Identification load fatigue fracture automatic interpretation method

Abstract

The invention discloses an identification load fatigue fracture automatic interpretation method. The method comprises the following steps: obtaining a fatigue fracture optical image; according to the convolutional neural network, carrying out segmented detection on a curve-shaped identification line in the fatigue fracture optical image to obtain an identification line discrete point set; according to the spatial density distribution of the discrete point set, the discrete point set is clustered into a plurality of point clusters, and noise points are removed; according to the spatial symmetry trend of each point cluster relative to the crack source, matching the point clusters belonging to the same crack front edge to form a complete crack front edge; the discrete points of the complete crack front edge are connected in sequence, a smooth curve is output as an interpretation result, and therefore automatic interpretation of the fatigue fracture identification line is completed. According to the method, manual marking is not needed in the whole process, the speed is high, the result is objective, the shape precision is high, and the method can be directly used for automatic interpretation of the fatigue fracture marking line in the reconstruction crack propagation process.

Description

Technical Field

[0001] This invention belongs to the field of automatic fatigue fracture identification technology, and in particular relates to an automatic identification method for fatigue fracture surfaces under identified loads. Background Technology

[0002] Fatigue fracture is a complex evolutionary process of a material / structure under alternating loads, from crack initiation and propagation to eventual fracture. Real-world structures are subjected to complex loads, and fatigue cracks in materials and structures under complex stress states typically exhibit a three-dimensional morphology. Crack propagation paths often undergo complex, non-planar evolution, including path inflections, bifurcation, and the initiation of secondary cracks. The crack plane may tilt, and the crack front morphology displays highly irregular geometric features. Accurately and completely reproducing the three-dimensional crack propagation path, crack front shape, and crack size is of great value for elucidating complex three-dimensional crack propagation behavior, revealing crack propagation laws, and establishing fatigue life assessment methods.

[0003] Given the complexity of structures and fatigue cracks, quantitative interpretation based on fracture surfaces has become the most effective method and has attracted widespread attention. Since alternating loads leave fatigue bands on the fracture surface, scanning electron microscopy (SEM) can provide a direct and effective analysis of the crack propagation process. However, due to the small spacing of the fatigue bands, SEM cannot provide a complete picture, and the interaction of high and low loads under random spectra makes accurate identification of the bands impossible.

[0004] To address this, a method for quantitatively interpreting fracture surfaces using identifying loads has been developed. By periodically inserting "identifying loads" into the fatigue load spectrum, identifying lines (optical microscope images) are left on the fatigue fracture surface for interpretation, thereby determining three-dimensional crack propagation information. This method has been widely applied in aircraft structures; however, the identifying load parameters rely on experience. To obtain the geometric characteristics of the identifying lines, manual fracture interpretation under an optical microscope or scanning electron microscope is required. Manual interpretation of fatigue fracture surfaces is time-consuming, labor-intensive, and inefficient.

[0005] To address the aforementioned problems in the existing technology, there is an urgent need to propose an automatic identification method for fatigue fracture surfaces under identified loads. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes an automatic identification method for identifying fatigue fracture surfaces under load, thereby improving the efficiency and accuracy of crack identification and resolving the problems existing in the prior art.

[0007] To achieve the above objectives, the present invention provides an automatic identification method for identifying fatigue fracture surfaces, comprising the following steps:

[0008] Acquire optical images of fatigue fracture surfaces;

[0009] The convolutional neural network is used to segment and detect the curved marker lines in the optical image of fatigue fracture, thereby obtaining a discrete set of marker line points.

[0010] Based on the spatial density distribution of the discrete point set, the discrete point set is clustered into several point clusters and noise points are removed;

[0011] Based on the spatial symmetry of each point cluster relative to the crack source, point clusters belonging to the same crack front are matched to form a complete crack front.

[0012] The discrete points at the complete crack front are connected sequentially, and a smooth curve is output as the interpretation result, thereby completing the automatic interpretation of the fatigue fracture mark line.

[0013] Optionally, the process of segmenting and detecting curved marker lines in the optical image of the fatigue fracture using a convolutional neural network to obtain a discrete set of marker lines includes:

[0014] A sliding window approach is used to sequentially extract local sub-images from the fracture image; each local sub-image is input into a trained convolutional neural network, which outputs a bounding box of the local segment of the marker line; the coordinates of the center points of all bounding boxes are extracted and summarized into a discrete point set of the marker line.

