence. The binary sequence is then converted into decimal values ​​for subsequent processing. For example, if the grayscale value of a boundary point in the image before the disturbance is 130, the grayscale value of the pixel above it is 128, the grayscale value below it is 135, the grayscale value to the left is 132, and the grayscale value to the right is 125, then the upward direction is weakening, the downward direction is enhancing, the left is enhancing, and the right is weakening. Accordingly, in the encoding, the upward direction is recorded as 0, the downward direction as 1, the left as 1, and the right as 0. This process is repeated to complete the encoding and recording of all directions, and finally a complete boundary difference direction data group is formed in the entire boundary area.

[0065] S202: Call the boundary difference direction data group, perform angle difference calculation on the direction encoding between boundary points, determine the direction consistency mark, extract the connected set of pixels corresponding to continuous and consistent regions, and generate a perturbed continuous pixel patch set.

[0066] First, the grayscale change trend of each boundary point in space is analyzed based on its directional encoding. The directional pattern represented by each encoding is read, and the encoding values ​​of adjacent points are obtained from the boundary path. The consistency of directional changes between the current point and its adjacent points is compared one by one. For directional values ​​at the same position in the encoding, if the two points have the same trend in more than half of the directions, they are judged to be directionally continuous, and a directional consistency flag is established. The search for the next boundary point continues in this direction, and the directional consistency between the new point and the previous point is compared again. If the condition of continuous consistency is met, it is continuously included in the current consistent region set. If the continuous consistency is interrupted, the expansion of the region ends, and the search for other consistent directional regions starts again from the new point. For each continuous consistent region, its starting point is recorded. The system identifies the starting and ending points of a continuous region and counts the number of boundary points within it. If the number of points in the continuous region exceeds a preset threshold (e.g., 5 or more), it is considered a valid perturbation continuous region. Otherwise, the region is considered to be due to noise and is discarded. Each perturbation continuous region that passes the screening is integrated into a pixel connected set, and its actual coordinate position in the original image and its boundary path number are recorded. For example, in the boundary path with grain number 12, there is a segment of 7 consecutive pixels from point (215, 322) to (221, 322) whose directional encoding all show a downward increasing trend. The directional encoding matching degree is high and meets the consistency judgment rule. This segment is identified as a set of perturbation continuous pixel patches. Finally, all regions in the image that meet the conditions are collected to form a perturbation continuous pixel patch set.

[0067] S203: Call the set of continuous pixel patches of disturbance, aggregate the gray-scale differences of pixels before and after disturbance in the corresponding disturbance region, organize the interference pattern of the boundary region, and generate the disturbance difference structure;

[0068] The algorithm sequentially reads the pixel coordinates contained in each patch and locates the corresponding positions of these points in the images before and after perturbation. It then obtains the grayscale values ​​of these points in both images, calculates the grayscale differences of all pixels in each patch, and obtains the grayscale change information of all points in that patch. The algorithm also statistically analyzes the overall distribution characteristics of all grayscale differences within a patch. For each patch, a difference aggregation process is performed, merging all grayscale difference data within it, summing them by pixel count to calculate the average, and extracting the maximum and minimum value ranges. Based on the average grayscale difference obtained after aggregation, the patches are classified according to preset intervals, for example, setting 0-10 as the first-level difference interval, and 11... -30 represents the second-level difference range, and 31 and above represents the third-level difference range. If the average grayscale difference of all pixels in a certain area is 26, it indicates that the area belongs to the second-level perturbation level. This classification result is marked and attached to the area data. Then, according to the boundary path number to which each area belongs, the perturbation level information of all areas is reclassified and integrated, and mapped back to the corresponding position in the grain boundary map. The perturbation level is displayed on the boundary structure map by color filling or layer overlay. The perturbation areas in each grain region are numbered and uniformly organized into a summary of boundary region interference graphics data. The final output is a structured perturbation difference structure, showing the perturbation situation of the corresponding boundary structure in the image.

[0069] Please see Figure 4 The specific steps of S3 are as follows:

[0070] S301: Based on the location of the disturbance difference structure, read the stress change map recorded during the loading process, extract the gray value of the corresponding coordinate position of the structure point in the stress map, determine whether there is an overlap between the location of the disturbance structure and the non-zero stress area in the stress map, mark all coordinate points in the overlapping area, and generate a coordinate set of the disturbance stress overlap area.

