Defect detection method and system applied to friction stir additive manufacturing process

Abstract

The invention discloses a defect detection method and system applied to a friction stir additive manufacturing process, and relates to the technical field of defect detection.The defect detection method comprises the steps that a temperature gradient direction field is obtained; according to the topological structure difference of two continuous frames of temperature gradient direction fields, an initial thermal anomaly region is determined; synchronously acquiring vibration acceleration signals; based on the multi-frequency-band coherence coefficient matrix difference of the vibration acceleration signal, determining an initial vibration abnormal area; obtaining a thermal vibration coupling topological graph according to the initial thermal anomaly region and the initial vibration anomaly region; determining a candidate defect area according to the thermal vibration coupling topological graph; aiming at each position in the candidate defect area, constructing a temperature vibration combined topological graph; obtaining a real defect area and a corresponding defect type according to the nonlinear correlation between the nodes of the temperature vibration combined topological graph; according to the method, real-time accurate recognition and active classification of defects in the friction stir additive manufacturing process are achieved, and the quality stability and reliability of manufactured parts are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of defect detection technology, and in particular to a defect detection method and system applied in a friction stir additive manufacturing process. Background Art

[0002] Friction stir additive manufacturing is an advanced manufacturing technology based on the principle of solid-state welding. It has been widely used in the fields of aerospace, automobile manufacturing, and high-end equipment manufacturing. This process deposits materials layer by layer through continuous friction heat and mechanical stirring, which can effectively avoid the thermal cracks, holes, and material oxidation problems caused by traditional melting additive manufacturing processes, and has excellent part performance. However, the friction stir additive manufacturing process is affected by the multiple coupling of material property differences, equipment vibration interference, and process parameter fluctuations. It is easy for subtle and difficult-to-detect microscopic defects such as holes, cracks, or local unbonded areas to occur inside the parts. These defects not only seriously reduce the mechanical properties and service life of the parts, but also greatly increase the safety risks of the parts in the service environment.

[0003] Currently, defect detection methods for friction stir additive manufacturing (FSAM) generally rely on subsequent nondestructive testing (NDT) techniques. However, these techniques are significantly limited in terms of both real-time and precision defect detection. This makes it difficult to accurately locate and proactively intervene in defect formation early in the manufacturing process, leading to the continuous accumulation of defects and ultimately severely impacting part quality. Furthermore, traditional monitoring methods based on a single physical field cannot accurately describe the complex coupling of thermal and vibration fields during the manufacturing process, resulting in inaccurate defect identification.

[0004] Therefore, the industry urgently needs a method that can integrate multi-physical field information and accurately locate manufacturing process defects in real time, so as to achieve accurate positioning and identification when the defects just appear or have not yet been fully formed. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a defect detection method applied to a stir friction additive manufacturing process.

[0006] To achieve the above objectives, the present invention provides, on the one hand, a defect detection method for use in a friction stir additive manufacturing process, comprising:

[0007] S101: Collecting a continuous infrared thermal image sequence of the friction stir additive manufacturing process, and calculating the corresponding temperature gradient direction field based on each frame of the infrared thermal image; determining the initial thermal anomaly area based on the topological structure difference of the temperature gradient direction field between two consecutive frames;

[0008] S102: synchronously collecting vibration acceleration signals at multiple spatial locations during the friction stir additive manufacturing process; and determining an initial vibration abnormality region based on differences in multi-frequency coherence coefficient matrices of the vibration acceleration signals at each spatial location;

[0009] S103: obtaining a thermal-vibration coupling topology map based on the spatial position topological correlation between the initial thermal abnormality region and the initial vibration abnormality region; and determining a candidate defect region based on the nonlinear manifold characteristics of the thermal-vibration coupling topology map.

[0010] S104: For each position in the candidate defect area, a temperature-vibration joint topology map is constructed according to the corresponding temperature gradient direction field and the vibration signal coherence coefficient matrix; and the real defect area and the corresponding defect type are obtained according to the nonlinear correlation between the nodes of the temperature-vibration joint topology map.

[0011] In another aspect, the present invention provides a defect detection system for use in a friction stir additive manufacturing process, which is based on the above-mentioned defect detection method for use in a friction stir additive manufacturing process and comprises:

[0012] The first acquisition module is used to collect a series of continuous infrared thermal images of the friction stir additive manufacturing process and calculate the corresponding temperature gradient direction field based on each frame of the infrared thermal image; based on the topological structure difference of the temperature gradient direction field between two consecutive frames, the initial thermal anomaly area is determined;

[0013] The second acquisition module is used to synchronously acquire vibration acceleration signals at multiple spatial positions during the friction stir additive manufacturing process;

[0014] Based on the difference in the coherence coefficient matrix of the vibration acceleration signals at each spatial position, the initial vibration abnormality area is determined;

[0015] A screening module is used to obtain a thermal-vibration coupling topology map based on the spatial position topological correlation between the initial thermal anomaly region and the initial vibration anomaly region; and to determine candidate defect regions based on the nonlinear manifold characteristics of the thermal-vibration coupling topology map;

[0016] The detection module is used to construct a temperature-vibration joint topology map for each location in the candidate defect area based on the corresponding temperature gradient direction field and the vibration signal coherence coefficient matrix; based on the nonlinear correlation between the nodes in the temperature-vibration joint topology map, the real defect area and the corresponding defect type are obtained.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] The present invention integrates the temperature field and vibration field information in the stir friction additive manufacturing process to construct a temperature-vibration joint topological analysis method, which can accurately determine the microscopic defect areas and corresponding defect types that appear in the manufacturing process in real time, effectively overcoming the delay and insufficient recognition accuracy problems of traditional subsequent detection methods.

