Method and device for detecting cracks of aluminum alloy component

Abstract

The invention relates to the technical field of crack detection, and discloses an aluminum alloy component crack detection method and device.The method comprises the steps that an aluminum alloy component is subjected to CT scanning and preprocessing, and a preprocessed CT image is obtained; performing crack judgment through an integrated classification model of a bilinear convolutional neural network and a support vector machine classifier to obtain an image which is judged to have cracks; performing crack width transformation to generate a crack width graph; carrying out region growing algorithm search to obtain a complete crack region and a corresponding crack quantitative parameter set; performing vibration analysis on the aluminum alloy component to obtain vibration characteristic data; a stress intensity factor is calculated by using a continuous damage mechanical model, and fatigue crack propagation analysis is performed to obtain residual life prediction data, so that the accuracy of crack identification is improved, accurate extraction of crack features is realized, and reliable residual life prediction is performed on the basis.

Description

Technical Field

[0001] The present application relates to the technical field of crack detection, and in particular to a method and device for detecting cracks in aluminum alloy components. Background Art

[0002] Aluminum alloys are widely used in high-end manufacturing fields such as aerospace and automobile manufacturing due to their excellent strength-to-weight ratio and corrosion resistance. However, during service, aluminum alloy components are prone to fatigue cracks, threatening the safety and reliability of the structure. Traditional crack detection methods such as visual inspection and ultrasonic testing often have difficulty accurately identifying tiny cracks and cannot effectively predict the crack expansion trend.

[0003] With the development of computer vision and artificial intelligence technology, automated crack detection methods based on CT images have gradually become a research hotspot. However, existing methods still have problems such as low recognition accuracy and high missed detection rate when processing high-noise, low-contrast CT images. In addition, for cracks with complex shapes, traditional image processing algorithms are difficult to accurately extract crack features, which affects subsequent quantitative analysis and life prediction. Summary of the invention

[0004] The present application provides a method and device for detecting cracks in aluminum alloy components, thereby improving the accuracy of crack identification, achieving precise extraction of crack features, and on this basis, making reliable remaining life predictions.

[0005] The first aspect of the present application provides a method for detecting cracks in an aluminum alloy component, the method comprising: Performing CT scanning and preprocessing on the aluminum alloy component to obtain a preprocessed CT image; Inputting the preprocessed CT image into an integrated classification model of a bilinear convolutional neural network and a support vector machine classifier for crack discrimination, thereby obtaining an image judged to have cracks; Perform crack width transformation on the image determined to have cracks, calculate the distance from the pixel to the edge, extract the center line, and generate a crack width map; Based on the crack width map, a regional growth algorithm search is performed to connect the disconnected crack segments and fill the small holes inside the cracks to obtain a complete crack region and a corresponding crack quantitative parameter set; Performing vibration analysis on the aluminum alloy component according to the complete crack region, calculating the natural frequency and mode, and obtaining vibration characteristic data; According to the vibration characteristic data and the crack quantitative parameter set, a continuous damage mechanics model is used to calculate the stress intensity factor, and a fatigue crack propagation analysis is performed to obtain the remaining life prediction data of the aluminum alloy component.

[0006] A second aspect of the present application provides a device for detecting cracks in an aluminum alloy component, the device comprising: A scanning module, used for performing CT scanning and preprocessing on the aluminum alloy component to obtain a preprocessed CT image; A discrimination module, used for inputting the preprocessed CT image into an integrated classification model of a bilinear convolutional neural network and a support vector machine classifier for crack discrimination, and obtaining an image judged to have cracks; A transformation module, used to perform crack width transformation on the image determined to have cracks, calculate the distance from the pixel to the edge, extract the center line, and generate a crack width map; A search module is used to perform a regional growth algorithm search based on the crack width map, connect the disconnected crack segments, fill the small holes inside the cracks, and obtain a complete crack region and a corresponding crack quantitative parameter set; An analysis module, used to perform vibration analysis on the aluminum alloy component according to the complete crack area, calculate the natural frequency and mode, and obtain vibration characteristic data; A calculation module is used to calculate the stress intensity factor according to the vibration characteristic data and the crack quantitative parameter set using a continuous damage mechanics model, and perform fatigue crack propagation analysis to obtain the remaining life prediction data of the aluminum alloy component.

[0007] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned method for detecting cracks in aluminum alloy components.

[0008] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned method for detecting cracks in aluminum alloy components.

[0009] Compared with the prior art, the present application has the following beneficial effects: by combining the integrated classification model of bilinear convolutional neural network and support vector machine classifier, it is possible to effectively process high-noise, low-contrast CT images, improve the accuracy of crack identification, and reduce the missed detection rate. By adopting crack width transformation and regional growth algorithm, the geometric features of complex-shaped cracks, including length, width, direction and curvature, can be accurately extracted, providing reliable quantitative parameters for subsequent analysis. By connecting the disconnected crack segments and filling the internal small holes, a more complete crack area can be obtained, improving the accuracy of subsequent analysis. Based on the complete crack area, finite element modeling and modal analysis can be performed to obtain the vibration characteristic data of aluminum alloy components, providing a basis for evaluating the impact of cracks on the overall performance of the structure. Combining the continuous damage mechanics model and Monte Carlo simulation method, the randomness of material parameters and loads can be considered, and the probability distribution of the remaining life can be given, thereby improving the accuracy of crack identification, realizing the accurate extraction of crack features, and on this basis, reliable remaining life prediction can be performed. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0011] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0012] Figure 1 It is a schematic flow chart of a method for detecting cracks in an aluminum alloy component provided by an embodiment of the present invention; Figure 2 It is a schematic block diagram of the structure of a device for detecting cracks in an aluminum alloy component provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0015] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0016] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 In the embodiment of the present application, an embodiment of the method for detecting cracks in an aluminum alloy component includes: Step 100, performing CT scanning and preprocessing on the aluminum alloy component to obtain a preprocessed CT image; It is understandable that the execution subject of the present application may be a detection device for cracks in aluminum alloy components, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0017] Specifically, a high-resolution CT scan is performed on the aluminum alloy component to obtain the original CT image sequence, which contains detailed information on the internal structure of the aluminum alloy component. The original CT image sequence is converted into a 16-bit grayscale PNG format image to retain more grayscale information and enhance the detail expression of the image. The 16-bit grayscale PNG format image is subjected to adaptive median filtering. By calculating the median of the pixel values ​​in the local area of ​​the image and replacing the value of each pixel with the median of the area, the image is denoised to obtain a denoised CT image. The global grayscale histogram is calculated based on the denoised CT image to determine the appropriate grayscale value stretching range. The contrast-enhanced CT image is obtained by linearly mapping the grayscale values. The contrast-enhanced CT image is subjected to multi-scale anisotropic diffusion filtering. This filtering technology highlights the edge features in the image by smoothing and enhancing the image at different scales to obtain an edge-enhanced CT image. Based on the edge-enhanced CT image, the global optimal threshold is calculated, and a preliminary binary image is generated in combination with the local adaptive threshold processing technology. On this basis, the connected domain analysis is performed on the preliminary binary image to identify and segment the crack area to obtain the preprocessed CT image. Connected domain analysis can effectively identify coherent regions in an image and help separate cracks from other irrelevant regions.

[0018] Step 200, inputting the preprocessed CT image into an integrated classification model of a bilinear convolutional neural network and a support vector machine classifier for crack discrimination, and obtaining an image that is judged to have cracks; Specifically, the preprocessed CT images are processed in blocks, and each image is divided into multiple overlapping sub-image blocks to obtain a set of CT image blocks. Through block processing, the analysis capability of local details is increased, and crack detection is made more refined, which helps to capture small or blurred crack features. The CT image block set is input into two different convolutional neural networks, ResNet50 and DenseNet121, for feature extraction. In the ResNet50 network, the image blocks are processed by five residual blocks, each of which contains three convolutional layers and short-circuit connections, so as to obtain the global features of the image through a deeper network structure and avoid the gradient disappearance problem. After this process, a 4096-dimensional global feature vector is obtained to fully characterize the macro information in the image block. At the same time, the DenseNet121 network extracts features from the same set of CT image blocks through four dense blocks. The dense blocks realize the transmission and sharing of features through the dense connection of the convolutional layer, which can extract more detailed local feature information and obtain a 1024-dimensional local feature vector. The advantage of DenseNet121 is that its network structure can effectively utilize low-level feature information, making the extracted features richer and more diverse. The global feature vector from ResNet50 and the local feature vector from DenseNet121 are bilinearly pooled. The two types of features are fused through outer product operations and feature reshaping to generate a 16384-dimensional fused feature vector. Through bilinear pooling, the high-order interaction information of the two types of features is retained to enhance the expressiveness of the features. The fused feature vector is reduced in dimension to reduce the computational complexity and storage requirements, and a 2048-dimensional reduced feature vector is obtained. The reduced feature vector is input into the support vector machine classifier, and the radial basis function kernel is used to perform nonlinear mapping of the features to find the optimal separation hyperplane and achieve preliminary classification of cracks and non-cracks. With its ability to construct the optimal hyperplane in high-dimensional space, the support vector machine can effectively distinguish the subtle differences between complex features. Based on the preliminary classification results, the classification results of adjacent image blocks are contextually optimized, and the classification results of each image block are adjusted and optimized by minimizing the energy function. Context optimization uses the spatial relationship and category consistency between image blocks to effectively reduce classification errors and obtain optimized classification results. The optimized classification results are analyzed for confidence, and the classification output of the support vector machine is converted into a probability distribution to estimate the probability of the existence of cracks in each image block. By analyzing these probabilities, especially in overlapping areas, decision fusion is performed using a weighted average method, and an appropriate threshold is set to finally determine the presence of cracks in the image.

