Quality evaluation method and system for composite board

Abstract

The invention discloses a quality evaluation method and system for a composite board, and relates to the technical field of quality detection, and the method comprises the steps: constructing a latent variable feature space representing interface binding energy abnormity based on an initial response data field through a physical constraint relation between interface local response features and binding energy distribution; extracting a topological correlation structure between local latent variable features in the latent variable feature space through a graph convolutional network to obtain a local binding energy anomaly initial candidate region; active response mode enhanced excitation is carried out on the local area of the initial candidate area, the dynamic enhanced response of each local area is obtained, and a binding energy abnormal area is determined; determining the specific position and range of the hidden pseudo solder strip according to the correlation topological structure difference between the local feature in the binding energy abnormal region and the normal region; according to the method, precise identification and high-precision positioning of the titanium-steel composite plate interface hidden cold solder strip are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of quality detection, and in particular to a quality assessment method and system for composite panels. Background Art

[0002] Titanium-steel composite plates are widely used in aerospace, chemical equipment, shipbuilding and other fields. Their interfacial bonding performance has a decisive impact on the safety and durability of the overall structure. However, in actual production, the interfaces of titanium-steel composite plates often have bonding defects, such as cold welds and weak bonding areas. The presence of these defects not only reduces the mechanical properties of the material, but may also cause premature failure or even catastrophic damage during service.

[0003] Traditional interface inspection methods typically rely on mechanical shear tests and ultrasonic flaw detection. However, these methods suffer from limited accuracy, difficulty accurately locating hidden microscopic defects, time-consuming and labor-intensive processes, and a lack of real-time monitoring capabilities. Furthermore, existing numerical simulations and non-destructive testing techniques often fail to effectively integrate the deep connections between the mechanical response of interfaces and their physical properties, making it difficult to accurately reveal the true distribution of interface binding energy and abnormal regions at the microscale.

[0004] Therefore, there is an urgent need to propose a new method that can accurately identify and locate abnormal areas of interface bonding between titanium and steel composite plates, so as to address the shortcomings of existing detection methods and improve the accuracy and reliability of interface bonding status assessment. Summary of the Invention

[0005] The purpose of the present invention is to provide a quality assessment system and method for composite panels to solve the problems in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for evaluating the quality of a composite plate, comprising:

[0008] S101: applying a preset excitation field to the interface of the titanium-steel composite plate to obtain an initial response data field of the local interface;

[0009] S102: Based on the initial response data field, constructing a latent variable feature space characterizing the interface binding energy anomaly through the physical constraint relationship between the interface local response characteristics and the binding energy distribution;

[0010] S103: Extracting the topological correlation structure between the local latent variable features in the latent variable feature space through a graph convolutional network to obtain the initial candidate region of local binding energy anomaly;

[0011] S104: performing active response mode enhancement excitation on the local area of ​​the initial candidate region to obtain a dynamic enhanced response of each local area, and determining the abnormal binding energy area according to the dynamic enhanced response change rate;

[0012] S105: Determine the specific location and range of the hidden cold weld zone in the titanium-steel composite plate interface based on the difference in the associated topological structure between the local characteristics in the abnormal binding energy region and the normal region.

[0013] In a second aspect, the present invention provides a composite plate quality assessment system, which is implemented based on the composite plate quality assessment method described above, and includes:

[0014] The excitation acquisition module is used to apply a preset excitation field to the interface of the titanium-steel composite plate to obtain the initial response data field of the local interface;

[0015] A feature construction module is used to construct a latent variable feature space characterizing the interface binding energy anomaly based on the initial response data field and through the physical constraint relationship between the interface local response characteristics and the binding energy distribution;

[0016] The topology extraction module is used to extract the topological correlation structure between the local latent variable features in the latent variable feature space through the graph convolutional network, and obtain the initial candidate area of ​​local binding energy anomaly;

[0017] A dynamic determination module is used to perform active response mode enhancement excitation on the local area of ​​the initial candidate region to obtain a dynamic enhanced response of each local area, and determine the abnormal binding energy area according to the change rate of the dynamic enhanced response;

[0018] The cold weld positioning module is used to determine the specific location and range of the hidden cold weld zone in the titanium-steel composite plate interface based on the difference in the associated topological structure between the local characteristics in the abnormal binding energy area and the normal area.