[0071] First, obtain the coordinates of all structural points with perturbation features in the constructed perturbation difference structure layer, and map them one-to-one with the complete image coordinate system. Then, read the stress change map recorded during the loading process to ensure that the stress map and the perturbation difference structure map have a unified pixel distribution and coordinate reference. Use image registration to perform synchronous correction processing on the two images. After correction, retrieve the pixel position corresponding to the perturbation structure point in the stress map point by point, and extract the corresponding gray value information. This gray value reflects the stress level experienced by the point in the image. A gray value of 0 indicates a stress-free area, and a value greater than 0 indicates a stress-distributed area. Traverse all perturbation structure coordinate points and map each point in the stress map. The corresponding grayscale value is read and determined to be non-zero. If the grayscale value of a point is non-zero, it is considered that the point is in the stress participation area, and the result is that there is an overlap relationship. If the grayscale value is 0, it is determined that there is no overlap, and the judgment process of the next point continues. All coordinate points that meet the overlap conditions are recorded in the overlap set. Finally, the stress overlap screening process of the entire disturbance structure diagram is completed, and the coordinate set of the disturbance stress overlap area is generated. For example, if a disturbance structure point is located at coordinates (226, 318), its corresponding grayscale value in the stress diagram is 42, which is greater than 0. This point is recorded as an overlap point. If a total of 137 valid overlap coordinate points are obtained in the final screening process, the set contains 137 spatial coordinate data.

[0072] S302: Call the coordinate set of the overlapping area of ​​disturbance stress, extract the disturbance difference direction data and stress map gray change direction data in the corresponding area, perform the same direction judgment, filter the coordinate set of points with matching directions, and re-sort according to the structural point index to generate a sequence of points with consistent disturbance stress.

[0073] Each overlapping point's coordinates are extracted from the set, and the corresponding perturbation difference direction code value is located using this coordinate index. Then, using this coordinate as a reference, the grayscale values ​​of adjacent pixels in eight directions around the stress map are obtained. The direction of grayscale change between each adjacent point and the center point is analyzed, and the trend is determined to be either increasing or decreasing. The change state is recorded in each direction, forming a set of direction codes reflecting the stress change trend. This stress direction code is compared with the original perturbation difference direction code, using a positional correspondence judgment method. That is, the change trends in the two direction codes are judged at corresponding positions in eight directions. If the change direction is the same in five or more directions, the point is identified as having a perturbation direction consistent with the stress. If the force direction is consistent, the point is recorded as a point with consistent direction; otherwise, the point is discarded. After filtering all overlapping points, a set of points with consistent direction is obtained. Then, according to their original number or path index value in the perturbation difference structure, the points in the set are sorted so that they can reflect the coherent spatial distribution in the image after sorting. For example, all points are arranged according to the rule of horizontal priority, so that points located on a horizontal path are arranged in adjacent positions in the set. Finally, the filtered points are rearranged to generate a sequence of points with consistent perturbation stress. For example, after filtering and rearrangement, points (120, 225), (121, 225), and (122, 225) are arranged continuously and maintain a consistent direction, and are uniformly included in the set of point sequences.

[0074] S303: Call the consistent stress point sequence, divide it into continuously distributed sub-blocks, extract the number of boundary points of each sub-block, and calculate the ratio of the range of number variation to the area of ​​the region. If the ratio is lower than the spatial stability threshold, mark the block, aggregate all marked blocks and reconstruct the graphic outline to generate the stress coupling region.

[0075] First, points within 2 pixels of each other and with a continuous arrangement are divided into sub-blocks. All points in the sequence are traversed, and a network of connections between them is established. If any two points are adjacent in a row or column, or if the diagonal distance is √2 pixels, they are considered to belong to the same sub-block and added to the current block. This process continues until no more connected points remain, completing the division of a sub-block. The next undivided point is then processed, and the above process is repeated. After all points have been divided into sub-blocks, a boundary point count operation is performed on each sub-block. This involves finding the number of pixels in the block that border at least one external pixel and recording them as the boundary point count of that sub-block. Then, the total number of pixels in each sub-block is counted to obtain... The actual area of ​​the block depends on the total number of pixels contained in the sub-block. Then, the ratio between the number of boundary points and the area is calculated. The boundary complexity index of the sub-block is obtained by dividing the number of boundary points by the area. The spatial stability threshold is set to 0.45. If the ratio of a sub-block is less than this threshold, it is considered that its boundary distribution is smooth and the internal disturbance range is concentrated. It is judged as a stress structure unstable region and is marked. All sub-blocks that meet the condition that the ratio is less than the threshold are classified into the abnormal region set, and their corresponding boundary points are extracted. The contour is constructed according to the coordinate order of the boundary points to form a closed polygon region. Finally, all abnormal sub-blocks are merged and the overall graphic boundary structure is constructed to generate the stress coupling region graphic.