[0019] The CT image block set is normalized, the pixel values ​​are scaled to the range of [0, 1], and the pre-calculated mean is subtracted and divided by the standard deviation to ensure the consistency of the input data and the stability of the model, and the standardized CT image block is obtained. The standardized CT image block is input into the initial convolution layer of the ResNet50 network, in which a 7×7 convolution kernel and a convolution operation with a stride of 2 are used. Subsequently, the initial convolution features of the image are extracted by the maximum pooling operation to obtain the first convolution feature map. The first convolution feature map is input into the first residual block group, which contains three residual units. Each residual unit consists of a 1×1 convolution layer, a 3×3 convolution layer, and a 1×1 convolution layer. These convolution layers are connected by short-circuit connections to achieve identity mapping. The features are fused by element-by-element addition, and important feature information is retained and transmitted to obtain the second convolution feature map. The second convolution feature map is downsampled, and a convolution with a stride of 2 is used to reduce the resolution of the feature map while maintaining the integrity of important features. The downsampled feature map is input into the second residual block group, which contains four residual units. Each unit continues to use the 1×1, 3×3, and 1×1 convolution sequence, and uses projection short circuit to match the input and output dimensions to obtain the third convolution feature map. The third convolution feature map is input into the third residual block group, which contains six residual units. In order to expand the receptive field, the convolution layer in the third residual block group uses dilated convolution, and the dilation rate increases from 1 to 2, so that the network can capture feature information in a larger range while maintaining the details of the feature map, and obtain the fourth convolution feature map. The fourth convolution feature map is input into the fourth residual block group, which contains three residual units. Each unit introduces the SE module, obtains global information through global average pooling, and then recalibrates the weight of each channel through two fully connected layers to enhance the model's attention to important feature channels, and generate the fifth convolution feature map. The fifth convolutional feature map is input to the last residual block group, which also contains three residual units. In order to increase the diversity and interactivity of features while keeping the number of parameters unchanged, each residual unit uses group convolution and channel rearrangement operations to ensure that the complex relationship between features is fully captured and expressed, and the sixth convolutional feature map is generated. The sixth convolutional feature map is subjected to global average pooling, and the two-dimensional feature map is converted into a one-dimensional vector to extract global features. The feature dimension is adjusted through a fully connected layer, and the dimension of the feature vector is reduced to 4096 dimensions, resulting in a 4096-dimensional global feature vector.

[0020] Data enhancement is performed on the CT image block set. Through random rotation, flipping, brightness adjustment and other operations, the diversity of the data set is expanded to prevent overfitting and enhance the generalization ability of the model, and the enhanced CT image block is obtained. The enhanced CT image block is input into the initial convolution layer of DenseNet121, which uses a 7×7 convolution kernel for convolution operation and the step size is set to 2. This configuration can quickly extract the preliminary features of the image. Then, the size of the feature map is reduced by a 3×3 maximum pooling operation to obtain the initial feature map. The initial feature map is input into the first dense block of DenseNet121, which consists of 6 dense layers. Each dense layer includes a 1×1 convolution layer and a 3×3 convolution layer. All convolution layers use the connection function to splice the output of each layer with the feature maps of all previous layers to obtain the first residual feature map. The first residual feature map is compressed by 1×1 convolution and downsampled by 2×2 average pooling operation to reduce the amount of calculation and the size of the feature map while retaining key feature information. The compressed and downsampled feature map is input into the second dense block, which contains 12 dense layers. Dilated convolution is used instead of standard 3×3 convolution, and the dilation rate varies alternately in each layer, ranging from 1 to 2. The use of dilated convolution can expand the receptive field while maintaining the size of the feature map, extract contextual information in a larger range, and obtain the second residual feature map. After channel compression and downsampling, the second residual feature map enters the third dense block, which contains 24 dense layers, and an attention mechanism is introduced after each dense layer, adding a channel attention module. This module enhances the focus on key features by dynamically adjusting the importance of feature channels, so that the model can focus more accurately on crack-related image areas, and generate the third residual feature map. The third residual feature map is channel compressed and downsampled, and input into the fourth dense block, which consists of 16 dense layers. In order to enhance the interactivity of features while maintaining the efficiency of model parameters, each dense layer uses group convolution and channel shuffling operations. These techniques can more effectively break the independence between channels, increase the fusion and expression capabilities of features, and obtain the fourth residual feature map. The fourth residual feature map is input into the transition layer, which contains a 1×1 convolution layer to reduce the number of channels to simplify the structure of the feature map. The two-dimensional feature map is then converted into a one-dimensional feature vector through a global average pooling operation to form a global image representation. The dimension of the one-dimensional feature vector is adjusted to 1024 through a fully connected layer, and Dropout regularization is applied to reduce overfitting, resulting in a 1024-dimensional local feature vector.

[0021] Step 300: Perform crack width transformation on the image determined to have cracks, calculate the distance from the pixel to the edge, extract the center line, and generate a crack width map; It should be noted that edge detection is performed on the image determined to have cracks. Through the edge detection algorithm, a binary image of the crack edge is obtained. Based on the binary image of the crack edge, the distance from each pixel to the nearest edge pixel is calculated to obtain a distance transformation image. Each pixel value in the distance transformation image represents the distance between the pixel and the nearest crack edge. According to the generated distance transformation image, the initial center line of the crack is extracted. The center line represents the geometric center of the crack in the image. The center line is identified by the peak pixels in the distance transformation image. These peak pixels are usually located on the central axis of the crack area. The center line is extracted to obtain the initial crack center line image. The initial crack center line image is optimized to eliminate noise and discontinuous points in the initial center line image, so that the crack center line is smoother and continuous. By removing noise and repairing breakpoints, the optimized crack center line is obtained. Based on the optimized crack center line, the vertical distance from each point on the center line to the crack edge is calculated. The vertical distance represents the local width value of the crack at that position. By calculating the entire crack center line, the local width distribution of the crack is obtained. In order to improve the stability and accuracy of the local crack width values, they are smoothed. Smoothing helps eliminate width fluctuations caused by local irregularities or measurement errors, making the width curve more coherent and smooth. The smoothed crack width values ​​are mapped back to the size of the original image to generate the final crack width map.

[0022] Step 400: Perform a regional growth algorithm search based on the crack width map, connect the disconnected crack segments, fill the small holes inside the cracks, and obtain a complete crack region and a corresponding crack quantitative parameter set; Specifically, the crack width map is threshold segmented. By setting a preset threshold, pixels with a width value greater than the threshold are marked as seed points, an initial seed point set is obtained, and the area where the crack may exist is identified. The region growing algorithm is executed based on the initial seed point set. The region growing algorithm gradually incorporates pixels adjacent to the seed point and with similar width values ​​into the growth area by expanding the seed point to form a preliminary crack area. The skeleton of the preliminary crack area is extracted to obtain a crack skeleton map, and the endpoints and branch points on the skeleton are marked. The skeleton extraction simplifies the crack area into one or more center lines, which is convenient for subsequent analysis and connection processing. On this basis, the Euclidean distance and direction angle between adjacent endpoints are calculated, and the endpoint pairs with a distance less than the threshold and a close direction angle are connected to repair the broken crack segments, make the crack skeleton more continuous, and obtain the connected crack skeleton. According to the connected crack skeleton, the crack area is reconstructed. Through the expansion operation, the connected skeleton is expanded back to the original crack width, the complete shape of the crack is restored, and the reconstructed crack area is obtained. In order to eliminate the small holes and gaps that may exist inside the crack, the reconstructed crack area is subjected to morphological closing operation. The morphological closing operation can effectively fill the tiny gaps and breakpoints inside the crack to obtain a complete crack area. The geometric features of the crack are extracted based on the complete crack area, including parameters such as the length, average width, maximum width, direction and curvature of the crack, and the geometric parameter set of the crack is obtained. The crack depth and volume are calculated by combining the crack geometric parameter set and the grayscale information of the original CT image. By analyzing the relationship between the grayscale change in the CT image and the actual depth and volume of the crack, the quantitative parameter set of the crack is finally obtained.