[0015] Optionally, the convolutional neural network adopts the YOLOv8 architecture, and the backbone network of the YOLOv8 architecture includes the CSPDarknet network, the PAN feature fusion network, and the anchorless detection head.

[0016] Optionally, the process of clustering the discrete point set into several point clusters and removing noise points based on the spatial density distribution of the discrete point set includes:

[0017] Take any point as the center and count the number of neighboring points within a preset radius; if the number of neighboring points reaches a preset minimum value, then the point is determined to be a core point; group all densely connected core points and their neighboring points into a point cluster and remove isolated noise points.

[0018] Optionally, the process of matching point clusters belonging to the same crack front to form a complete crack front, based on the spatial symmetry of each point cluster relative to the crack initiation, includes:

[0019] Using the crack initiation as the pole, the planar coordinates of each point cluster are converted to polar coordinates. Under polar coordinates, a quadratic function is used to jointly fit the point clusters of the same candidate combination. If the fitting residual is less than a preset threshold, the combination is determined to belong to the same crack initiation.

[0020] Optionally, the process of obtaining the fitting residual includes:

[0021] Sum the squares of the radial distance differences between points at the same polar angle and the fitted curve; divide the sum of squares by the number of points involved in the fitting, and then divide by the square of the maximum height of the fitted curve to obtain the normalized average residual.

[0022] Optionally, the process of connecting the discrete points of the complete crack front edge in sequence and outputting a smooth curve as the interpretation result includes:

[0023] The matched point clusters are arranged in ascending order of polar angle; the polar angle-polar radius relationship in polar coordinates is fitted using a quadratic function, and the fitting result is converted back to Cartesian coordinates to obtain a smooth curve of the crack front.

[0024] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0025] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0026] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method.

[0027] Compared with the prior art, the present invention has the following advantages and technical effects:

[0028] This invention utilizes a complete technology chain, including convolutional neural network segmented detection, density clustering for noise reduction, polar coordinate quadratic function matching and quadratic function fitting, and inverse coordinate transformation, to automatically extract curved marker lines from fracture images into discrete points, cluster them into point clusters, complete the matching with the leading edge according to the symmetry of the crack source, and then generate continuous smooth curves in polar angle order. This achieves automatic interpretation of fatigue fracture marker lines without manual marking, with high speed, objective results, high shape accuracy, and can be directly used to reconstruct the crack propagation process. Attached Figure Description

[0029] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0030] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0031] Figure 2 This is a schematic diagram illustrating the general morphology of the marking line on the fatigue fracture surface and the method for defining the identification target according to an embodiment of the present invention.

[0032] Figure 3 This is a schematic diagram illustrating the annotation of the identified target according to an embodiment of the present invention;

[0033] Figure 4 This is a schematic diagram of the DBSCAN algorithm flow according to an embodiment of the present invention;

[0034] Figure 5 This is a schematic diagram of the approximately symmetrical marking lines of a single crack source according to an embodiment of the present invention;

[0035] Figure 6 This is a schematic diagram of the crack leading edge fitting effect in an embodiment of the present invention, wherein (a) is a schematic diagram of the crack leading edge fitting effect when the scatter clusters come from the same crack leading edge; and (b) is a schematic diagram of the crack leading edge fitting effect when the scatter clusters come from different crack leading edges.

[0036] Figure 7 This is a schematic diagram of unit residual statistics based on log-normal distribution according to an embodiment of the present invention, wherein (a) is a schematic diagram of probability density function; and (b) is a schematic diagram of cumulative distribution function and empirical distribution data points.

[0037] Figure 8 This is a sample dimension diagram of an embodiment of the present invention;

[0038] Figure 9 Images of fatigue fracture surfaces of 33 specimens from an embodiment of the present invention;

[0039] Figure 10 The images shown are the identification results of the marking lines in the optical images of the fracture surface in the test set of this invention, where (a)-(c) are schematic diagrams of the identification results of the three test set images, respectively.

[0040] Figure 11 This is a schematic diagram of the evaluation training model process curve in an embodiment of the present invention, wherein (a) is a schematic diagram of the training set loss curve; and (b) is a schematic diagram of the validation set loss curve.