[0076] Please see Figure 5 The specific steps of S4 are as follows:

[0077] S401: Based on the corresponding region in the feature map of all coordinate points in the stress coupling region, extract all channel values ​​in the image response matrix, identify the path structure elements corresponding to the coordinate points, establish spatial distribution connection relationships, and generate the image response path coordinate set;

[0078] First, iterate through all valid pixel coordinates in the stress coupling region. For each point, perform a spatial mapping operation to map its position to the feature map coordinate system where the image response matrix is ​​located. Then, read the response values ​​of that point in all channels from the image response matrix according to the coordinates. The response matrix usually contains multiple physical or computational attribute channels, such as grayscale intensity channels, texture direction channels, local contrast channels, etc. Each channel has an independent numerical distribution on the image, and each point has a unique response value in each channel. After reading, combine the channel values ​​to form the response vector of that point. Next, analyze whether the point constitutes a continuous path structure element in the feature map. The judgment method is: if the response value change of this point and at least three points in the eight adjacent directions remains continuous in a certain channel (…), then… If the change in response value is within a set threshold range, then the point is considered a path structure element and marked as a path point. Then, adjacent path points with the same direction are sequentially retrieved. By checking the trend and directional distribution of response value changes, the connection relationship between points is established, and finally a path structure connection map is formed. For example, the grayscale value of point (245, 360) is 135, and the values ​​of its right point (245, 361) and the lower point (246, 360) are 136 and 134 respectively. The change is within the continuity judgment threshold ±2, so the point is considered to form a path connection with the surrounding area. The connection direction and channel response information of the point and its adjacent points are recorded. After processing all points one by one, the image response path coordinate set is finally generated, which contains the spatial position and structural connection information of all path points.

[0079] S402: Call the image response path coordinate set, extract the values ​​of path points and adjacent regions in the image response matrix, perform point-by-point subtraction on the response values, determine whether the result is positive or negative, record the response direction, perform aggregation analysis on the direction sequence, and generate a set of path points with consistent response directions.

[0080] The coordinates of each path point are extracted sequentially, and its location is determined in the image response matrix. Then, the raw response values ​​of that path point in all channels are extracted, constructing a local region centered on that point. The neighboring points in eight directions are extracted, and their response values ​​are read from the response matrix. A subtraction operation is performed between the response values ​​of each neighboring point and the center point in each direction, recording the sign of the difference. If the neighboring point's response value is greater than the center point's, the difference is negative, indicating a downward direction; otherwise, it is positive, indicating an upward direction. The difference states in the eight directions are sequentially arranged into a direction sequence, representing the response change trend of the current path point. All points in the path set are then processed sequentially to complete the process. After constructing the path point direction sequence, further aggregation analysis is performed on the direction sequence to find points with continuous and consistent response directions in the image response path. The judgment criterion is: if several continuous path points have consistent positive and negative change states in at least five directions, these points are considered to be a direction-consistent group, which is then classified and marked with a direction-consistent flag. For example, path points (310, 402), (311, 403), and (312, 404) are all response value decreasing directions in seven directions, so they are judged to be path points with consistent directions. Their start and end coordinates are recorded and included in the response direction consistent path point set. After completing the direction judgment and clustering processing of all paths, a point set data structure containing multiple direction-consistent path segments is obtained.

[0081] S403: Call the path point set with consistent response direction, aggregate continuous point columns according to spatial connectivity, expand all aggregated structures into continuous graphic regions in the image channel, record the path boundary contours and mark the structure index, and generate a response path configuration map.