[0023] Step 500: Perform vibration analysis on the aluminum alloy component according to the complete crack area, calculate the natural frequency and mode, and obtain vibration characteristic data; Specifically, a three-dimensional finite element model is constructed based on the geometric information of the complete crack area and the aluminum alloy component. In this model, the mesh is refined for the crack area to ensure that the geometric details of the crack are accurately captured and the calculation accuracy is improved to obtain a discretized structural model. According to the discretized structural model, the mass matrix and stiffness matrix are established. The mass matrix reflects the mass distribution of the structure, while the stiffness matrix describes the stiffness characteristics of the structure. By solving the eigenvalue problems of these matrices, the natural frequencies and their corresponding eigenvectors are obtained. The natural frequencies represent the natural vibration frequencies of the structure without external forces, while the eigenvectors correspond to the vibration modes at these frequencies. The eigenvectors are normalized, and the node displacements are converted into continuous modal shapes to obtain the vibration modes of the aluminum alloy component, reflecting the vibration modes of the aluminum alloy component at different natural frequencies. Based on the natural frequencies and vibration modes, the modal participation factors and effective masses are calculated. These parameters are used to evaluate the contribution of each mode to the overall structural response. By analyzing these contributions, the importance ranking of each mode is obtained. According to the order of modal importance, the first N main modes are selected, and the proportional damping matrix is ​​constructed using the Rayleigh damping model. This damping model can simplify the calculation of damping force and facilitate harmonic response analysis. Through harmonic response analysis, the frequency response function at different excitation frequencies is calculated to obtain the harmonic response characteristics of aluminum alloy components. The harmonic response characteristics provide dynamic response information of the structure under different frequency excitations. Combining the natural frequency, vibration mode and harmonic response characteristics, characteristic parameters such as frequency offset, modal curvature and transfer function are extracted. These parameters reflect the specific impact of cracks on the vibration characteristics of the structure, which can help evaluate the health status of aluminum alloy components and obtain comprehensive vibration characteristic data.

[0024] Step 600: Calculate the stress intensity factor using a continuous damage mechanics model based on the vibration characteristic data and the crack quantitative parameter set, and perform fatigue crack growth analysis to obtain the remaining life prediction data of the aluminum alloy component.

[0025] Specifically, a continuous damage mechanics model of aluminum alloy components is constructed based on vibration characteristic data and crack quantitative parameter sets. The construction of this model combines the geometric characteristics, material properties and vibration response of the crack area to accurately describe the stress field around the crack. In order to solve the stress intensity factor at the crack tip, the J integral method is used to analyze the stress field at the crack tip. As a path-independent integral method, the J integral can effectively calculate the stress intensity factor at the crack tip. Its accuracy is verified by the strain energy release rate to ensure that the calculation of the stress intensity factor reflects the actual stress field distribution at the crack tip. According to the stress field distribution at the crack tip, the fatigue crack growth rate equation is established using the Paris law. The Paris law is an empirical formula for fatigue crack growth analysis, and its main parameters include the stress intensity factor range and crack growth rate. The Walker correction is used to consider the influence of stress ratio on crack growth rate, and the corrected crack growth rate model is obtained. Based on the corrected crack growth rate model, the numerical integration method with adaptive step size is used to simulate the crack growth path. The adaptive step size method can dynamically adjust the step size during crack growth to improve the calculation efficiency and accuracy. Through the simulation process, the evolution curve of crack length with the number of cycles is obtained, which reflects the law of gradual crack expansion under fatigue load. The crack length evolution curve is fitted piecewise linearly, and the crack extension parameters of each stage are extracted from it to describe the behavior characteristics of the crack at different expansion stages. Combined with the Forman equation, the influence of the critical stress intensity factor is considered to establish a crack extension life prediction model. The Forman equation can effectively link the crack extension rate with the upper limit of the stress intensity factor, and more accurately predict the crack extension rate when it is close to failure. Based on the crack extension life prediction model, combined with the Monte Carlo simulation method, considering the randomness of material parameters and loads, multiple groups of life prediction results are generated. The Monte Carlo method can simulate various uncertain factors under actual use conditions through random sampling technology to obtain the probability distribution of the remaining life. The probability distribution of the remaining life is analyzed for reliability, the remaining life under a given failure probability is calculated, and the key factors affecting the life prediction are determined through sensitivity analysis. Sensitivity analysis helps to identify which parameters have the greatest impact on the remaining life, so that these factors can be monitored in actual operation, and finally the remaining life prediction data of aluminum alloy components can be obtained.

[0026] In the embodiment of the present application, by combining the integrated classification model of bilinear convolutional neural network and support vector machine classifier, it is possible to effectively process high-noise, low-contrast CT images, improve the accuracy of crack identification, and reduce the missed detection rate. By using crack width transformation and regional growth algorithm, the geometric features of complex shape cracks, including length, width, direction and curvature, etc., can be accurately extracted, providing reliable quantitative parameters for subsequent analysis. By connecting the disconnected crack segments and filling the internal small holes, a more complete crack area can be obtained, improving the accuracy of subsequent analysis. Based on the complete crack area, finite element modeling and modal analysis can be performed, and the vibration characteristic data of aluminum alloy components can be obtained, providing a basis for evaluating the impact of cracks on the overall performance of the structure. Combined with the continuous damage mechanics model and Monte Carlo simulation method, the randomness of material parameters and loads can be considered, and the probability distribution of the remaining life can be given, thereby improving the accuracy of crack identification, realizing the accurate extraction of crack features, and on this basis, reliable remaining life prediction can be performed.

[0027] In a specific embodiment, the process of executing step 100 may specifically include the following steps: Perform high-resolution CT scanning on aluminum alloy components to obtain original CT image sequences, and convert the original CT image sequences into 16-bit grayscale PNG format images; Adaptive median filtering is performed on the 16-bit grayscale PNG format image. The median of the pixel values ​​in the local area is calculated and the pixel value is replaced by the median to obtain the denoised CT image. The global grayscale histogram is calculated based on the denoised CT image, the stretching range is determined, and the grayscale value is linearly mapped to obtain a contrast-enhanced CT image; Perform multi-scale anisotropic diffusion filtering on the contrast-enhanced CT image to obtain an edge-enhanced CT image; Based on the edge-enhanced CT image, the global optimal threshold is calculated, and combined with the local adaptive threshold to obtain a preliminary binary image. The connected domain analysis is performed on the preliminary binary image to obtain the preprocessed CT image.

[0028] Specifically, the internal structural information of the component is obtained through high-resolution CT scanning to obtain the original CT image sequence. The original CT image sequence is converted into a 16-bit grayscale PNG format image, which retains rich grayscale information, has a wider dynamic range, and better represents the subtle differences in aluminum alloy components. Adaptive median filtering is performed on the 16-bit grayscale PNG format image to eliminate the noise that may be introduced during the CT scanning process. Adaptive median filtering calculates the median of the pixel values ​​in the local area of ​​the image and replaces the value of each pixel with the median of the area, effectively removing noise while retaining the edges and details in the image. The principle of adaptive median filtering can be expressed as: ; in, Indicates that the denoised image is at position The pixel value of Indicates The neighborhood pixel value set centered on is a pixel. Through filtering operations, the noise components in the image are significantly reduced, and a clearer denoised CT image is obtained. The denoised CT image is subjected to grayscale value stretching and contrast enhancement processing. The grayscale distribution of the image is determined by calculating the global grayscale histogram. The calculation of the global grayscale histogram can help determine the appropriate grayscale value stretching range, so as to linearly map the grayscale value in subsequent image processing. The formula for linear mapping is as follows: ; in, Indicates that the enhanced CT image is at position The pixel value of represents the original gray value of the denoised CT image, and Represent the minimum and maximum grayscale values ​​in the image, respectively. and Represent the minimum and maximum grayscale values ​​after linear mapping. Through the linear mapping operation, the contrast of the image is significantly improved. Multi-scale anisotropic diffusion filtering is performed on the contrast-enhanced CT image. By performing diffusion operations at different scales, the edge features in the image are effectively enhanced while reducing background noise and irrelevant details. Multi-scale anisotropic diffusion filtering is accomplished by solving the nonlinear diffusion equation, which is expressed as: ; in, represents the image gradient, is a function used to control the diffusion speed, which depends on the size of the image gradient. In the form of, the noise is removed while retaining the edge features to obtain an edge-enhanced CT image. The edge-enhanced CT image is binarized to highlight the crack area and separate it from the background. The global optimal threshold of the image is calculated, which can be achieved by methods such as the Otsu algorithm. Combined with the local adaptive threshold processing technology, the image is binarized to obtain a preliminary binary image. In this process, the connected domain analysis is used to cluster the pixels in the binary image, identify and separate the interconnected areas, and obtain the preprocessed CT image. Connected domain analysis can effectively separate the crack area from other irrelevant features by analyzing the connectivity between pixels.

[0029] In a specific embodiment, the process of executing step 200 may specifically include the following steps: The preprocessed CT images are subjected to block processing, and each preprocessed CT image is divided into a plurality of overlapping sub-image blocks to obtain a CT image block set; The CT image block set is input into the ResNet50 network, and features are extracted through 5 residual blocks. Each residual block contains 3 convolutional layers and short-circuit connections to obtain a 4096-dimensional global feature vector. The CT image block set is input into the DenseNet121 network, and features are extracted through 4 dense blocks. Each dense block contains multiple convolutional layers and dense connections to obtain a 1024-dimensional local feature vector. Perform bilinear pooling operations on the global feature vector and the local feature vector, obtain a 16384-dimensional fused feature vector through outer product operations and feature reshaping, and perform dimensionality reduction processing on the fused feature vector to obtain a 2048-dimensional reduced-dimensional feature vector; The reduced-dimensional feature vector is input into the support vector machine classifier, and the radial basis function kernel is used for nonlinear mapping and optimal separation hyperplane solution to obtain the preliminary classification results; Based on the preliminary classification results, the classification results of adjacent image blocks are contextually optimized to minimize the energy function and obtain the optimized classification results. Conduct confidence analysis on the optimized classification results, convert the output of the support vector machine into a probability distribution, and obtain the probability of crack existence in each image block; Based on the probability of crack existence, decision fusion is performed on the overlapping areas, and the image judged to have cracks is obtained by calculating the weighted average probability and setting the threshold.