[0019] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0020] The present invention adopts finite element numerical simulation and combines it with multi-physical field excitation methods, and uses genetic algorithms to optimize and select the most sensitive excitation type in the local interface, thereby constructing a high-precision initial data field that includes both displacement and stress responses. It effectively overcomes the defects of traditional detection methods that have single response information and insufficient accuracy, significantly improves the quality and reliability of the initial interface defect feature extraction, and lays a solid data foundation for the subsequent accurate identification of hidden interface defects.

[0021] By combining the variational autoencoder deep learning model with a clear physical energy function, the present invention constructs a latent variable feature space that can accurately reflect the abnormal characteristics of interface binding energy, and uses physical constraint regularization to achieve iterative optimization and precise correction of latent variable characteristics, effectively solving the problems of insufficient feature representation capabilities and poor physical interpretability of traditional machine learning, making the identification and characterization of interface microscopic binding defects more accurate and reliable.

[0022] The present invention accurately locates the specific position and range of the abnormal interface binding energy area through the topological correlation structure analysis of latent variable features based on graph convolutional networks, combined with the dynamic enhancement excitation and topological structure difference cross-validation method of the real-time active response mode, especially realizes the fine identification and spatial continuity definition of the continuous hidden weld zone area, which significantly improves the accuracy, real-time performance and engineering application value of the interface quality defect detection and monitoring of titanium steel composite plates. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0024] Figure 1 This is a flow chart of a quality assessment method for a composite board according to the present invention;

[0025] Figure 2 This is a framework diagram of a quality assessment system for composite panels according to the present invention. DETAILED DESCRIPTION

[0026] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be more comprehensive and complete, and will fully convey the concepts of the example embodiments to those skilled in the art. The accompanying drawings are merely schematic illustrations of the disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures indicate identical or similar parts, and thus any repetitive description thereof will be omitted.

[0027] In addition, the described features, structures or characteristics can be combined in one or more example embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the example embodiments disclosed in this application. However, those skilled in the art will appreciate that the technical solutions disclosed in this application can be practiced while omitting one or more of the specific details, or other methods, components, steps, etc. can be adopted. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring the various aspects disclosed in this application.

[0028] Example 1 like Figure 1 As shown, this embodiment discloses a method for evaluating the quality of a composite plate, comprising:

[0029] S101: applying a preset excitation field to the interface of the titanium-steel composite plate to obtain an initial response data field of the local interface;

[0030] In a specific implementation, the obtaining of the initial response data field of the local interface area includes:

[0031] Establishing an initial finite element simulation model of the titanium-steel composite plate according to the elastic modulus, density, and acoustic impedance distribution data of the titanium-steel composite plate;

[0032] Specifically, firstly, the titanium steel composite plate is sampled and tested to obtain the elastic modulus distribution. , density distribution , acoustic impedance distribution The specific value of forms a set of material parameters; among them, the elastic modulus The determination of density can be done by tensile test or ultrasonic measurement; Measured by weighing or density meter; acoustic impedance It is obtained by direct measurement through ultrasonic detector;

[0033] Subsequently, finite element analysis software (such as ABAQUS and ANSYS) was used to construct an initial finite element simulation model of the titanium-steel composite plate based on the geometric dimensions of the actual titanium-steel composite plate and the spatial distribution data obtained above. Specifically, the geometric model of the composite plate was meshed using a cell size of 1 mm × 1 mm × 0.5 mm as an example, and the obtained material parameters were assigned to each corresponding grid cell one by one to accurately reflect the mechanical properties of the actual titanium-steel interface.

[0034] For example, define any unit The material property set is:

[0035]

[0036] Where i, j, and k represent the spatial position index of the unit respectively.

[0037] Thermal field, ultrasonic stress wave, and electromagnetic field excitations were applied to the initial finite element simulation model. The most sensitive excitation type was determined by comparing the differences in displacement response distribution after the three excitations.

[0038] Specifically, determining the most sensitive stimulus type includes:

[0039] Taking the interface binding energy threshold of the normal bonding area and the virtual welding zone area as the constraint condition, the displacement sensitivity matrix of each local area under different excitation types is calculated respectively.