[0082] For each path segment, a spatial connectivity aggregation operation is performed. First, the image coordinates of all points in each path segment are extracted, and a two-dimensional coordinate grid is established. Adjacent points are determined according to the horizontal, vertical, and diagonal directions in the image space to see if they are within a continuous pixel distribution range. If any two points are adjacent in the horizontal, vertical, or diagonal direction without gaps, they are considered connected points and aggregated into the same connected sub-region. After traversing all points and performing the aggregation operation, the original multiple independent path segments are combined into continuously distributed image sub-structure regions. Then, a contour detection operation is performed on the spatial edges of each connected structure to identify which points are on the outer edge of the structure, i.e., those points have no adjacent path points on at least one side. These edge points are extracted to form a complete structural wheel. The contour line is arranged sequentially according to the connection relationship between the points to form a closed boundary path. Then, a unique structure index number is assigned to each connected structure. This number is defined according to the spatial distribution of the path, for example, the numbering is sequentially increased from the upper left corner to the lower right corner. The structure index and boundary path information are recorded and marked in the response image channel. Finally, all path structure information is integrated to generate a response path configuration map. This map reflects the spatial distribution characteristics and structural organization of all regions with the same response direction. For example, the region number R07 corresponds to a point set range of (200, 340) to (208, 348), and the contour line length is 32 points. The structure number and response path are stored together in the configuration map structure to form a complete output.

[0083] Please see Figure 6 The specific steps of S5 are as follows:

[0084] S501: Based on all marked path points in the response path configuration diagram, extract the path index information of interference identifiers, add restriction markers to the path points, filter out the path set with restriction markers, and classify the remaining path numbers uniformly to establish a path enhancement marker set;

[0085] First, the structural information of all response paths in the graph is traversed, and the path number index corresponding to each path is extracted. Then, a structural state identification operation is performed on this number, including whether interference features such as inconsistent direction, non-closed boundaries, or stress response mismatch occurred in the preceding stages. Path numbers possessing any of these features are added to the interference identifier set and marked as restricted path numbers. Subsequently, a comparison operation is performed on each path in the response path configuration graph according to its index. For paths matching the restricted identifier number, a restricted marker status value is added to all points on that path to indicate that the path does not have the stability basis to participate in subsequent enhancement processing. Finally, all path structures with restricted markers are filtered out from the original path structure set to ensure that these paths are no longer called in subsequent processing. After screening, all numbers in the remaining path structure are reordered to ensure continuity. A set of path numbers without restrictions is established as the basis for path enhancement candidates. Then, the path set is uniformly marked with enhancement identifiers. A marker field is added to the path data structure to indicate that it is a member of the enhanced object path set. For example, path numbers P04, P08, and P13 are marked as restricted paths in the structure identification due to messy direction distribution, deviation in response value, or too many breakpoints. After adding the restriction identifier, they are excluded from the original set. Other paths such as P01, P03, P06, and P10 are all structurally coherent and have stable responses. They are included in the enhanced marked path set. Finally, the initial screening of the path structure and the establishment of the enhancement set are completed, resulting in the path enhancement marker set.

[0086] S502: Call the path enhancement tag set, aggregate the spatial positional relationships of all path points, verify the connection structure of each path, reconstruct the path channel structure morphology, and generate a response channel structure index set;

[0087] Each enhancement path number is read sequentially from the set, and its corresponding image path point coordinate sequence is retrieved. Spatial position unfolding is performed on each path point, and its arrangement in the image's two-dimensional coordinate grid is analyzed point by point. The horizontal, vertical, and diagonal displacements between adjacent points in each path are used as the basis for path connectivity. The integrity of the path's internal structure is verified according to continuity rules: if the spatial displacement between all adjacent points is within 1 pixel or √2 pixels, the path connectivity structure is verified. If the coordinate difference between two consecutive points exceeds this range, the path is considered broken and marked as a structurally abnormal path, which is then removed or corrected before participating in subsequent channel structure reconstruction. For structurally complete paths, channel structure morphology construction is performed according to their point connection order, preserving the original path point order and incorporating it into the image response. The continuous channel region is generated synchronously in the response matrix. The area occupied by the path in each response channel is drawn as a linear or curved structure, while keeping the original path direction characteristics unchanged. If a path has local dense turns, spatial interpolation is used to supplement points between the turns to ensure the integrity of the path. The point supplementation rule is set according to the maximum spacing threshold. For example, points are supplemented for discontinuous areas exceeding 2 pixels. Then, the supplemented points are integrated into the path to complete the reconstruction of the connected structure. For example, the path numbered P10 has a cubic polyline segment at coordinates (320, 410) to (327, 417). By inserting points (324, 414) and (325, 415) to smooth the corners, the structure is restored. Finally, the structure of all paths is aggregated and organized to establish a response channel structure index set, recording the spatial position and boundary of each path in the image response channel.