[0019] The present invention adopts nonlinear topological structure analysis and manifold feature recognition technology of continuous infrared thermal image sequence and vibration acceleration signal, so that the positioning of defect area no longer relies solely on a single physical quantity or simple threshold judgment, which significantly improves the objectivity, accuracy and stability of defect area identification.

[0020] By real-time identification and precise classification of defect areas, the present invention can timely implement process parameter adjustments and manufacturing process control, inhibiting the formation and development of defects at the source, greatly improving the overall quality of stir friction additive manufacturing parts, and reducing the cost and difficulty of subsequent processing and testing. It has obvious economic benefits and engineering practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 any creative work.

[0022] Figure 1 Schematic diagram of the process of the present invention;

[0023] Figure 2 Schematic diagram of the structure of the system of the present invention. DETAILED DESCRIPTION

[0024] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] See also Figure 1 According to a first aspect of the present invention, a defect detection method for a friction stir additive manufacturing process is provided, comprising:

[0026] S101: Acquire a continuous infrared thermal image sequence of the friction stir additive manufacturing process, obtain the corresponding temperature gradient direction field based on each frame of the infrared thermal image; determine the initial thermal anomaly area based on the topological structure difference between the temperature gradient direction fields of two consecutive frames;

[0027] It should be noted that: first, an infrared thermal imager (such as the FLIR A655sc or FLIR X6900sc) installed above the working area of ​​the additive manufacturing equipment is used to collect a continuous infrared thermal image sequence of the manufacturing area in real time at a fixed sampling frequency (for example, 1 frame / second) during the operation of the equipment;

[0028] The infrared thermal image is a two-dimensional pixel matrix, where each pixel corresponds to the temperature value at a specific location during the friction stir additive manufacturing process.

[0029] For each frame of infrared thermal image collected, the temperature gradient direction field of the frame image is calculated based on the image processing method. The specific calculation method is as follows:

[0030] Taking each pixel of the current frame heat map as the calculation object, the Sobel operator is used to perform convolution calculation on the image to obtain the gradient components of the image in the horizontal and vertical directions:

[0031] G x =Sobel x *I,G y =Sobel y *I;

[0032] Where I is the heat map matrix of the current frame, Sobel x 、Sobel x They are the horizontal and vertical templates of the Sobel operator respectively;

[0033] Further calculate the temperature gradient direction θ of each pixel:

[0034]

[0035] At this point, each frame of infrared thermal image generates a temperature gradient direction field of the same size, where the value range of each pixel is [-π / 2,π / 2];

[0036] Wherein, determining the initial thermal anomaly area includes:

[0037] The temperature gradient direction field of two consecutive frames is divided into multiple spatial sub-regions, and the direction vector field of each sub-region is extracted;

[0038] Specifically, the temperature gradient direction field of two consecutive frames is processed using a grid division method, which includes:

[0039] The temperature gradient direction field of each frame is evenly divided into multiple spatial sub-regions. For example, an image with a size of 640×480 pixels is divided into 16×12 sub-regions, and each sub-region is 40×40 pixels in size.

[0040] For all pixels in each sub-region, the corresponding temperature gradient direction vector is extracted to form the local direction vector field of the sub-region for subsequent topological analysis;

[0041] Calculate the cosine of the angle between the central pixel of each sub-region and the direction vector of the adjacent pixels, and construct a local topological adjacency matrix;

[0042] The specific process of constructing the local topological adjacency matrix includes:

[0043] The central pixel of each sub-region is selected as a reference, and the cosine value of the temperature gradient angle between the central pixel and the surrounding 8 adjacent pixels (upper, lower, left, right and four diagonal directions) is calculated;

[0044] Specifically, the specific calculation formula is:

[0045]

[0046] Among them, V c Represents the direction vector of the center pixel, V n represents the direction vector of any pixel among the surrounding 8 adjacent pixels, and α is the angle between them;

[0047] If the cosine value of the angle is greater than a preset threshold (for example, the threshold is set to 0.85), a node connection relationship is established between the central pixel and the corresponding adjacent pixel, otherwise it is not established;

[0048] A 0-1 local topological adjacency matrix is ​​formed based on the node connection relationship, where 1 represents established connection and 0 represents no connection;

[0049] For example, the adjacency matrix constructed with the center pixel as the starting point is as follows:

[0050]

[0051] Compare the topological adjacency matrices of each pair of corresponding sub-regions in two consecutive frames to determine the node connection differences between the topological adjacency matrices of the corresponding sub-regions;

[0052] Specifically, the local topological adjacency matrices of each pair of corresponding spatial sub-regions in two consecutive frames are compared element by element to determine the difference in connectivity (0 or 1) in the corresponding elements. The total number of differences in node connectivity in each spatial sub-region is statistically obtained as a measure of the degree of topological structure change.