[0041] Figure 12 This is a schematic diagram of the DBSCAN clustering and denoising results of the scatter coordinates of the marker lines in an embodiment of the present invention, wherein (a)-(c) are schematic diagrams of the DBSCAN clustering and denoising results of the scatter coordinates of the marker lines in three test set images, respectively.

[0042] Figure 13 This is a schematic diagram of the scatter cluster matching result based on Res value in an embodiment of the present invention, wherein (a)-(c) are schematic diagrams of the scatter cluster matching result based on Res value for three test set images, respectively. Detailed Implementation

[0043] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0044] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0045] Example 1

[0046] The marked load method, as the most effective method for reproducing the three-dimensional fatigue crack propagation process, is widely used in the fatigue crack propagation theory and fatigue test research of reusable aircraft structures. However, it suffers from problems such as low efficiency in interpreting fatigue fracture marking lines and susceptibility to subjective interference. Therefore, this embodiment establishes an automatic fatigue fracture interpretation method based on image recognition and neural network algorithms to improve the interpretation efficiency and crack extraction accuracy of the marked load method, while reducing subjective interference in the interpretation results. First, based on the image convolutional neural network (CNN) algorithm and combined with the geometric characteristics of the marking lines, an automatic identification method for fracture marking lines is established, identifying continuous marking lines as discrete scatter points. Then, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering algorithm and the scatter point cluster matching algorithm are used to cluster and match the scatter points. Thus, an automatic fracture interpretation method based on marking line identification is established. Finally, the method is verified for marked load fractures of 7050-T7451 porous plate material under random spectrum.

[0047] This embodiment uses a convolutional neural network model to train the shape features of the marker lines and locates the scatter coordinates of the marker lines on the test set. Then, a density-based clustering algorithm is used to eliminate a small number of false positives and cluster the scatter points into different marker lines. Subsequently, based on the fitting residuals, the scatter point clusters are matched according to their respective crack leading edges, thereby completing the automatic identification of marker lines (i.e., crack leading edges). The method flowchart is as follows. As shown.

[0048] This embodiment provides a method for automatically identifying fatigue fracture surfaces under load, comprising the following steps:

[0049] Acquire optical images of fatigue fracture surfaces;

[0050] The convolutional neural network is used to segment and detect the curved marker lines in the optical image of fatigue fracture, thereby obtaining a discrete set of marker line points.

[0051] Based on the spatial density distribution of the discrete point set, the discrete point set is clustered into several point clusters and noise points are removed;

[0052] Based on the spatial symmetry of each point cluster relative to the crack source, point clusters belonging to the same crack front are matched to form a complete crack front.

[0053] The discrete points at the complete crack front are connected sequentially, and a smooth curve is output as the interpretation result, thereby completing the automatic interpretation of the fatigue fracture mark line.

[0054] As a specific implementation method, the following steps are included:

[0055] Define the target to be identified:

[0056] Unlike conventional object detection problems, marker lines appear as large-scale curves on images. Current object detection methods are almost entirely region detection algorithms. These algorithms extract local regions of an image into two-dimensional arrays (or colored three-dimensional arrays), analyze the array features for learning and training, and ultimately achieve automatic object detection. As shown, if the target is defined as a complete marker line, the algorithm's output is shown in the blue rectangle. This result is useless for analyzing the crack lead shape and constructing the crack propagation process. It's impossible to obtain the shape and accurate coordinates of the crack lead at that location using only the blue rectangle. However, if each marker line is segmented and identified, with only a small portion identified each time, and the results of these small rectangles are then pieced together, sufficiently accurate scattered coordinates of the marker lines can be obtained, such as... As shown in the red rectangle in the middle.

[0057] To improve recognition accuracy, when defining the recognition target, it is essential to ensure that the points within the rectangular frame lie on the marker line as much as possible. During post-processing of the recognition results, only the points within each recognition frame need to be extracted to obtain the scattered coordinates of the marker line.

[0058] Building the dataset:

[0059] In this embodiment, the dataset required for the image recognition method consists of optical photographs of fracture surfaces containing clearly marked lines. Following the marked load method, 33 fatigue tests were conducted on forged 7050-T7451 aluminum alloy under three random spectra, resulting in 33 optical photographs of the fracture surfaces. The dataset was divided into three parts: a training set, a validation set, and a test set. Three photographs were randomly selected as the test set, and 80% of the remaining 30 photographs were randomly selected as the training set, with 20% used as the validation set.