[0030] Specifically, the preprocessed CT image is divided into blocks, and each preprocessed CT image is divided into multiple overlapping sub-images to obtain a set of CT image blocks. The set of CT image blocks is input into the ResNet50 network for feature extraction. The ResNet50 network consists of five residual blocks, each of which contains three convolutional layers and short-circuit connections. Through these residual blocks, ResNet50 can effectively extract deep features in the image. The residual block introduces short-circuit connections to alleviate the gradient vanishing problem in the deep network. The basic operation of the residual block can be expressed as: ; in, represents the output of the residual block, For input, It is a nonlinear transformation function composed of multiple convolutional layers and activation functions. Represents the weight parameters of the convolutional layer. With such a structure, ResNet50 can not only capture the complex features in the image at a deep level, but also maintain the stability of the calculation. After being processed by five residual blocks, the extracted global features are integrated into 4096-dimensional global feature vectors, which condense the key information in the image blocks and reflect the overall structural characteristics of the CT image. The same set of CT image blocks is input into the DenseNet121 network for feature extraction. The DenseNet121 network consists of four dense blocks, each of which contains multiple convolutional layers, which are connected by dense connections. The characteristic of dense connection is that the output of each layer is not only passed to the next layer, but also connected to all subsequent layers to form a dense connection structure. This connection method can effectively utilize low-level features and transfer more information between the layers of the network, thereby enhancing the expression ability of features. The operation of dense blocks can be expressed as: ; in, Represents the first Layer output, is the output of the previous layer, Representative Convolution operation of the layer. Through this structure, DenseNet121 can extract more detailed local features and obtain a 1024-dimensional local feature vector. The global feature vector and local feature vector extracted from ResNet50 and DenseNet121 are subjected to bilinear pooling. Bilinear pooling fuses global features with local features through outer product operations and feature reshaping to generate a 16384-dimensional fused feature vector. Bilinear pooling can capture high-order interaction information between different features and improve the expressiveness of the model. The calculation of the fused feature vector can be expressed as: ; in, represents the fused feature vector, and are global features and local features respectively. represents the outer product operation, is a vectorization operation. Through this operation, a high-dimensional fused feature vector is generated. In order to reduce the computational complexity and the risk of overfitting, the 16384-dimensional fused feature vector is reduced in dimension to obtain a 2048-dimensional reduced feature vector. The reduced feature vector is input into the support vector machine classifier for preliminary classification of cracks. The radial basis function kernel (RBF kernel) is used to perform nonlinear mapping on the feature vector, and the optimal separation hyperplane is constructed in the high-dimensional space to maximize the boundaries between different classes. The classification decision function of the support vector machine can be expressed as: ; in, is the classification decision function, is the support vector, is the Lagrange multiplier, is the kernel function (here is the RBF kernel), is the bias term. Through this process, the preliminary classification results of each image block are obtained. In order to improve the classification accuracy, the classification results of adjacent image blocks are contextually optimized based on the preliminary classification results. By minimizing the energy function and adjusting the classification consistency between adjacent image blocks, the optimized classification results are obtained, which effectively eliminates the noise and errors that may exist in the classification process. The confidence analysis is performed on the optimized classification results, and the output of the support vector machine is converted into a probability distribution to obtain the probability of crack existence in each image block. Based on these probability values, the overlapping areas are fused. By calculating the weighted average probability and setting a reasonable threshold, it is finally determined which areas have cracks and the final crack detection result image is generated.

[0031] In a specific embodiment, the step of inputting the CT image block set into the ResNet50 network, performing feature extraction through 5 residual blocks, each residual block including 3 convolutional layers and short-circuit connections, and obtaining a 4096-dimensional global feature vector may specifically include the following steps: Normalize the CT image block set, scale the pixel values ​​to the range of [0, 1], subtract the pre-calculated mean, and divide by the standard deviation to obtain a standardized CT image block; The standardized CT image block is input into the initial convolution layer of ResNet50, and the first convolution feature map is obtained using a 7×7 convolution kernel, a stride of 2, and maximum pooling; The first convolution feature map is input into the first residual block group, which contains 3 residual units. Each residual unit uses a 1×1, 3×3, and 1×1 convolution sequence and identity mapping, and fuses features by element-by-element addition to obtain the second convolution feature map. The second convolution feature map is downsampled, a convolution operation with a step size of 2 is used, and the second residual block group is input, which contains 4 residual units. Each residual unit uses the same convolution sequence and projection short circuit to obtain the third convolution feature map; The third convolution feature map is input into the third residual block group, which contains 6 residual units. The dilated convolution is used instead of the standard convolution. The dilation rate is increased from 1 to 2 to increase the receptive field, and the fourth convolution feature map is obtained. The fourth convolutional feature map is input into the fourth residual block group, which includes 3 residual units. Each residual unit uses the SE module, and the channel weights are recalibrated through global average pooling and two fully connected layers to obtain the fifth convolutional feature map. The fifth convolution feature map is input into the last residual block group, which contains 3 residual units. Each residual unit uses group convolution and channel rearrangement operations to increase feature interaction while maintaining the parameter amount, and obtains the sixth convolution feature map; Global average pooling is performed on the sixth convolutional feature map to convert the two-dimensional feature map into a one-dimensional vector, and then the feature dimension is adjusted to 4096 through a fully connected layer to obtain a 4096-dimensional global feature vector.

[0032] Specifically, the CT image block set is normalized to scale the pixel values ​​of the image to a uniform range, usually [0, 1], to eliminate the brightness differences between different images and obtain standardized CT image blocks. The standardized CT image blocks are input into the initial convolution layer of the ResNet50 network. The initial convolution layer uses a 7×7 convolution kernel with a step size of 2 and is combined with a maximum pooling operation. This combination can effectively reduce the resolution of the image while extracting preliminary edge and texture features to obtain the first convolution feature map. Maximum pooling reduces the size of the data by taking the maximum value of the local area, retaining significant features and filtering out unimportant details. The first convolution feature map is input into the first residual block group of ResNet50. The residual block group contains three residual units, each of which consists of three convolution layers, namely a 1×1 convolution layer, a 3×3 convolution layer, and a 1×1 convolution layer. The residual block is designed to solve the common gradient vanishing problem in deep neural networks. It directly bypasses the convolutional layer through a short-circuit connection (or identity mapping) and adds the convolution result to the output. The output of the residual unit can be expressed as: ; in, is the output of the residual unit, is the input, represents the transformation after the convolution layer and activation function, is the weight of the convolution layer. The features after convolution transformation are directly added to the input through element-by-element addition to obtain the second convolution feature map. This structure allows the network to be deepened without significantly increasing the complexity, thereby improving the ability of feature extraction. The second convolution feature map is downsampled. In order to reduce the resolution of the feature map, a convolution operation with a stride of 2 is used, so that the size of the feature map is reduced by half while maintaining the main structure of the feature. The downsampled feature map is input into the second residual block group, which contains four residual units. Similar to the first residual block group, each residual unit uses a sequence of 1×1, 3×3, and 1×1 convolutions, and uses projection short-circuit connections to match the input and output sizes to generate the third convolution feature map. In order to expand the receptive field of the network, the third convolution feature map is input into the third residual block group. This residual block group contains six residual units, in which dilated convolution is used instead of standard convolution operations. Dilated convolution can expand the receptive field without increasing the amount of computation by increasing the spacing between convolution kernels (i.e., the dilation rate), thereby capturing a wider range of contextual information. The operation of dilated convolution can be expressed as: ; Where d represents the dilation rate. The dilation rate increases from 1 to 2, so that the feature map can retain more contextual information, and finally the fourth convolution feature map is obtained. The fourth convolution feature map is input into the fourth residual block group, which contains three residual units. In order to improve the sensitivity of the network to different channel features, the SE module (Squeeze-and-Excitation module) is introduced into each residual unit in the fourth residual block group. The SE module recalibrates the weight of each channel through global average pooling and two fully connected layers, so that the network can dynamically adjust the importance of each channel to obtain the fifth convolution feature map. The fifth convolution feature map is input into the last residual block group. This group also contains three residual units, each of which enhances the interactivity between features by using group convolution and channel rearrangement operations. Group convolution reduces the computational complexity and enhances the diversity of features by grouping channels and performing convolution separately while keeping the number of parameters unchanged. Channel rearrangement breaks the independence between channels and increases the mixing between channels to obtain the sixth convolution feature map. Perform global average pooling on the sixth convolutional feature map to convert the two-dimensional feature map into a one-dimensional vector. Perform global average pooling on each channel to obtain the global feature representation of each channel, thereby greatly reducing the feature dimension. The feature dimension is adjusted to 4096 through the fully connected layer, and finally a 4096-dimensional global feature vector is obtained.