[0040] Specifically, we first define the interface binding energy threshold of the normal binding region , the interface bonding energy threshold of the virtual solder joint area is ; The threshold value can be measured by standard interfacial peeling or shearing test;

[0041] Then, in the simulation model, unit thermal field excitation (temperature rise T = 1 ° C, unit ultrasonic stress wave excitation (stress amplitude ), unit electromagnetic field excitation (magnetic field strength ), calculate and record the displacement response value of each unit under each excitation;

[0042] Furthermore, the displacement sensitivity matrix S is constructed, and the specific elements are defined as:

[0043]

[0044] in, represents the displacement response value of the i-th unit, represents the unit applied intensity of the jth excitation type;

[0045] It should be noted that the index i∈[1,N] represents the number of the i-th spatial grid unit in the composite plate, and N is the total number of units; the index j∈{1,2,3} represents the correspondence between the three different excitation types applied:

[0046] .

[0047] With the goal of maximizing the entropy of the local binding energy difference of titanium-steel composite plates, a genetic algorithm is used to optimize the displacement sensitivity matrix and determine the most sensitive excitation type.

[0048] Specifically, the interface binding energy difference entropy value H is defined as:

[0049]

[0050] in, is the ratio of the displacement sensitivity difference value of unit k to the sum of the sensitivity differences of all units, and is calculated as:

[0051]

[0052] in, is the displacement sensitivity value of the kth unit, The average of all unit displacement sensitivities;

[0053] Furthermore, the genetic algorithm is used to determine the incentive type combination with the largest entropy value:

[0054] (1) Represent each excitation type as a chromosome using binary coding;

[0055] (2) Randomly initialize the chromosome population;

[0056] (3) Calculate the entropy value corresponding to each chromosome as fitness;

[0057] (4) Select chromosomes with fitness greater than the preset value for crossover and mutation;

[0058] (5) Repeat the iteration until the entropy value converges and the optimal incentive type is determined.

[0059] Cross-feeding the sensitivity matrix corresponding to the determined most sensitive excitation type to the initial finite element simulation model, and updating and determining the most sensitive excitation type by re-correcting the boundary conditions of the simulation model;

[0060] Specifically, the displacement sensitivity matrix corresponding to the optimal excitation type determined in the previous step is used as a feedback basis. The boundary load conditions (including position, direction, and intensity) of the simulation model are adjusted according to the local areas with high displacement sensitivity, and the displacement response is simulated again.

[0061] It should be noted that: through several feedback iterations, the excitation boundary conditions of the simulation model were optimized and stabilized, and the most sensitive excitation type was finally determined, obtaining a high-quality simulation response;

[0062] According to the most sensitive excitation type determined, the initial displacement response of the local interface under this excitation is extracted, and the initial displacement response field is established;

[0063] Specifically, the most sensitive excitation type (for example, ultrasonic stress wave excitation) is determined with an excitation intensity that can be actually implemented. (For example ) is applied to the finite element simulation model, and the initial displacement response value of each interface local unit under this excitation is obtained through numerical simulation;

[0064] Assume that any local unit in the simulation model The initial displacement response is After integrating the displacement response data of all units, the initial displacement response field is formed ):

[0065]

[0066] It should be noted that the initial displacement response field is indexed by the unit center coordinates to ensure the spatial position consistency of subsequent data processing and analysis.

[0067] Inverting the initial stress response field using the material mechanics equilibrium equation according to the initial displacement response field;

[0068] Specifically, the elastic equilibrium differential equation in material mechanics is used:

[0069]

[0070] Under static or quasi-static conditions (neglecting inertia terms), the equation simplifies to:

[0071]

[0072] Where: is the stress tensor; For physical strength, it can usually be ignored; is the density; is the displacement field.

[0073] The displacement field Substitute the material constitutive equation (for example, the isotropic material constitutive equation):

[0074]

[0075] in: , is the Lame constant of the material, which can be determined by the elastic modulus E and Poisson's ratio v, specifically:

[0076]

[0077] Where: I is the unit tensor; is the displacement gradient tensor.

[0078] The initial displacement response field and the initial stress response field are merged to form an initial response data field that characterizes the local area of ​​the interface;

[0079] Specifically, the initial displacement response field is obtained The initial stress response field obtained in step S101.4 Performing one-to-one spatial fusion, the initial response data of each local unit of the interface can be expressed as:

[0080]

[0081] The response data set of all local units is further defined as the initial response data field :

[0082]

[0083] It should be noted that this initial response data field contains both displacement and stress mechanical characteristics, providing a more comprehensive, objective and reliable data basis for subsequent abnormal area detection.