[0053] For example, if the adjacency matrix of a sub-region in the t-th frame is:

[0054]

[0055] The adjacency matrix of the sub-region corresponding to the t+1th frame is:

[0056]

[0057] Then the number of differences between the two is the total number of differences between the corresponding elements;

[0058] Count the number of node connection differences in each sub-region and cluster them using the DBSCAN clustering algorithm;

[0059] Specifically, the number of node connection differences in all sub-regions is input as a feature vector into the DBSCAN clustering algorithm for unsupervised clustering. DBSCAN clustering parameters such as the neighborhood radius (ε) can be selected as 0.5 times the mean of the number of differences, and the minimum number of samples in the neighborhood (MinPts) can be set to 5. Through cluster analysis, all sub-regions are automatically divided into multiple different clusters, and the sub-regions within each cluster have similar topological structure change characteristics.

[0060] The spatial sub-region corresponding to the cluster with the largest number of differences in the clustering results is determined as the initial thermal anomaly region;

[0061] S102: synchronously collecting vibration acceleration signals at multiple spatial locations during the friction stir additive manufacturing process; and determining an initial vibration abnormality region based on differences in multi-frequency coherence coefficient matrices of the vibration signals at each spatial location;

[0062] It should be noted that during the friction stir additive manufacturing process, multiple high-precision vibration sensors (such as the PCB 352C33 accelerometer) are installed at different spatial locations on the processing platform surface (for example, above the welding gun, on the workpiece fixture, on the side of the processing platform, etc.). Each sensor synchronously collects vibration acceleration signals at different locations during the manufacturing process. The sampling frequency is generally set to 10kHz to ensure that high-time-resolution vibration data is obtained. The vibration acceleration signals at all locations are simultaneously recorded using a synchronous data acquisition device (such as the NIPXIe-4499 high-speed data acquisition system) to ensure strict timing synchronization between the signals.

[0063] The determining of the initial vibration abnormal area includes:

[0064] Perform EEMD (Ensembled Empirical Mode Decomposition) on the vibration acceleration signal at each spatial position to obtain the intrinsic mode functions (IMFs) of multiple frequency bands. The specific implementation method is as follows:

[0065] Use EEMD decomposition tools (such as the EEMD algorithm package in Matlab) to decompose the vibration signal at each spatial position into a series of Intrinsic Mode Functions (IMFs);

[0066] The EEMD parameters are set as follows: the amplitude of the auxiliary white noise is 0.2 times the standard deviation of the original signal, and the number of iterations is set to 100 to ensure the robustness of the IMF component;

[0067] After decomposition, the vibration signal at each spatial position obtains about 6 to 8 IMF components. Each IMF component corresponds to the vibration characteristics in a different frequency range, and is arranged in order from high-frequency noise to low-frequency trends;

[0068] Calculate the coherence coefficient matrix between all IMF components at each spatial position and use the sum of the main diagonal of the matrix as the eigenvalue;

[0069] The specific process of calculating the coherence coefficient matrix between the IMF components includes:

[0070] Perform short-time Fourier transform on any two IMF components obtained by decomposing the vibration signal at each spatial position to obtain the spectrum distribution respectively. The specific steps are as follows:

[0071] STFT parameter settings: the window type is Hanning window, the window length is 256 sampling points, and the window overlap rate is 50%;

[0072] For each IMF component, the time-spectral power matrix is ​​calculated;

[0073] Calculate the cross-spectral density function between any two IMF component spectral distributions:

[0074] CPSD xy (f) = E[X(f)Y * (f)];

[0075] Where X(f) and Y(f) are the complex representations of the two IMF components in the frequency domain, * represents the complex conjugate, and E[·] represents the statistical expectation operation of multiple time windows;

[0076] Normalize the cross-spectral density function to obtain the standardized coherence coefficient value, and use it to construct a complete coherence coefficient matrix;

[0077] Specifically, based on the cross-spectral density function, the normalized coherence function is calculated:

[0078]

[0079] Where, PSD x (f) PSD y (f) are the autopower spectral density functions of the two IMF components;

[0080] The average value of the coherence function within the frequency range is then taken as the final normalized coherence coefficient between the two IMF components;

[0081] Repeat the above steps to construct a complete coherence coefficient matrix for all IMF component combinations. The coherence coefficient matrix is ​​a symmetric matrix with all diagonal elements being 1.