[0088] S503: Call the response channel structure index set, call the image response matrix data for each channel path, read the channel response value according to the path sequence and perform mean statistics, combine the path spatial distribution attributes and numerical characteristics, calculate the representative index of the path structure, and generate the performance evaluation result of aluminum-vanadium alloy.

[0089] The process involves sequentially retrieving the index number of each path in the channel structure and its corresponding image coordinate point sequence. Numerical data for each path point in different response channels are retrieved from the image response matrix. A response value sequence is constructed according to the order of the path points. The mean of all response values ​​in the sequence is calculated by summing the response values ​​of all points in the sequence and dividing by the number of path points. This yields the average response intensity of the path in the corresponding channel. After the statistical analysis, the structural attributes of the path are comprehensively summarized, including the pixel area covered in the image space, path length, directional distribution, and boundary tightness. Furthermore, the structural attributes are combined with the mean response to construct a representative index of the path structure. This index reflects the path's response... The energy concentration and direction in the matrix are consistent. Standard judgment intervals are set. For example, paths with a mean response value above 120 and fewer than 3 directional changes are considered first-level representative paths, those between 90 and 120 are second-level representative paths, and those below 90 are third-level representative paths. If path number P07 has a mean value of 126 in the grayscale response channel and only changes direction 2 times, it is classified as a first-level path. After performing the same calculation on all paths, the representative indicators of the paths are summarized. Combined with the distribution of the stress coupling region where the path is located, the path structure characteristics in the whole image are summarized as a whole. Finally, a complete performance evaluation result of aluminum-vanadium alloy is generated. The result indicates the response capability and spatial distribution characteristics of different paths, which is used to support subsequent material property comparison and application scheme adjustment.

[0090] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A deep learning-based method for evaluating the performance of aluminum-vanadium alloys for aerospace applications, characterized in that, Includes the following steps: S1: Acquire microscopic images of aluminum-vanadium alloy, determine structural boundaries by the gray-scale difference between adjacent pixels, track gray-scale abrupt change points and number continuous closed boundaries, construct two-dimensional boundary graphics, and generate grain boundary maps. S2: Based on the grain boundary map, locate the images before and after the disturbance, compare the consistency of the gray-scale difference direction between the boundary point and the adjacent point, construct the disturbance continuity surface, organize the boundary coincidence interference map, and generate the disturbance difference structure. S3: Read the stress map corresponding to the location of the disturbance difference structure, filter the region where the disturbance and stress change are consistent, determine the change of the boundary point of the intersection block, analyze the spatial stability, and generate the stress coupling region. The specific steps for S3 are as follows: S301: Based on the location of the disturbance difference structure, read the stress change map recorded during the loading process, extract the gray value of the corresponding coordinate position of the structure point in the stress map, determine whether there is an overlap relationship between the location of the disturbance structure and the non-zero stress area in the stress map, mark all coordinate points of the overlapping area, and generate a coordinate set of the disturbance stress overlap area. S302: Call the coordinate set of the overlapping area of ​​disturbance stress, extract the disturbance difference direction data and stress map gray change direction data in the corresponding area, perform the same direction judgment, filter the coordinate set of points with matching directions, and re-sort according to the structural point index to generate a sequence of points with consistent disturbance stress. S303: Call the sequence of consistent disturbance stress points, divide it into continuously distributed sub-blocks, extract the number of boundary points of each sub-block, and calculate the ratio of the range of change of the number to the area of ​​the region. If the ratio is lower than the spatial stability threshold, mark the block, aggregate all marked blocks and reconstruct the graphic outline to generate the stress coupling region. S4: Locate the stress coupling region in the corresponding region of the feature map, extract the image response path and compare it with the neighboring response values ​​point by point, integrate paths with consistent differences, construct a channel extension map, and generate a response path configuration map. S5: Use the response path configuration graph to mark paths, record interference paths and impose restrictions, filter out unmarked paths to form an enhanced mark set, summarize all path structures, and generate aluminum-vanadium alloy performance evaluation results.