[0060] For each image in the training and validation sets, perform the following steps: The target labels are shown. This embodiment uses the "makesense" online labeling tool for labeling; similarly, the Python labelimg open-source toolkit or other labeling tools can also be used. The labeling tool returns a label type (in this embodiment, there is only one label type, i.e., label line) and the coordinates of the four vertices of the rectangle for each local label line image. The dataset images and their corresponding labels constitute the input to the convolutional neural network model.

[0061] Neural network models and training methods:

[0062] Yolov8 is an end-to-end supervised learning convolutional neural network model. Its neural model consists of three main parts: the backbone network is the CSPDarknet (cross-stage partial darknet) network, the PAN (PathAggregation Network) feature fusion network, and the anchorless detection head.

[0063] The backbone network is used for feature extraction. For an input image, I∈R H×W×3 Features of the target image are extracted through convolutional layers, BN (Batch Normalization) layers, and SiLU (sigmoid liner unit) activation layers.

[0064]

[0065] To improve the detection capability for targets of different sizes, the feature values ​​are then fed into a multi-scale feature fusion network (PAN). This network combines bottom-up and top-down information flow methods to fuse multi-scale features and construct a feature pyramid (P) based on the feature values ​​(Yolov8 uses a three-layer feature pyramid). If the feature value is a target with a defined label, the target's feature pyramid can be obtained.

[0066]

[0067] For an image to be detected, the above process is performed at each location in the image. Then, the feature values ​​are fed into the Anchor-Free detection head for target discrimination, and a prediction result is output as shown in the following formula.

[0068]

[0069] Where cls is the class probability, xy is the coordinates of the center point of the output object detection bounding box, wh is the width and height of the detection box, and objectness is the probability of the foreground object. Then, the prediction results are compared with the validation set, and the cumulative total loss function is calculated:

[0070]

[0071] The gradient of the loss function is then calculated using automatic differentiation, and the AdamW optimizer is used to update the model weights. This process is repeated to complete the model training.

[0072] Scatter clustering and matching methods for marker lines:

[0073] The DBSCAN algorithm is a density-based spatial clustering algorithm that divides regions with sufficient density into clusters and discovers clusters of arbitrary shapes in a noisy spatial database. It locates a cluster as the largest set of density-connected points.

[0074] The algorithm defines a radius r eps Circular neighborhood (planar problem) and a minimum number of samples n min This is a critical threshold used to describe the density of sample distribution within a neighborhood. If, centered at a certain point, the number of samples in the neighborhood is greater than n... min If a point is found to be a core object, then that point is considered a core object. Points in that neighborhood are said to be directly reachable from the core object density. If there exists a sample sequence p1, p2, p3, ..., p... K Furthermore, each subsequent sample in the sequence can be directly reached from the density of the preceding sample (i.e., p). k+1 Can be derived from p k If the density is directly reached, then it is called p. K Density reachability is determined by p1. If there exists a core object such that two samples are both density reachable from this core object, then these two samples are said to be density-connected. The set of samples with the highest density reachability derived from the density reachability relation is clustered into a cluster. That is, select a core object that has not yet been classified as a seed, and then find the set of all samples density-reachable from this core object; these can be clustered into a cluster, such as... Figure 4 As shown.

[0075] Matching algorithm based on the physical characteristics of the marking lines:

[0076] The DBSCAN algorithm is a clustering method based on scatter density, which does not consider the complete curve trend of the crack lead. For a single crack source (this embodiment only solves the single crack source problem), the marker line has good approximate symmetry with respect to the crack source, such as... As shown. By utilizing this symmetrical trend of the marker lines, matching relationships between clusters of scattered points can be established.