[0082] Finally, the eigenvalue of the coherence coefficient matrix is ​​the sum of the elements on the main diagonal, which is used to characterize the overall frequency domain correlation characteristics of the vibration signal at that spatial position;

[0083] The eigenvalues ​​of the coherence coefficient matrix at all spatial positions are used to form an eigenvector, and the Mahalanobis distance between the eigenvector and the reference eigenvector is calculated;

[0084] Specifically, based on the eigenvalues ​​of the coherence coefficient matrix at each spatial position, the initial abnormal area is further determined:

[0085] Arrange the eigenvalues ​​of the coherence coefficient matrix at all spatial positions in sequence to form eigenvectors;

[0086] Calculate the coherence matrix eigenvalue vectors of multiple sets of vibration signal data collected in normal working conditions (when there are no abnormalities in the manufacturing process) and take the average as the reference eigenvalue vector;

[0087] The Mahalanobis distance method is used to calculate the distance between each spatial position feature vector and the reference feature vector. The specific calculation formula is:

[0088]

[0089] Where X is the spatial position eigenvector to be measured, μ is the reference eigenvector, and Σ is the covariance matrix of the eigenvector under normal working conditions;

[0090] Select the spatial position corresponding to the area where the Mahalanobis distance exceeds the preset threshold as the initial vibration abnormal area;

[0091] It should be noted that the threshold of the Mahalanobis distance is set by the historical statistical method, and 1.5 to 2 times the standard deviation of the mean of the Mahalanobis distance calculated from the normal operating data is usually taken as the abnormality discrimination threshold; if the Mahalanobis distance calculated from the eigenvector of a certain spatial position exceeds the preset threshold, it is considered that the vibration characteristics of the spatial position are abnormal and it is determined to be an initial vibration abnormal area.

[0092] S103: Obtaining a thermal-vibration coupling topology map based on the spatial position topological correlation between the initial thermal abnormality region and the initial vibration abnormality region; and determining candidate defect regions based on the nonlinear manifold characteristics of the thermal-vibration coupling topology map.

[0093] The step of obtaining a thermal-vibration coupling topology map based on a spatial position topological correlation between the initial thermal anomaly region and the initial vibration anomaly region includes:

[0094] Extracting the geometric center coordinates of the initial thermal anomaly region and the initial vibration anomaly region, and calculating the two-dimensional spatial distance matrix between the centers of the thermal anomaly region and the vibration anomaly region;

[0095] Specifically, the geometric center coordinates are accurately extracted for the initial thermal anomaly region determined in step S101 and the initial vibration anomaly region determined in step S102:

[0096] For all spatial sub-regions of the initial thermal anomaly area, calculate the geometric center coordinates (x h ,y h ), which is calculated as follows:

[0097]

[0098] Where: (x h ,y h ) is the pixel coordinate center of the i-th sub-region in the initial thermal anomaly area, N h is the number of sub-regions contained in the initial thermal anomaly region;

[0099] Similarly, for the initial vibration abnormal area, the geometric center coordinates (x v ,y v );

[0100] After obtaining the geometric center coordinates of the thermal anomaly area and the vibration anomaly area, the two-dimensional Euclidean space distance matrix D between the two is further calculated:

[0101]

[0102] Where: D ij represents the spatial distance from the center of the i-th sub-region of the thermal anomaly area to the center of the j-th region of the vibration anomaly area;

[0103] The spectral clustering method is used to perform cluster analysis on the spatial distance matrix to determine the optimal topological correspondence between the thermal anomaly area and the vibration anomaly area;

[0104] The specific process of determining the optimal topological correspondence relationship includes:

[0105] Calculate the normalized Laplacian matrix of the spatial distance matrix;

[0106] Specifically, the similarity matrix W is constructed based on the distance matrix D, which is:

[0107]

[0108] In the formula, the parameter σ is generally taken as the mean or median of all distance values ​​to ensure that the W value has a reasonable numerical discrimination;

[0109] Then, calculate the degree matrix M, which is a diagonal matrix with each element being:

[0110]

[0111] Further calculate the normalized Laplace matrix L norm :

[0112]

[0113] Where: I is the unit matrix;

[0114] Perform eigenvalue decomposition on the Laplace matrix and select the eigenvectors corresponding to the second and third smallest eigenvalues ​​to form a low-dimensional embedding space;

[0115] Specifically, the above normalized Laplace matrix is ​​decomposed into eigenvalues. The specific implementation method is as follows: all eigenvalues ​​and eigenvectors of the matrix are calculated; the eigenvalues ​​are arranged in ascending order, and the eigenvectors corresponding to the second and third smallest eigenvalues ​​are selected to form a two-dimensional low-dimensional embedding space; through this operation, the original high-dimensional distance features are mapped to a low-dimensional space that is easy to cluster, so as to more clearly show the topological relationship characteristics between regions.

[0116] In the low-dimensional embedding space, nodes are clustered using the K-means algorithm to obtain the optimal node topology correspondence; the specific implementation method is:

[0117] In the two-dimensional embedding space, all regional center nodes are divided into several clusters using the K-means clustering algorithm;

[0118] K value selection method: Use the Elbow method to determine the optimal number of clusters, usually between 2 and 5, with the highest node compactness within the cluster as the standard;

[0119] After clustering is completed, the thermal anomaly areas and vibration anomaly areas that exist simultaneously in each cluster are paired with nodes with clear corresponding relationships, forming a clear topological correspondence;

[0120] The pairs of thermal anomaly areas and vibration anomaly areas with the best topological correspondence are used as nodes, and the corresponding relationships are used as connecting edges to construct a thermal-vibration coupling topology graph with clear nodes and edges.