[0033] In a specific embodiment, the execution step inputs the CT image block set into the DenseNet121 network, and extracts features through four dense blocks, each of which contains multiple convolutional layers and dense connections. The process of obtaining a 1024-dimensional local feature vector may specifically include the following steps: Performing data enhancement processing on the CT image block set, including random rotation, flipping and brightness adjustment, to obtain enhanced CT image blocks; The enhanced CT image block is input into the initial convolution layer of DenseNet121, and a convolution operation with a 7×7 convolution kernel and a stride of 2 is used, followed by a 3×3 maximum pooling to obtain the initial feature map; The initial feature map is input into the first dense block, which contains 6 dense layers. Each dense layer consists of 1×1 convolution and 3×3 convolution. The output of each layer is concatenated with the feature maps of all previous layers through the connection function to obtain the first residual feature map. The first residual feature map is downsampled by 1×1 convolution and 2×2 average pooling, and then input into the second dense block, which contains 12 dense layers. The dilated convolution is used instead of the standard 3×3 convolution, and the dilation rate is changed alternately from 1 to 2 to obtain the second residual feature map. The second residual feature map is compressed and downsampled again, and then input into the third dense block, which contains 24 dense layers. The attention mechanism is introduced, and a channel attention module is added after each dense layer to dynamically adjust the importance of the feature channel to obtain the third residual feature map. The third residual feature map is channel compressed and downsampled, and input into the fourth dense block, which contains 16 dense layers. Each dense layer uses group convolution and channel shuffle operations to enhance feature interaction while maintaining parameter efficiency, so as to obtain the fourth residual feature map; The fourth residual feature map is input into a transition layer, which includes a 1×1 convolution to reduce the number of channels and a global average pooling operation to convert the two-dimensional feature map into a one-dimensional feature vector; The one-dimensional feature vector is adjusted to 1024 dimensions through a fully connected layer, and Dropout regularization is applied to obtain a 1024-dimensional local feature vector.

[0034] Specifically, data enhancement processing is performed on the set of CT image blocks to increase the diversity of the data and improve the generalization ability of the model. Data enhancement processing includes operations such as random rotation, flipping, and brightness adjustment. Through these operations, a set of enhanced CT image blocks are obtained. The enhanced CT image blocks are input into the initial convolution layer of the DenseNet121 network. The initial convolution layer of DenseNet121 uses a 7×7 convolution kernel with a step size of 2, followed by a 3×3 maximum pooling operation. This combination enables the network to quickly reduce the size of the image while extracting basic edge and texture features. The formula for the convolution operation can be expressed as: ; in, Indicates that the convolution output feature map is at position The value of is the weight of the convolution kernel, For the input image at position The pixel value of is the bias term. After being processed by the initial convolutional layer, the initial feature map obtained retains the main structural features of the image and passes these features to subsequent layers. The initial feature map is input into the first dense block of DenseNet121. The dense block contains six dense layers, each of which consists of a 1×1 convolutional layer and a 3×3 convolutional layer. The core feature of the dense block is that the output of each layer is not only passed to the next layer but also spliced ​​with the output feature maps of all previous layers. In this way, the network can make full use of low-level features and enhance the transmission of information. The formula for the splicing operation can be expressed as: ; in, Indicates The output of the layer, For the Nonlinear transformation of layers, Represents the concatenation operation of the feature map. Through the connection function, DenseNet can extract richer features while maintaining computational efficiency. After processing in the dense block, the first residual feature map is obtained. In order to reduce the size and computational complexity of the feature map, the first residual feature map is subjected to channel compression and downsampling. A 1×1 convolution layer is used to compress the number of channels, and a 2×2 average pooling operation is used for downsampling, thereby effectively reducing the data dimension while maintaining the main features. The processed feature map is input into the second dense block, which contains 12 dense layers, similar to the first dense block. The standard 3×3 convolution is replaced by a dilated convolution, and the dilation rate varies alternately in each layer, ranging from 1 to 2. The introduction of the dilated convolution can expand the receptive field, capture a wider range of contextual information, and obtain the second residual feature map. The second residual feature map is processed by channel compression and downsampling and input into the third dense block. The dense block contains 24 dense layers. In order to improve the expressiveness and accuracy of the features, the attention mechanism is introduced. A channel attention module is added after each dense layer to dynamically adjust the importance of feature channels so that the model can focus more on key features. The formula for the attention mechanism is: ; in, is the input feature, and is the weight matrix, is the Sigmoid activation function. Through this mechanism, the network can more flexibly adjust the weights of each channel to obtain the third residual feature map. After channel compression and downsampling, the third residual feature map is input into the fourth dense block, which contains 16 dense layers. In order to enhance feature interaction while maintaining parameter efficiency, each dense layer uses group convolution and channel shuffling operations. Group convolution reduces the number of parameters by dividing the channel into multiple groups and performing convolutions separately, while the channel shuffling operation enhances the diversity and expression of features by disrupting the channel order to generate the fourth residual feature map. The fourth residual feature map is input into the transition layer. The transition layer contains a The convolution layer is used to reduce the number of channels and convert the two-dimensional feature map into a one-dimensional feature vector through the global average pooling operation. Global average pooling reduces the dimension of the feature by averaging each channel and summarizing the features of each channel into one value. The one-dimensional feature vector is passed through the fully connected layer to adjust its dimension to 1024. In order to prevent overfitting, Dropout regularization is applied after the fully connected layer to randomly set some neurons to zero, enhance the robustness and generalization ability of the model, and obtain a 1024-dimensional local feature vector.

[0035] In a specific embodiment, the process of executing step 300 may specifically include the following steps: Perform edge detection on the image determined to have cracks to obtain a binary image of the crack edge; Based on the crack edge binary image, the distance to the nearest edge pixel is calculated for each pixel to obtain a distance transformation image; According to the distance transformation diagram, the initial center line of the crack is extracted to obtain the initial crack center line diagram; Optimize the initial crack centerline diagram to eliminate noise and discontinuities, and obtain the optimized crack centerline; Based on the optimized crack centerline, the vertical distance from each point on the centerline to the crack edge is calculated to obtain the local crack width value; The local crack width values ​​are smoothed and mapped to the original image size to obtain the crack width map.

[0036] Specifically, edge detection is performed on the image determined to have cracks, and the boundaries of the cracks are identified and extracted. Common edge detection methods include the Sobel operator, Canny edge detection, etc., which determine the edge position by detecting the sharp change of the grayscale value of the image. For example, by applying the Canny edge detector, the original image is converted into a crack edge binary image. In the crack edge binary image, the edge pixels of the crack are marked as 1, and the pixels in other areas are marked as 0, thereby clearly separating the outline of the crack. Based on the crack edge binary image, the distance to the nearest edge pixel is calculated for each pixel to generate a distance transformation map. The distance transformation map assigns a value to each non-edge pixel, representing the distance from the pixel to the nearest edge pixel. The distance transformation can be calculated using Euclidean distance, Manhattan distance, or Chebyshev distance. Assuming that Euclidean distance is used, the formula is as follows: ; in, Represents pixels in the distance transform map The value of represents the set of all edge pixels, is the coordinate of the edge pixel. By calculation, each pixel value in the distance transform map represents the distance between the pixel and the nearest crack edge. Using the generated distance transform map, the initial centerline of the crack is extracted. The centerline of the crack is usually located on the central axis of the crack and represents the geometric center of the crack. The initial centerline can be achieved by finding the local maximum in the distance transform map, and the maximum value is often located at the center of the crack. By connecting the local maxima, an initial crack centerline map is formed to show the center trajectory of the crack. The initial centerline map is optimized to eliminate unnecessary details and breakpoints. The optimization can be achieved through morphological operations (such as corrosion and expansion) and median filtering. These operations can smooth the centerline, repair the broken parts, and obtain the optimized crack centerline. Based on the optimized crack centerline, the vertical distance from each point on the centerline to the crack edge is calculated to obtain the local crack width value. The distance from each centerline point to the crack edge is measured on both sides. These distances can be calculated in the direction perpendicular to the centerline, and the formula is as follows: ; in, Indicates the centerline point The crack width value at and Respectively indicate points The distance to the left and right edges of the crack. The width of the crack at different positions is calculated by the formula to obtain a series of local crack width values. The local crack width values ​​are smoothed. By moving average or Gaussian filtering, high-frequency noise in the local width values ​​is effectively eliminated to generate a smoother width curve. The smoothed width values ​​are mapped back to the size of the original image to generate a crack width map.

[0037] In a specific embodiment, the process of executing step 400 may specifically include the following steps: Perform threshold segmentation on the crack width map, mark pixels with width values ​​greater than a preset threshold as seed points, and obtain an initial seed point set; Based on the initial seed point set, the region growing algorithm is executed to include pixels adjacent to the seed point and with similar width values ​​into the growth region to obtain the preliminary crack region; Perform skeleton extraction on the preliminary crack area to obtain the crack skeleton, and mark the endpoints and branch points on the crack skeleton; Based on the crack skeleton, the Euclidean distance and direction angle between adjacent endpoints are calculated, and the endpoint pairs with a distance less than a threshold and a close direction angle are connected to obtain the connected crack skeleton; Reconstruct the crack region according to the connected crack skeleton, and use the expansion operation to expand the skeleton to the original crack width to obtain the reconstructed crack region; Perform morphological closing operation on the reconstructed crack area to fill the internal small holes and gaps to obtain the complete crack area; Based on the complete crack area, the geometric characteristics of the crack, including length, average width, maximum width, direction and curvature, are extracted to obtain the crack geometric parameter set; The crack geometric parameter set and the grayscale information of the original CT image are combined to calculate the crack depth and volume, and obtain the crack quantitative parameter set.