2. The deep learning-based performance evaluation method for aerospace aluminum-vanadium alloys according to claim 1, characterized in that, The grain boundary map includes boundary location distribution, grain morphology characteristics, and boundary numbering information. The perturbation difference structure includes grayscale variation regions, directional consistency blocks, and boundary perturbation patterns. The stress coupling region includes stress response coordinates, perturbation stress superposition points, and stability variation intervals. The response path configuration map includes a set of response trajectory points, a group of difference-consistent paths, and a channel extension structure. The aluminum-vanadium alloy performance evaluation results include perturbed path classification, path enhancement marker set, and channel performance mapping map.

3. The deep learning-based performance evaluation method for aerospace aluminum-vanadium alloys according to claim 1, characterized in that, The perturbation continuity surface refers to a continuous region in the image before and after the perturbation where the direction of the grayscale difference change between the boundary point and its neighboring points remains consistent.

4. The deep learning-based performance evaluation method for aerospace aluminum-vanadium alloys according to claim 1, characterized in that, The change in boundary points of the intersection block refers to the phenomenon where the position and shape of the boundary points change in the overlapping blocks of the region where the disturbance and stress change are consistent.

5. The deep learning-based performance evaluation method for aerospace aluminum-vanadium alloys according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire microscopic images of aluminum-vanadium alloy samples, calculate grayscale difference sequences by calling grayscale values ​​of adjacent pixels, determine whether the grayscale difference exceeds the threshold for structural abrupt change judgment, mark the coordinates of abrupt change pixels, and generate a set of grayscale abrupt change point coordinates. S102: Call the gray-scale abrupt change point coordinate set, retrieve the path continuity of the abrupt change point, determine whether the Euclidean distance between adjacent points is less than the closed structure determination threshold, and generate a closed boundary path number matrix. S103: Call the closed boundary path number matrix, aggregate the pixel coordinates of the same number and perform the boundary connection operation to establish the corresponding boundary polygon structure and generate the grain boundary map.

6. The deep learning-based performance evaluation method for aerospace aluminum-vanadium alloys according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the position marked by the grain boundary map, read the image before and after the disturbance, extract the gray values ​​of overlapping pixels in the same area, calculate the gray value difference between the boundary point and the adjacent pixel, encode the direction of the difference, and generate a boundary difference direction data group. S202: Call the boundary difference direction data group, perform angle difference calculation on the direction encoding between boundary points, determine the direction consistency mark, extract the connected set of pixels corresponding to the continuous and consistent region, and generate a set of perturbed continuous pixel patches. S203: Call the set of continuous perturbation pixel patches, aggregate the grayscale differences of pixels before and after the perturbation in the corresponding perturbation region, organize the interference pattern of the boundary region, and generate the perturbation difference structure.

7. The deep learning-based performance evaluation method for aerospace aluminum-vanadium alloys according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the corresponding region in the feature map of all coordinate points in the stress coupling region, extract all channel values ​​in the image response matrix, identify the path structure elements corresponding to the coordinate points, establish spatial distribution connection relationships, and generate an image response path coordinate set. S402: Call the image response path coordinate set, extract the values ​​of path points and adjacent regions in the image response matrix, perform point-by-point subtraction on the response values, determine whether the result is positive or negative, record the response direction, perform aggregation analysis on the direction sequence, and generate a set of path points with consistent response directions. S403: Call the set of path points with consistent response direction, aggregate continuous point columns according to spatial connectivity, expand all aggregated structures into continuous graphic regions in the image channel, record the path boundary contours and mark the structure index, and generate a response path configuration map.

8. The deep learning-based performance evaluation method for aerospace aluminum-vanadium alloys according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on all marked path points in the response path configuration diagram, extract the path index information of interference identifiers, add restriction identifiers to the path points, filter out the path set with restriction identifiers, and classify the remaining path numbers uniformly to establish a path enhancement identifier set. S502: Call the path enhancement mark set, aggregate the spatial positional relationships of all path points, verify the connection structure of each path, reconstruct the path channel structure morphology, and generate a response channel structure index set; S503: Call the response channel structure index set, call the image response matrix data for each channel path, read the channel response value according to the path sequence and perform mean statistics, combine the path spatial distribution attributes and numerical characteristics, calculate the representative index of the path structure, and generate the performance evaluation result of aluminum-vanadium alloy.

9. The deep learning-based performance evaluation method for aerospace aluminum-vanadium alloys according to claim 8, characterized in that, The restriction markers refer to the markers used during the path extraction process to identify path points that are affected by interference and are not suitable for subsequent analysis.

Images