[0121] After the optimal topological correspondence is determined, a thermal-vibration coupling topology diagram with clear nodes and connection edges is further constructed. The specific implementation method is:

[0122] Each thermal anomaly area and vibration anomaly area with a clear corresponding relationship form a pair of nodes, which are expressed as:

[0123] Node set: V = {(H1,V1),(H2,V2),…,(H n ,V n )}, where H i ,V i represent the i-th pair of nodes in the thermal anomaly region and the vibration anomaly region respectively;

[0124] On the basis of the clear topological correspondence, the correspondence between each pair of nodes is used as the connecting edge E of the graph, that is:

[0125] Edge set: E={e ij |e ij Representative node pair (H i ,V i )};

[0126] Finally, the graph structure formed by the node set V and the edge set E is defined as the thermal-vibration coupling topological graph G(V,E) to clearly represent the spatial topological relationship between the thermal and vibration abnormal areas.

[0127] It should be noted that after obtaining the above-mentioned thermal-vibration coupling topology map G(V, E), in order to identify potential candidate defect areas in the map, this embodiment first extracts nonlinear manifold features from the thermal-vibration coupling topology map. The specific process is as follows:

[0128] The thermal vibration coupling topology graph G(V,E) is converted into the adjacency matrix A. The matrix is ​​defined as follows:

[0129] If the node V i and V j There is a direct topological connection between them, then the adjacency matrix element A ij =1;

[0130] If the node V i and V j There is no direct topological connection between them, then the adjacency matrix element A ij =0;

[0131] Since the thermal-vibration coupling topology graph constructed in this embodiment is an undirected graph, the adjacency matrix A is a symmetric matrix;

[0132] Furthermore, in order to reveal the nonlinear relationship hidden in the thermal-vibration coupling topology, the adjacency matrix A is subjected to manifold learning dimensionality reduction. The specific method is as follows:

[0133] The ISOMAP manifold learning algorithm is used for dimensionality reduction. The specific implementation method is as follows:

[0134] Calculate the shortest path distance matrix G between graph nodes based on the adjacency matrix A dist , the distance is calculated by Dijkstra algorithm; for the shortest path distance matrix G dist Perform multidimensional scaling to reduce the dimension to a two-dimensional or three-dimensional embedding space; after ISOMAP processing, the coordinates of each node in the low-dimensional space are marked as Y i =(y i1 ,y i2 ,…,y id ), where d represents the embedding space dimension, which is usually 2 or 3;

[0135] After obtaining the low-dimensional spatial embedding representation of the nodes, the nonlinear manifold density characteristics of each node are further analyzed to identify potential defect areas in the thermal-vibration coupling topology. The specific implementation is as follows:

[0136] For the low-dimensional embedding space coordinates of each node, the Gaussian kernel density estimation method is used to calculate the local density ρ i , the specific formula is:

[0137]

[0138] Where, ‖Y i -Y j ‖ represents the Euclidean distance between node i and node j, N is the total number of nodes, σ is the bandwidth parameter of the density kernel function, and is usually the average distance between nodes;

[0139] Calculate the local density values ​​of all nodes in the thermal-vibration coupling topology respectively and form the local density eigenvector p = [ρ1,ρ2,…,ρ N ];

[0140] To further measure the local topological compactness of the nodes, the clustering coefficient C of each node is calculated. i , defined as follows:

[0141]

[0142] Where N i is the set of adjacent nodes of node i, E(N i ) represents N i The actual number of connected edges between nodes, k i Represents the number of adjacent nodes of node i; get the clustering coefficient feature vector C=[C1,C2,…,C N ];

[0143] The local density features and clustering coefficient features obtained by the above calculations are integrated to form the nonlinear manifold comprehensive feature vector F of the node i :

[0144] Fi =[ρ i ,C i ],i=1,2,…,N;

[0145] After obtaining the node nonlinear manifold comprehensive feature vector, the candidate defect area is further determined. The specific determination method is as follows: the nonlinear manifold feature vector set {F i} perform density clustering, and the specific implementation method is as follows: the DBSCAN algorithm parameters are set as follows: the neighborhood radius ∈ is automatically determined by the k-distance curve between the eigenvectors, and the 95% quantile distance in the eigenvector set is usually taken as the neighborhood radius; the minimum number of sample points in a cluster, MinPts, is set to 3 to 5 nodes; after the DBSCAN algorithm completes clustering, the cluster with a large number of nodes and local density and topological clustering coefficient characteristics significantly higher than the overall mean is identified as a candidate defect cluster; the corresponding area of ​​the nodes in the candidate defect cluster is traced back, and according to the node definition of the aforementioned topological map, each node corresponds to a clear spatial position of the initial thermal anomaly area and the vibration anomaly area; the initial thermal anomaly area and the initial vibration anomaly area corresponding to the candidate defect cluster node are merged to form a complete candidate defect area; finally, the candidate defect area with a clear spatial boundary is marked as the specific object for in-depth defect classification and verification in the next stage.