[0038] Specifically, the crack width map is threshold segmented. By setting a predetermined threshold, pixels with width values ​​greater than the threshold are marked as seed points, and key points that may belong to the main crack area are identified. These seed points constitute the initial seed point set, representing the crack parts with larger width in the image. Based on the initial seed point set, the region growing algorithm is executed. Region growing is a commonly used image segmentation technology. Its principle is to start from the seed point and gradually incorporate pixels adjacent to the seed point and with similar features into the growth region. Feature similarity refers to the fact that the width values ​​of adjacent pixels are close to the width value of the initial seed point. The formula for region growing can be expressed as: ; in, represents the growth area, is the seed point set, Represents pixels Neighborhood of and Pixels and The width value of is the allowed width value difference threshold. Through this process, the preliminary crack area is gradually expanded to obtain the preliminary crack area image. The preliminary crack area is subjected to skeleton extraction. Skeleton extraction is an image processing technique whose goal is to compress the shape information in the area into a center line, that is, the skeleton of the crack. Skeleton extraction can be achieved through morphological operations or distance transformations. The obtained skeleton image retains the main geometric shape of the crack and removes redundant information. Mark the endpoints and branch points on the crack skeleton. The endpoint is the end position of the skeleton, while the branch point indicates the bifurcation of the crack. Based on the extracted crack skeleton, the Euclidean distance and direction angle between adjacent endpoints are calculated. The Euclidean distance measures the straight-line distance between two endpoints, while the direction angle reflects the relative direction between the two endpoints. The formula is as follows: ; in, Representation endpoint and The Euclidean distance between them. By comparing these distances and angles, the endpoint pairs whose distance is less than the set threshold and whose direction angles are close are connected to repair possible fractures and obtain the connected crack skeleton. The crack area is reconstructed according to the connected crack skeleton. The skeleton is expanded to the width of the original crack. The morphological expansion operation is used, and its formula is: ; in, represents the reconstructed crack area, is the skeleton image, represents the morphological dilation operation, is the structural element. The expansion operation expands the skeleton to a width similar to the original crack to generate a reconstructed crack area image. The reconstructed crack area is subjected to a morphological closing operation. The closing operation fills the small holes and gaps inside the crack by first dilating and then corroding, making the crack area more complete and smooth. The closing operation can effectively eliminate the tiny holes caused by noise and enhance the overall coherence of the crack to obtain a complete crack area. Based on the complete crack area, the geometric features of the crack are extracted, including the length, average width, maximum width, direction and curvature of the crack. The formula is as follows: ; in, represents the total length of the crack, The first The coordinates of the points. Combined with the extracted crack geometry parameter set and the grayscale information of the original CT image, the depth and volume of the crack are calculated. The depth can be estimated by analyzing the performance of the crack in different slices, while the volume can be obtained by accumulating the area of ​​the crack in each slice. The formula is: ; in, represents the volume of the crack, For the The area of ​​the crack in each slice, is the slice spacing. Through calculation, a quantitative parameter set of the crack is obtained, including geometric characteristics and volume information.

[0039] In a specific embodiment, the process of executing step 500 may specifically include the following steps: Based on the geometric information of the complete crack area and the aluminum alloy component, a three-dimensional finite element model is constructed, and the mesh of the crack area is refined to obtain a discretized structural model. According to the discretized structural model, the mass matrix and stiffness matrix are established, and the eigenvalue problem is solved to obtain the natural frequency and the corresponding eigenvector; The eigenvectors are normalized and the node displacements are converted into continuous mode shapes to obtain the vibration modes of the aluminum alloy component. Based on the natural frequency and vibration mode, the modal participation factor and effective mass are calculated, the contribution of each mode to the overall response is evaluated, and the modal importance ranking is obtained; According to the order of modal importance, the first N main modes are selected, the proportional damping matrix is ​​constructed using the Rayleigh damping model, and the harmonic response analysis is performed to calculate the frequency response function under different excitation frequencies to obtain the harmonic response characteristics of the aluminum alloy component; Combining the natural frequency, vibration mode and harmonic response characteristics, characteristic parameters such as frequency shift, modal curvature and transfer function are extracted to obtain vibration characteristic data.

[0040] Specifically, a three-dimensional finite element model is constructed based on the complete crack area and the geometric information of the component. By inputting the geometric shape and material properties of the component into the finite element software, a preliminary model describing the mechanical behavior of the component is generated. In this model, the crack area is meshed and refined, and the density of nodes and units is increased at the crack to capture the effect of the crack on the structural performance of the component, and a discretized structural model is obtained. According to the discretized structural model, the mass matrix and stiffness matrix are established. The mass matrix represents the inertial characteristics of the system, while the stiffness matrix describes the elastic characteristics of the system. The mass matrix is ​​usually a diagonal matrix, which represents the mass distribution of each node; the stiffness matrix is ​​formed by assembling the elastic characteristics of each unit. Solving the eigenvalue problem is to solve the following characteristic equation: ; in, is the stiffness matrix, is the mass matrix, is the characteristic value, is the corresponding eigenvector. The square root of is the natural frequency of the system, and the eigenvector It corresponds to the vibration mode at this natural frequency. By solving this equation, the natural frequency of the component and its corresponding vibration mode are obtained. The eigenvector is normalized. The eigenvector is standardized to meet specific specification conditions for subsequent calculations. Usually, the modulus of the normalized eigenvector is set to 1, or the mass of the eigenvector is normalized, that is: ; After normalization, the eigenvector can more accurately reflect the relative movement of each node of the component. The node displacement is converted into a continuous modal shape to obtain the vibration mode of the aluminum alloy component at each natural frequency. The modal shape shows the deformation mode of the component under different modes. Based on the natural frequency and vibration mode, the modal participation factor and effective mass are calculated. The modal participation factor indicates the response intensity of a mode under a specific load, while the effective mass measures the contribution of the mode to the total mass. Through calculation, the importance of each mode is ranked to obtain the modal importance ranking. According to the modal importance ranking, the top N main modes are selected, and the proportional damping matrix is ​​constructed using the Rayleigh damping model. The Rayleigh damping model assumes that the damping of the system is proportional to the mass and stiffness, and the damping matrix can be expressed as: ; in, is the damping matrix, and is the proportionality coefficient, is the mass matrix, is the stiffness matrix. By setting appropriate coefficients and , and obtain the damping characteristics that meet the actual engineering needs. Perform harmonic response analysis on the component and calculate the frequency response function at different excitation frequencies. The frequency response function reflects the response amplitude and phase of the component under a specific frequency excitation. Combined with the calculated natural frequency, vibration mode and harmonic response characteristics, a series of characteristic parameters are extracted, such as frequency offset, modal curvature and transfer function. Frequency offset refers to the difference between the frequency of the actual system and the theoretical natural frequency; modal curvature reflects the degree of curvature of the modal shape; and the transfer function describes the relationship between the input excitation and the output response. Through these characteristic parameters, the vibration behavior of the component is understood and complete vibration characteristic data is obtained.

[0041] In a specific embodiment, the process of executing step 600 may specifically include the following steps: Based on the vibration characteristic data and the crack quantitative parameter set, a continuous damage mechanics model of aluminum alloy components is constructed, and the stress intensity factor at the crack tip is calculated using the J-integral method. Its accuracy is verified by the strain energy release rate, and the stress field distribution at the crack tip is obtained. According to the stress field distribution at the crack tip, the fatigue crack growth rate equation is established using the Paris law, and the influence of stress ratio is considered through Walker correction to obtain the corrected crack growth rate model. Based on the modified crack growth rate model, the numerical integration method with adaptive step size is used to simulate the crack growth path and obtain the evolution curve of crack length with the number of cycles. The crack length evolution curve is fitted piecewise linearly to extract the crack growth parameters at each stage, and the influence of the critical stress intensity factor is considered in combination with the Forman equation to obtain the crack growth life prediction model. Based on the crack growth life prediction model, combined with the Monte Carlo simulation method, considering the randomness of material parameters and loads, multiple groups of life prediction results are generated to obtain the probability distribution of the remaining life; Reliability analysis is performed on the probability distribution of the remaining life, the remaining life under a given failure probability is calculated, and the target influencing factors are determined through sensitivity analysis to obtain the remaining life prediction data of aluminum alloy components.