[0077] Using the crack initiation point as the pole (it should be noted that in this embodiment, the crack initiation location in the fracture image still needs to be manually specified, and an automated crack initiation identification method has not yet been established), the (x,y) coordinates of the scattered points on the marker line are transformed into polar coordinates (θ,r). For a complete cluster of scattered points on a marker line, plotting (θ,r) on a Cartesian coordinate system still maintains good symmetry. In this embodiment, a quadratic function is used to fit the (θ,r) relationship, and then transformed back into (x,y) coordinates, which can fit the crack leading edge well. When the scattered point cluster comes from the same crack leading edge, the residual between the fitted curve and the scattered point cluster is small. When the scattered point cluster comes from different crack leading edges, the curve fits the scattered points very poorly, with a large residual, such as... As shown.

[0078] Therefore, different scatter clusters can be fitted together. If the residual is less than a certain critical value, these scatter clusters are considered to originate from the same crack source. Otherwise, if the residual is too large, they are determined not to belong to the same crack front. In this embodiment, the residual is defined as the sum of the squares of the difference between the polar radius of the scatter points and the curve at the same polar angle θ in polar coordinates. To avoid the influence of the size of the marker line and the number of scatter points on the residual evaluation, the sum of squares of the residuals is first divided by the number of coordinate points, and then divided by the square of the height of the fitted curve for normalization to obtain the normalized average residual, which is used to evaluate formula (5).

[0079]

[0080] In order to obtain the critical residual Res0 for determining the matching relationship, this embodiment extracts the scatter coordinate set of 62 marker lines, and fits the scatter points from the same crack front edge according to the above method, calculates the residuals to obtain statistical regularity, and thus determines the critical residual Res0.

[0081] like Figure 7 As shown, this assumes the average residual follows a log-normal distribution (base 10), i.e., the formula... The probability density function is shown. According to the formula... The estimated distribution parameters are μ = -1.31772 and σ = 0.0408. In this embodiment, the 0.95 quantile is taken as the critical value, resulting in Res0 = 0.0562. That is, if Res is greater than 0.0562, the scatter plot clusters are considered to originate from different crack leads. In the sample, the Res value for the scatter cluster originating from the same crack front is 0.0325; while The Res value of the scatter cluster originating from two adjacent crack leads is 0.0761. The latter fails the matching test and can be identified as originating from different crack leads. The greater the distance between the crack leads to which the scatter cluster belongs, the larger the Res value; when the two are separated by a marker line, the Res value reaches 0.0871. The scatter cluster matching algorithm based on Res is quite effective.

[0082]

[0083]

[0084] Example:

[0085] This embodiment uses 7050-T7451 perforated plate material and conducts fatigue tests according to the indicated load method described by Wang et al. The material and specimen structure are exactly the same as Wang's. The material composition and specimen dimensions are as follows: and As shown.

[0086] surface

[0087] Si Fe Cu Mn Mg Cr Zn Ti 0.12 0.15 2.0-2.6 0.10 1.9-2.6 0.04 5.7-6.7 0.06

[0088] In this embodiment, 34 fatigue tests were conducted using the aforementioned materials and specimens under three random spectra and three stress levels. The fatigue tests under the random spectra (including the labeled load) were conducted on an Instron 8801 microscope, and fracture surface images were captured using a ZEISS optical microscope at 50x magnification. In this embodiment, the captured fracture surface optical photographs were 636×481 pixels. Figure 9 As shown.

[0089] Define the target to be identified and build the dataset:

[0090] The dataset is divided into three parts: training set, validation set, and test set. Three photos are randomly selected as the test set, and of the remaining 30 photos, 80% are randomly selected as the training set and 20% as the validation set. The 30 photos from the training and validation sets were labeled and compiled into a txt file containing the image name, label name, and coordinates of the corner points of the target rectangle.

[0091] Neural network model training:

[0092] The image set was fed into a YOLOv8 neural network model for training. After 100 training steps, the optimal model obtained during training was used to perform line recognition tests on the test set images. The final line recognition results are as follows. As shown in the figure, the direct output of the neural network is a blue rectangle. The coordinates of each point within the rectangle are extracted as the final recognition result of the marker line (green scatter dots). The results show that the trained neural network model has excellent recognition ability.

[0093] During training, the training set and validation set were recorded in the formula. The changes in the three types of loss values ​​as the training steps increase are as follows: As shown, the loss values ​​of both the training and validation sets decrease and tend to flatten as the number of learning steps increases. While the validation set loss exhibits some oscillations and fluctuations, the trained model still does not suffer from overfitting or underfitting, indicating that the model is relatively reliable.