[0146] S104: For each location within the candidate defect area, a temperature-vibration joint topology map is constructed based on the corresponding temperature gradient direction field and the vibration signal coherence coefficient matrix; based on the nonlinear correlation between nodes in the temperature-vibration joint topology map, the actual defect area and the corresponding defect type are obtained;

[0147] The method of obtaining the real defect area and the corresponding defect type based on the nonlinear correlation between the nodes of the temperature-vibration joint topology graph includes:

[0148] Extract the temperature gradient directional field characteristics and vibration signal coherence coefficient matrix characteristics at each node position in the candidate defect area and combine them into a multidimensional feature vector;

[0149] Specifically, the temperature gradient direction field feature is as follows: with each node position in the candidate defect area as the center, a 5×5 local temperature gradient direction field is selected to calculate the angle distribution characteristics between the center position and the temperature gradient direction of the surrounding pixels; the statistics of the angle cosine, including the mean, variance and skewness, are extracted to form a 3D temperature gradient direction field feature vector; the vibration signal coherence coefficient matrix feature is as follows: the vibration signal of the spatial position corresponding to each node position in the candidate defect area is extracted, and the corresponding coherence coefficient matrix is ​​calculated; the matrix eigenvalues ​​(such as the first three eigenvalues) are extracted from the coherence coefficient matrix to form a 3D vibration feature vector; the multi-dimensional feature vector is constructed: the temperature gradient direction field feature vector is combined with the vibration signal coherence coefficient matrix feature vector to form a 6-dimensional comprehensive feature vector F for each node position. i ;

[0150] On this basis, all nodes in the candidate defect area are taken as topological map nodes, the comprehensive feature vector is taken as the node feature, and the connection relationship between nodes is constructed based on the spatial proximity relationship (such as the 8-neighborhood method) to clarify the node and connection relationship of the temperature-vibration joint topological map;

[0151] Calculate the nonlinear mutual information matrix of the multi-dimensional feature vectors between all nodes in the candidate defect area;

[0152] The specific process of calculating the nonlinear mutual information matrix of the multi-dimensional feature vectors between the candidate defect area nodes includes:

[0153] Perform Min-Max normalization on the multidimensional feature vector of each node. The specific formula is:

[0154]

[0155] Where, F min 、F max are the minimum and maximum values ​​of the corresponding feature dimensions of all nodes respectively;

[0156] For the normalized eigenvectors of any node pair, the kernel density estimation method is used to obtain the joint probability distribution function;

[0157] For the normalized eigenvectors of any node pair i, j, the kernel density estimation (KDE) method is used to calculate the joint probability distribution function The specific kernel function uses the Gaussian kernel function, and the bandwidth parameter σ of the Gaussian kernel selects the standard deviation of the distance between the eigenvectors; the joint probability density function calculation formula is:

[0158]

[0159] Where: N is the dimension of the feature vector (here 6), h is the kernel density bandwidth;

[0160] The nonlinear mutual information value of each node pair is calculated using the joint probability distribution function, and a complete nonlinear mutual information matrix is ​​formed;

[0161] According to the joint probability distribution function, the nonlinear mutual information value I(F i ; F j ), calculation formula:

[0162]

[0163] Where: is the probability density function of the single node feature vector, which is obtained by estimating the kernel density of the single node feature;

[0164] The nonlinear mutual information values ​​of all node pairs are combined into a nonlinear mutual information matrix MI, and the matrix elements MI ij =I(F i ; F j );

[0165] Perform Isomap manifold dimensionality reduction on the mutual information matrix to determine the nonlinear correlation between nodes;

[0166] The nonlinear mutual information matrix MI is used as the distance matrix, and the ISOMAP method is used to perform manifold dimensionality reduction on the nodes. The specific method is:

[0167] Calculate the shortest path distance between nodes based on the mutual information matrix and use the Dijkstra algorithm;

[0168] Perform multidimensional scaling (MDS) on the shortest path distance to reduce the dimension to two-dimensional space;

[0169] Obtain the manifold embedding coordinates of the nodes in the two-dimensional space, and use the Euclidean distance between the nodes in the two-dimensional space to represent the nonlinear correlation between the nodes. The closer the distance, the higher the correlation.

[0170] Based on the nonlinear correlation between nodes, a clustering algorithm is used to identify the node cluster with the most clustered nodes as the real defect area;

[0171] In the two-dimensional manifold space, a clustering algorithm is used to cluster nodes based on the nonlinear correlation between nodes to determine the true defect area:

[0172] The DBSCAN algorithm is used to cluster nodes. The DBSCAN algorithm parameter setting method is as follows: ∈ is selected as the 90th percentile of the two-dimensional spatial distance between nodes; MinPts (minimum number of nodes) is set to 4-6; the node cluster with the largest number of nodes in the cluster is determined as the true defect area cluster; based on the spatial position of the nodes in the true defect area cluster, the actual processing space is traced back to determine the specific spatial range of the true defect area;

[0173] Determine the corresponding defect type of the real defect area according to the maximum coherent component of the characteristics of each node in the real defect area cluster;

[0174] In order to determine the defect type of the real defect area, the principal component characteristics of the coherence coefficient matrix of the vibration signals corresponding to all nodes in the real defect area cluster are analyzed:

[0175] Perform principal component analysis (PCA) on the coherence coefficient matrix of the vibration signals of the nodes in the cluster;

[0176] Extract the eigenvalues ​​and eigenvectors corresponding to the principal components;

[0177] The frequency band feature corresponding to the principal component with the highest variance contribution rate (i.e., the maximum coherent component) is selected and determined as the typical defect feature of the real defect area;

[0178] Match the defect spectrum feature library in the historical defect database to determine the specific defect type, such as:

[0179] If the maximum coherent component frequency band is in the range of 100Hz to 200Hz, it is judged as an "internal pore defect";

[0180] If the maximum coherent component frequency band is in the range of 400Hz to 600Hz, it is judged as a "crack defect";

[0181] By analogy, the correspondence between other frequency bands and specific defect types is clarified, and ultimately, accurate defect identification in the stir friction additive manufacturing process is achieved.

[0182] See also Figure 2 Based on the same inventive concept, the second aspect of the present invention provides a defect detection system for use in a friction stir additive manufacturing process. For details not provided in this embodiment, please refer to the description of the relevant parts of Example 1. The system includes:

[0183] The first acquisition module 201 is used to acquire a continuous infrared thermal image sequence of the friction stir additive manufacturing process and calculate the corresponding temperature gradient direction field based on each frame of the infrared thermal image; and determine the initial thermal anomaly area based on the topological structure difference between the temperature gradient direction fields of two consecutive frames;

[0184] The second acquisition module 202 is used to synchronously acquire vibration acceleration signals at multiple spatial positions during the friction stir additive manufacturing process;

[0185] Based on the difference in the coherence coefficient matrix of the vibration acceleration signals at each spatial position, the initial vibration abnormality area is determined;

[0186] The screening module 203 is configured to obtain a thermal-vibration coupling topology map based on the spatial topological relationship between the initial thermal anomaly region and the initial vibration anomaly region; and determine candidate defect regions based on the nonlinear manifold characteristics of the thermal-vibration coupling topology map.

[0187] The detection module 204 is used to construct a temperature-vibration joint topology map for each position in the candidate defect area based on the corresponding temperature gradient direction field and the vibration signal coherence coefficient matrix; and obtain the real defect area and the corresponding defect type based on the nonlinear correlation between the nodes of the temperature-vibration joint topology map.

[0188] The above embodiments can be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, or a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0189] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only one type. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0190] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0191] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0192] Some of the data in the above formula are calculated by removing the dimensions and taking their numerical values. The formula is a formula that is closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0193] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A defect detection method for a friction stir additive manufacturing process, characterized in that: include: S101: collecting a continuous infrared thermal image sequence of the friction stir additive manufacturing process, and calculating the corresponding temperature gradient direction field based on each frame of the infrared thermal image; The initial thermal anomaly area is determined based on the topological structure difference of the temperature gradient direction field between two consecutive frames; S102: synchronously collecting vibration acceleration signals at multiple spatial locations during the friction stir additive manufacturing process; and determining an initial vibration abnormality region based on differences in multi-frequency coherence coefficient matrices of the vibration acceleration signals at each spatial location; S103: Obtaining a thermal-vibration coupling topology map based on a spatial position topological correlation relationship between the initial thermal anomaly region and the initial vibration anomaly region; Determine the candidate defect area based on the nonlinear manifold characteristics of the thermal-vibration coupling topology map; S104: For each position in the candidate defect area, a temperature-vibration joint topology map is constructed according to the corresponding temperature gradient direction field and the vibration signal coherence coefficient matrix; and the real defect area and the corresponding defect type are obtained according to the nonlinear correlation between the nodes of the temperature-vibration joint topology map.

2. A defect detection method for a friction stir additive manufacturing process according to claim 1, characterized in that: in, Determining the initial thermal anomaly area includes: The temperature gradient direction field of two consecutive frames is divided into multiple spatial sub-regions, and the direction vector field of each sub-region is extracted; Calculate the cosine of the angle between the central pixel of each sub-region and the direction vector of the adjacent pixels, and construct a local topological adjacency matrix; Compare the topological adjacency matrices of each pair of corresponding sub-regions in two consecutive frames to determine the node connection differences between the topological adjacency matrices of the corresponding sub-regions; Count the number of node connection differences in each sub-region and cluster them using the DBSCAN clustering algorithm; The spatial subregion corresponding to the cluster with the largest number of differences in the clustering results is determined as the initial thermal anomaly area.

3. A defect detection method for use in a friction stir additive manufacturing process according to claim 2, characterized in that: in, The specific process of constructing the local topology adjacency matrix includes: The central pixel of each sub-region is selected as a reference, and the cosine value of the temperature gradient angle between the central pixel and the surrounding 8 adjacent pixels is calculated; If the cosine value of the included angle is greater than a preset threshold, a node connection relationship is established between the central pixel and the corresponding adjacent pixel, otherwise it is not established; A 0-1 local topological adjacency matrix is ​​formed based on the node connection relationship, where 1 represents the establishment of a connection and 0 represents no connection.