[0042] Specifically, the vibration characteristic data (such as natural frequency, vibration mode) is combined with the crack geometric parameters (such as length, width, depth). Through finite element analysis or other numerical methods, a three-dimensional continuous damage mechanics model containing cracks is established to describe the stress concentration phenomenon around the crack and the influence of crack extension on the overall structural performance. The J integral method is used to calculate the stress intensity factor at the crack tip. The J integral is a path-independent integral method used to evaluate the stress intensity near the crack tip. The J integral is defined as: ; in, represents the energy release rate at the crack tip, is the integral path around the crack tip, is the strain energy density function, is the stress component, is the displacement component, is the normal vector, is the differential length element on the path. By solving the J integral, the stress intensity factor at the crack tip is obtained, which is an important parameter to characterize the stress field at the crack tip. In order to verify the accuracy of the J integral calculation, the strain energy release rate is used for verification. The strain energy release rate is an important indicator to describe material failure, which is defined as the strain energy released per unit length of crack extension. By comparing the results calculated by the J integral method with the results calculated by the strain energy release rate, the accuracy and consistency of the stress field distribution at the crack tip are ensured. According to the stress field distribution at the crack tip, the crack growth rate equation is established using the Paris law. The Paris law describes the relationship between the crack growth rate and the range of stress intensity factors, and its expression is: ; in, is the crack growth rate, is the crack length, is the number of cycles, is the stress intensity factor range, and is the material constant. It is corrected by Walker correction. The crack growth rate model expression after Walker correction is: ; in, is the stress ratio, is a correction index. Through this correction model, the crack expansion behavior under different stress conditions can be predicted more accurately. Based on the corrected crack growth rate model, the numerical integration method with adaptive step size is used to simulate the crack expansion path. The adaptive step size method can improve the efficiency and accuracy of numerical integration by dynamically adjusting the integration step size. Through simulation calculation, the evolution curve of crack length with the number of cycles is obtained, showing the process of gradual expansion of the crack under fatigue load. In order to analyze the crack growth process, the crack length evolution curve is piecewise linearly fitted. Piecewise linear fitting can identify the different stages in the crack growth process and extract the crack growth parameters (such as growth rate, crack starting point, etc.) of each stage. These parameters can reflect the expansion characteristics of the crack at different stages, and combine the Forman equation to consider the influence of the critical stress intensity factor, and establish a crack growth life prediction model. The Forman equation is an improved fatigue crack growth model that can provide more accurate predictions when approaching fracture. Its expression is: ; in, is the critical stress intensity factor. Through this model, the crack propagation life before reaching fracture is predicted. Based on the crack propagation life prediction model, the Monte Carlo simulation method is combined to consider the randomness of material parameters and loads. Monte Carlo simulation generates a large number of possible material properties and load conditions through random sampling to simulate the crack propagation behavior under different conditions. Through this method, multiple sets of crack propagation life prediction results are generated, and the probability distribution of the remaining life is obtained, reflecting the failure probability and remaining life of the component under different conditions. Reliability analysis is performed on the generated remaining life probability distribution. Calculate the remaining life under a given failure probability, that is, the remaining service life of the component at a specific reliability level. Determine which factors have the greatest impact on the remaining life through sensitivity analysis, which helps to identify key material properties or load conditions and take targeted measures to extend the service life of the component.

[0043] The above describes the method for detecting cracks in an aluminum alloy component in the embodiment of the present application. The following describes the detection device 10 for detecting cracks in an aluminum alloy component in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of a detection device 10 for cracks in an aluminum alloy component includes: The scanning module 11 is used to perform CT scanning and preprocessing on the aluminum alloy component to obtain a preprocessed CT image; The identification module 12 is used to input the preprocessed CT image into the integrated classification model of the bilinear convolutional neural network and the support vector machine classifier to perform crack identification, and obtain an image that is determined to have cracks; The transformation module 13 is used to transform the crack width of the image determined to have cracks, calculate the distance from the pixel to the edge, extract the center line, and generate a crack width map; A search module 14 is used to perform a regional growth algorithm search based on the crack width map, connect the disconnected crack segments, fill the small holes inside the cracks, and obtain a complete crack region and a corresponding crack quantitative parameter set; An analysis module 15 is used to perform vibration analysis on the aluminum alloy component according to the complete crack area, calculate the natural frequency and mode, and obtain vibration characteristic data; The calculation module 16 is used to calculate the stress intensity factor based on the vibration characteristic data and the crack quantitative parameter set using the continuous damage mechanics model, and perform fatigue crack propagation analysis to obtain the remaining life prediction data of the aluminum alloy component.

[0044] Through the synergy of the above components, the integrated classification model combining bilinear convolutional neural network and support vector machine classifier can effectively process high-noise and low-contrast CT images, improve the accuracy of crack identification, and reduce the missed detection rate. The crack width transformation and regional growth algorithm can accurately extract the geometric characteristics of complex-shaped cracks, including length, width, direction and curvature, and provide reliable quantitative parameters for subsequent analysis. By connecting the disconnected crack segments and filling the internal small holes, a more complete crack area can be obtained, which improves the accuracy of subsequent analysis. Based on the finite element modeling and modal analysis of the complete crack area, the vibration characteristic data of the aluminum alloy component can be obtained, which provides a basis for evaluating the impact of cracks on the overall performance of the structure. Combining the continuous damage mechanics model and Monte Carlo simulation method, the randomness of material parameters and loads can be considered, and the probability distribution of the remaining life can be given, thereby improving the accuracy of crack identification, realizing the accurate extraction of crack characteristics, and on this basis, reliable remaining life prediction can be carried out.

[0045] The present application also provides an electronic device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the aluminum alloy component crack detection method in the above-mentioned embodiments.

[0046] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein instructions are stored in the computer, and when the instructions are executed on a computer, the computer executes the steps of the method for detecting cracks in an aluminum alloy component.

[0047] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0048] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0049] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting cracks in an aluminum alloy component, characterized in that: The method comprises: Performing CT scanning and preprocessing on the aluminum alloy component to obtain a preprocessed CT image; Inputting the preprocessed CT image into an integrated classification model of a bilinear convolutional neural network and a support vector machine classifier for crack discrimination, thereby obtaining an image judged to have cracks; Perform crack width transformation on the image determined to have cracks, calculate the distance from the pixel to the edge, extract the center line, and generate a crack width map; Based on the crack width map, a regional growth algorithm search is performed to connect the disconnected crack segments and fill the small holes inside the cracks to obtain a complete crack region and a corresponding crack quantitative parameter set; Performing vibration analysis on the aluminum alloy component according to the complete crack region, calculating the natural frequency and mode, and obtaining vibration characteristic data; According to the vibration characteristic data and the crack quantitative parameter set, a continuous damage mechanics model is used to calculate the stress intensity factor, and a fatigue crack propagation analysis is performed to obtain the remaining life prediction data of the aluminum alloy component.

2. The method for detecting cracks in an aluminum alloy component according to claim 1, characterized in that: The step of performing CT scanning and preprocessing on the aluminum alloy component to obtain a preprocessed CT image includes: Performing high-resolution CT scanning on the aluminum alloy component to obtain an original CT image sequence, and converting the original CT image sequence into a 16-bit grayscale PNG format image; Adaptively median filtering the 16-bit grayscale PNG format image, calculating the median of the pixel values ​​in the local area, and replacing the pixel values ​​with the median to obtain a denoised CT image; Calculating a global grayscale histogram based on the denoised CT image, determining a stretching range, and performing linear mapping on the grayscale values ​​to obtain a contrast-enhanced CT image; Performing multi-scale anisotropic diffusion filtering on the contrast-enhanced CT image to obtain an edge-enhanced CT image; Based on the edge-enhanced CT image, a global optimal threshold is calculated, and a preliminary binarized image is obtained in combination with a local adaptive threshold, and a connected domain analysis is performed on the preliminary binarized image to obtain a preprocessed CT image.

3. The method for detecting cracks in an aluminum alloy component according to claim 1, characterized in that: The preprocessed CT image is input into an integrated classification model of a bilinear convolutional neural network and a support vector machine classifier for crack discrimination to obtain an image determined to have cracks, including: Performing block processing on the preprocessed CT image, dividing each preprocessed CT image into a plurality of overlapping sub-image blocks, to obtain a CT image block set; The CT image block set is input into the ResNet50 network, and features are extracted through 5 residual blocks, each of which contains 3 convolutional layers and short-circuit connections to obtain a 4096-dimensional global feature vector; The CT image block set is input into the DenseNet121 network, and features are extracted through 4 dense blocks, each of which contains multiple convolutional layers and dense connections to obtain a 1024-dimensional local feature vector; Performing a bilinear pooling operation on the global feature vector and the local feature vector, obtaining a 16384-dimensional fused feature vector through outer product operation and feature reshaping, and performing dimensionality reduction processing on the fused feature vector to obtain a 2048-dimensional reduced-dimensional feature vector; Inputting the dimension-reduced feature vector into a support vector machine classifier, using a radial basis function kernel to perform nonlinear mapping and solve the optimal separation hyperplane to obtain a preliminary classification result; Based on the preliminary classification results, context optimization is performed on the classification results of adjacent image blocks to minimize the energy function and obtain an optimized classification result; Performing confidence analysis on the optimized classification results, converting the output of the support vector machine into a probability distribution, and obtaining the probability of crack existence in each image block; Based on the crack existence probability, decision fusion is performed on the overlapping areas, and an image determined to have cracks is obtained by calculating the weighted average probability and setting a threshold.