[0094] Scatter clustering and matching:

[0095] In this embodiment, the captured optical photograph of the fracture surface is 636×481 pixels, and r is taken as... eps =20 pixels, n min =4, and clusters with fewer than 5 samples were removed. The test set image recognition results were clustered and optimized, and the results are as follows: As shown in the results, the DBSCAN algorithm effectively removes false positives in image recognition, optimizes the recognition results, and completes preliminary clustering of scatter point coordinates.

[0096] right The scatter clusters shown were matched and classified, and the result is shown in the "Error! Reference source not found" example. Therefore, the scatter clusters in the left image were correctly matched to their respective crack leads. This completes the automatic identification of the marker lines from the image of the fatigue fracture under identified load, as shown in the example. Figure 13 As shown.

[0097] Example 2

[0098] This embodiment also discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in Embodiment 1.

[0099] Example 3

[0100] This embodiment also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0101] Example 4

[0102] This embodiment also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0103] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An automatic reading method for identifying a load fatigue fracture, characterized by, The method comprises the following steps: obtaining an optical image of a fatigue fracture; segmenting and detecting a curve-shaped mark line in the optical image of the fatigue fracture according to a convolutional neural network to obtain a discrete point set of the mark line; clustering the discrete point set into a plurality of point clusters and removing noise points according to a spatial density distribution of the discrete point set; matching point clusters belonging to the same crack front to form a complete crack front according to a spatial symmetry trend of each point cluster relative to a crack source; sequentially connecting discrete points of the complete crack front, and outputting a smooth curve as an interpretation result, thereby completing automatic interpretation of the mark line of the fatigue fracture.

2. The method of claim 1, wherein the process of segmenting and detecting a curve-shaped mark line in the optical image of the fatigue fracture according to a convolutional neural network to obtain a discrete point set of the mark line comprises: adopting a sliding window method to sequentially intercept local subgraphs on the fracture image; inputting each local subgraph into a trained convolutional neural network to output a rectangular frame of a local segment of the mark line; and extracting the center point coordinates of all rectangular frames to be summarized as a discrete point set of the mark line.

3. The method of claim 1, wherein the convolutional neural network adopts a yolov8 architecture, and a backbone network of the yolov8 architecture comprises a CSPDarknet network, a PAN feature fusion network, and an anchor-free detection head.

4. The method of claim 1, wherein the process of clustering the discrete point set into a plurality of point clusters and removing noise points according to a spatial density distribution of the discrete point set comprises: counting the number of adjacent points within a preset radius range with any point as the center; if the number of adjacent points reaches a preset minimum value, the point is determined to be a core point; and all core points and adjacent points with the same density are grouped into a point cluster, and isolated noise points are removed.

5. The method of claim 1, wherein the process of matching point clusters belonging to the same crack front to form a complete crack front according to a spatial symmetry trend of each point cluster relative to a crack source comprises: converting the plane coordinates of each point cluster into polar coordinates with the crack source as the pole; in polar coordinates, a quadratic function is used to jointly fit the point clusters in the same candidate combination; and if the fitting residual is less than a preset threshold, it is determined that the combination belongs to the same crack front.

6. The method of claim 5, wherein the process of obtaining the fitting residual comprises: squaring the radial distance difference between the points under the same polar angle and the fitting curve; dividing the sum of squares by the number of points participating in fitting, and then dividing by the square of the maximum height of the fitting curve to obtain the normalized average residual.

7. The method of claim 1, wherein the process of sequentially connecting the discrete points of the complete crack front and outputting a smooth curve as an interpretation result comprises: arranging the matched point clusters in ascending order of polar angle; fitting the polar angle-polar radius relationship in polar coordinates using a quadratic function, and converting the fitting result back to plane rectangular coordinates to obtain a smooth curve of the crack front. The processor executes the computer program to implement the steps of the method of any one of claims 1-7. ​ ​ ​ ​ ​ 8. A computer apparatus comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program, when executed by the processor, causes the processor to perform the method of any one of claims 1 to 7. ​ 9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, which when executed by the processor, implements the steps of the method of any one of claims 1-7.

10. A computer program product comprising a computer program, characterized in that, The computer program, which when executed by the processor, implements the steps of the method of any one of claims 1-7.

Images