4. A defect detection method for use in a friction stir additive manufacturing process according to claim 3, characterized in that: in, Determining the initial abnormal vibration area includes: Perform EEMD decomposition on the vibration acceleration signal at each spatial position to obtain the intrinsic mode functions of multiple frequency bands; Calculate the coherence coefficient matrix between all IMF components at each spatial position and use the sum of the main diagonal of the matrix as the eigenvalue; The eigenvalues ​​of the coherence coefficient matrix at all spatial positions are used to form an eigenvector, and the Mahalanobis distance between the eigenvector and the reference eigenvector is calculated; The area corresponding to the spatial position where the Mahalanobis distance exceeds a preset threshold is selected as the initial vibration abnormality area.

5. The defect detection method for friction stir additive manufacturing according to claim 4, characterized in that: in, The specific process of calculating the coherence coefficient matrix between the IMF components includes: Perform short-time Fourier transform on any two IMF components obtained by decomposing the vibration signal at each spatial position to obtain the spectrum distribution respectively; Calculate the cross-spectral density function between any two IMF component spectral distributions; The cross-spectral density function is normalized to obtain the standardized coherence coefficient value, and the complete coherence coefficient matrix is ​​constructed based on it.

6. A defect detection method for use in a friction stir additive manufacturing process according to claim 5, characterized in that: in, The step of obtaining a thermal-vibration coupling topology map based on a spatial position topological correlation between an initial thermal anomaly region and an initial vibration anomaly region includes: Extracting the geometric center coordinates of the initial thermal anomaly region and the initial vibration anomaly region, and calculating the two-dimensional spatial distance matrix between the centers of the thermal anomaly region and the vibration anomaly region; The spectral clustering method is used to perform cluster analysis on the spatial distance matrix to determine the optimal topological correspondence between the thermal anomaly area and the vibration anomaly area; The pairs of thermal anomaly areas and vibration anomaly areas with the best topological correspondence are taken as nodes, and the corresponding relationships are used as connecting edges to construct a thermal-vibration coupling topology graph with clear nodes and edges.

7. A defect detection method for use in a friction stir additive manufacturing process according to claim 6, characterized in that: in, The specific process of using the spectral clustering method to perform cluster analysis and determine the optimal topological correspondence includes: Calculate the normalized Laplacian matrix of the spatial distance matrix; Perform eigenvalue decomposition on the Laplace matrix and select the eigenvectors corresponding to the second and third smallest eigenvalues ​​to form a low-dimensional embedding space; In the low-dimensional embedding space, the nodes are clustered using the K-means algorithm to obtain the optimal node topology correspondence.

8. The defect detection method for friction stir additive manufacturing according to claim 7, characterized in that: in, The method of obtaining the real defect area and the corresponding defect type based on the nonlinear correlation between the nodes of the temperature-vibration joint topology graph includes: Extract the temperature gradient directional field characteristics and vibration signal coherence coefficient matrix characteristics at each node position in the candidate defect area and combine them into a multidimensional feature vector; Calculate the nonlinear mutual information matrix of the multi-dimensional feature vectors between all nodes in the candidate defect area; Perform Isomap manifold dimensionality reduction on the mutual information matrix to determine the nonlinear correlation between nodes; Based on the nonlinear correlation between nodes, a clustering algorithm is used to identify the node cluster with the most clustered nodes as the real defect area; According to the maximum coherent component of each node feature in the real defect area cluster, the corresponding defect type of the real defect area is determined.

9. The defect detection method for friction stir additive manufacturing according to claim 8, characterized in that: in, The specific process of calculating the nonlinear mutual information matrix of the multi-dimensional feature vectors between candidate defect area nodes includes: Perform Min-Max normalization on the multidimensional feature vector of each node; For the normalized eigenvectors of any node pair, the kernel density estimation method is used to obtain the joint probability distribution function; The joint probability distribution function is used to calculate the nonlinear mutual information value of each node pair and form a complete nonlinear mutual information matrix.

10. A defect detection system for a friction stir additive manufacturing process, implemented based on a defect detection method for a friction stir additive manufacturing process according to any one of claims 1 to 9, characterized in that: include: The first acquisition module is used to collect a continuous infrared thermal image sequence of the friction stir additive manufacturing process and calculate the corresponding temperature gradient direction field based on each frame of the infrared thermal image; The initial thermal anomaly area is determined based on the topological structure difference of the temperature gradient direction field between two consecutive frames; The second acquisition module is used to synchronously acquire vibration acceleration signals at multiple spatial positions during the friction stir additive manufacturing process; Based on the difference in the coherence coefficient matrix of the vibration acceleration signals at each spatial position, the initial vibration abnormality area is determined; A screening module is used to obtain a thermal-vibration coupling topology map based on a spatial position topological correlation relationship between an initial thermal anomaly region and an initial vibration anomaly region; Determine the candidate defect area based on the nonlinear manifold characteristics of the thermal-vibration coupling topology map; The detection module is used to construct a temperature-vibration joint topology map for each location in the candidate defect area based on the corresponding temperature gradient direction field and the vibration signal coherence coefficient matrix; According to the nonlinear correlation between nodes in the temperature-vibration joint topology graph, the real defect area and the corresponding defect type are obtained.

Images