4. The method for detecting cracks in an aluminum alloy component according to claim 3, characterized in that: The CT image block set is input into the ResNet50 network, and features are extracted through 5 residual blocks, each of which contains 3 convolutional layers and short-circuit connections, to obtain a 4096-dimensional global feature vector, including: Normalizing the CT image block set, scaling pixel values ​​to a range of [0, 1], subtracting a pre-calculated mean, and dividing by a standard deviation to obtain a standardized CT image block; Input the standardized CT image block into the initial convolution layer of ResNet50, use a 7×7 convolution kernel, a convolution operation with a step size of 2, and a maximum pooling to obtain a first convolution feature map; Input the first convolution feature map into the first residual block group, which includes 3 residual units, each residual unit uses a 1×1, 3×3, 1×1 convolution sequence and identity mapping, and fuses features by element-by-element addition to obtain a second convolution feature map; Down-sample the second convolution feature map, use a convolution operation with a step size of 2, and input a second residual block group, including 4 residual units, each residual unit uses the same convolution sequence and projection short circuit, to obtain a third convolution feature map; Input the third convolutional feature map into a third residual block group, including 6 residual units, use dilated convolution instead of standard convolution, and increase the dilation rate from 1 to 2 to increase the receptive field, so as to obtain a fourth convolutional feature map; Inputting a fourth residual block group into the fourth convolutional feature map, including 3 residual units, each residual unit using an SE module, recalibrating channel weights through global average pooling and two fully connected layers, to obtain a fifth convolutional feature map; Inputting the fifth convolutional feature map into the last residual block group, which includes 3 residual units, each residual unit uses group convolution and channel rearrangement operations to increase feature interaction while maintaining the parameter amount, so as to obtain a sixth convolutional feature map; Global average pooling is performed on the sixth convolutional feature map, the two-dimensional feature map is converted into a one-dimensional vector, and then the feature dimension is adjusted to 4096 through a fully connected layer to obtain a 4096-dimensional global feature vector.

5. The method for detecting cracks in an aluminum alloy component according to claim 3, characterized in that: The CT image block set is input into the DenseNet121 network, and feature extraction is performed through 4 dense blocks, each of which contains multiple convolutional layers and dense connections, to obtain a 1024-dimensional local feature vector, including: Performing data enhancement processing on the CT image block set, including random rotation, flipping and brightness adjustment, to obtain enhanced CT image blocks; The enhanced CT image block is input into the initial convolution layer of DenseNet121, and a convolution operation with a 7×7 convolution kernel and a step size of 2 is performed, and a 3×3 maximum pooling is performed to obtain an initial feature map; Input the initial feature map into the first dense block, which includes 6 dense layers, each of which consists of 1×1 convolution and 3×3 convolution, and concatenate the output of each layer with the feature maps of all previous layers through a connection function to obtain a first residual feature map; The first residual feature map is downsampled by performing channel compression by 1×1 convolution and 2×2 average pooling, and is input into a second dense block, which contains 12 dense layers. A dilated convolution is used instead of a standard 3×3 convolution, and the dilation rate is changed alternately from 1 to 2 to obtain a second residual feature map. The second residual feature map is compressed and downsampled again, and input into the third dense block, which contains 24 dense layers. The attention mechanism is introduced, and a channel attention module is added after each dense layer to dynamically adjust the importance of the feature channel to obtain the third residual feature map; Perform channel compression and downsampling on the third residual feature map, and input the result into a fourth dense block, where the dense block includes 16 dense layers, each dense layer uses group convolution and channel shuffle operations to enhance feature interaction while maintaining parameter efficiency, so as to obtain a fourth residual feature map; Inputting the fourth residual feature map into a transition layer, the transition layer including a 1×1 convolution for reducing the number of channels and a global average pooling operation to convert the two-dimensional feature map into a one-dimensional feature vector; The one-dimensional feature vector is adjusted to a feature dimension of 1024 through a fully connected layer, and Dropout regularization is applied to obtain a 1024-dimensional local feature vector.

6. The method for detecting cracks in an aluminum alloy component according to claim 1, characterized in that: The crack width transformation is performed on the image determined to have cracks, the distance from the pixel to the edge is calculated, the center line is extracted, and the crack width map is generated, including: Performing edge detection on the image determined to have cracks to obtain a binary image of crack edges; Based on the crack edge binary image, the distance to the nearest edge pixel is calculated for each pixel to obtain a distance transformation image; Extracting the initial centerline of the crack according to the distance transformation diagram to obtain an initial crack centerline diagram; Optimizing the initial crack centerline diagram to eliminate noise and discontinuities to obtain an optimized crack centerline; Based on the optimized crack centerline, the vertical distance from each point on the centerline to the crack edge is calculated to obtain a local crack width value; The local crack width value is smoothed and mapped to the original image size to obtain a crack width map.

7. The method for detecting cracks in an aluminum alloy component according to claim 1, characterized in that: The method of performing a regional growth algorithm search based on the crack width map, connecting the disconnected crack segments, and filling the small holes inside the cracks to obtain a complete crack region and a corresponding crack quantitative parameter set includes: Performing threshold segmentation on the crack width map, marking pixels with width values ​​greater than a preset threshold as seed points, and obtaining an initial seed point set; Based on the initial seed point set, a region growing algorithm is executed to include pixels adjacent to the seed point and having similar width values ​​into the growing region to obtain a preliminary crack region; Performing skeleton extraction on the preliminary crack region to obtain a crack skeleton, and marking endpoints and branch points on the crack skeleton; Based on the crack skeleton, calculating the Euclidean distance and direction angle between adjacent endpoints, connecting the endpoint pairs whose distance is less than a threshold and whose direction angle is close, to obtain a connected crack skeleton; Reconstructing the crack region according to the connected crack skeleton, and using an expansion operation to expand the skeleton to the original crack width to obtain a reconstructed crack region; Performing a morphological closing operation on the reconstructed crack region to fill small holes and gaps inside to obtain a complete crack region; Based on the complete crack region, geometric features of the crack are extracted, including length, average width, maximum width, direction and curvature, to obtain a crack geometric parameter set; The crack geometric parameter set and the grayscale information of the original CT image are combined to calculate the depth and volume of the crack to obtain a crack quantitative parameter set.

8. The method for detecting cracks in an aluminum alloy component according to claim 1, characterized in that: The vibration analysis of the aluminum alloy component is performed according to the complete crack region, the natural frequency and mode are calculated, and the vibration characteristic data is obtained, including: Based on the geometric information of the complete crack region and the aluminum alloy component, a three-dimensional finite element model is constructed, and the crack region is meshed to obtain a discretized structural model; According to the discretized structural model, a mass matrix and a stiffness matrix are established, and an eigenvalue problem is solved to obtain a natural frequency and a corresponding eigenvector; Normalizing the eigenvectors and converting the node displacements into continuous modal shapes to obtain vibration modes of the aluminum alloy component; Based on the natural frequency and the vibration mode, the modal participation factor and the effective mass are calculated, the contribution of each mode to the overall response is evaluated, and the modal importance ranking is obtained; According to the order of importance of the modes, the first N main modes are selected, a proportional damping matrix is ​​constructed using the Rayleigh damping model, and a harmonic response analysis is performed to calculate the frequency response function under different excitation frequencies to obtain the harmonic response characteristics of the aluminum alloy component; The natural frequency, the vibration mode and the harmonic response characteristics are combined to extract characteristic parameters such as frequency shift, modal curvature and transfer function to obtain vibration characteristic data.

9. The method for detecting cracks in an aluminum alloy component according to claim 1, characterized in that: The method of calculating the stress intensity factor using a continuous damage mechanics model based on the vibration characteristic data and the crack quantitative parameter set, and performing fatigue crack propagation analysis to obtain the remaining life prediction data of the aluminum alloy component includes: Based on the vibration characteristic data and the crack quantitative parameter set, a continuous damage mechanics model of the aluminum alloy component is constructed, and the stress intensity factor at the crack tip is calculated by using the J-integral method, and its accuracy is verified by the strain energy release rate to obtain the stress field distribution at the crack tip; According to the stress field distribution at the crack tip, the fatigue crack growth rate equation is established using the Paris law, and the influence of the stress ratio is considered through Walker correction to obtain a corrected crack growth rate model; Based on the modified crack growth rate model, a numerical integration method with an adaptive step size is used to simulate the crack growth path and obtain an evolution curve of the crack length with the number of cycles; Performing piecewise linear fitting on the crack length evolution curve, extracting crack extension parameters at each stage, and considering the influence of critical stress intensity factor in combination with Forman equation to obtain a crack extension life prediction model; Based on the crack growth life prediction model, combined with the Monte Carlo simulation method, considering the randomness of material parameters and loads, multiple groups of life prediction results are generated to obtain the probability distribution of the remaining life; A reliability analysis is performed on the probability distribution of the remaining life, the remaining life under a given failure probability is calculated, and target influencing factors are determined through sensitivity analysis to obtain the remaining life prediction data of the aluminum alloy component.

10. A device for detecting cracks in aluminum alloy components, characterized in that: The method for detecting cracks in an aluminum alloy component according to any one of claims 1 to 9, wherein the device for detecting cracks in the aluminum alloy component comprises: A scanning module, used for performing CT scanning and preprocessing on the aluminum alloy component to obtain a preprocessed CT image; A discrimination module, used for inputting the preprocessed CT image into an integrated classification model of a bilinear convolutional neural network and a support vector machine classifier for crack discrimination, and obtaining an image judged to have cracks; A transformation module, used to perform crack width transformation on the image determined to have cracks, calculate the distance from the pixel to the edge, extract the center line, and generate a crack width map; A search module is used to perform a regional growth algorithm search based on the crack width map, connect the disconnected crack segments, fill the small holes inside the cracks, and obtain a complete crack region and a corresponding crack quantitative parameter set; An analysis module, used to perform vibration analysis on the aluminum alloy component according to the complete crack area, calculate the natural frequency and mode, and obtain vibration characteristic data; A calculation module is used to calculate the stress intensity factor according to the vibration characteristic data and the crack quantitative parameter set using a continuous damage mechanics model, and perform fatigue crack propagation analysis to obtain the remaining life prediction data of the aluminum alloy component.

Images