[0084] S102: Based on the initial response data field, constructing a latent variable feature space characterizing the interface binding energy anomaly through the physical constraint relationship between the interface local response characteristics and the binding energy distribution;

[0085] In a specific implementation, the construction of a latent variable feature space characterizing the abnormal interface binding energy includes:

[0086] Taking the initial response data field of the local interface as input, the latent space features are extracted based on the variational autoencoder (VAE) model;

[0087] Specifically, this embodiment uses a variational autoencoder (VAE) model to extract a low-dimensional latent space feature vector that can effectively express the local response characteristics of the interface from the obtained initial response data field:

[0088] (1) Input data definition:

[0089] The initial response data field The displacement-stress vector of each local element As the input feature vector of the VAE model;

[0090] (2) VAE model structure:

[0091] Encoder network structure:

[0092] Input layer: input initial response feature vector ;

[0093] First hidden layer: 128 neurons, activation function ReLU;

[0094] Second hidden layer: 64 neurons, activation function ReLU;

[0095] Output layer: Generate latent variable mean vector and the standard deviation vector , the dimensions are all 20;

[0096] The specific calculation formula is as follows:

[0097]

[0098]

[0099]

[0100] Among them, W and b represent the weight and bias coefficient matrices of each layer.

[0101] Decoder network structure:

[0102] Input layer: input latent variable vector z;

[0103] First hidden layer: 64 neurons, activation function ReLU;

[0104] Second hidden layer: 128 neurons, activation function ReLU;

[0105] Output layer: reconstruct the initial response data vector ;

[0106] The specific calculation formula is as follows:

[0107]

[0108]

[0109]

[0110] Among them, the meaning of each layer parameter is the same as that of the encoder network;

[0111] (3) Loss function definition:

[0112] The loss function of VAE in this embodiment includes two parts: reconstruction error and KL divergence:

[0113]

[0114] Among them, KL divergence is used to constrain the latent variable distribution to be close to the standard normal distribution , to ensure that the generated latent variable feature space has good continuity and expressiveness.

[0115] By utilizing the physical constraint relationship of the interface binding energy anomaly, the latent space features are regularized with physical constraints to generate latent variable features constrained by the interface binding energy distribution.

[0116] Specifically, the physical constraint regularization of latent space features includes:

[0117] According to the intrinsic physical relationship between the material interface binding energy and the strain energy density, a physical energy function of the binding energy is established; Specific implementation method:

[0118] According to fracture mechanics theory, the interfacial bonding energy and local strain energy density There is a clear linear relationship between them:

[0119]

[0120] in, and is the material interface property constant, determined by interface shear test and numerical fitting;

[0121] Strain energy density It is defined as the volume integral of the stress-strain tensor product within the local element:

[0122]

[0123] in, is the stress tensor, is the strain tensor, and “:” represents the inner product of the tensor;

[0124] Specifically, the strain tensor By displacement field The spatial gradient of is derived as follows:

[0125]

[0126] Among them, the displacement gradient Defined as:

[0127] ;

[0128] Taking latent space features as input, the strain energy density distribution of each local area is calculated through the physical energy function;

[0129] It should be noted that the multilayer perceptron network (MLP) used in this step is used to map the latent space features extracted by VAE to the strain energy density distribution, so as to achieve an effective connection between the latent space and the actual physical features;

[0130] The specific implementation is as follows:

[0131] (1) Construct a multi-layer perceptron network structure for calculating strain energy density, including:

[0132] Input layer: The dimension is the same as the latent space feature vector z (for example, 20 dimensions);

[0133] Hidden layer: 128 neurons, activation function ReLU;

[0134] Output layer: Output strain energy density value , single neuron, no activation function (linear activation);

[0135] The specific calculation formula is:

[0136]

[0137]

[0138] 2) Use a large amount of experimentally tested sample data to supervise the training of the above MLP network, and each sample input latent space feature , the supervisory label is the true strain energy density value obtained by finite element calculation or experimental measurement ;

[0139] The specific training loss function is defined as:

[0140]

[0141] Where N represents the total number of samples used for training. The conventional gradient descent method (such as the Adam algorithm) is used in the optimization process until convergence.

[0142] Based on the binding energy threshold condition, the physical relationship between the strain energy density distribution and the initial response field of the interface is mapped;

[0143] It should be understood that the core purpose of this step is to clearly determine whether there may be abnormalities in the local interface (such as cold solder joints or weak bonding) through the binding energy threshold condition:

[0144] (1) Using the local strain energy density distribution obtained in the above steps And substitute the established physical energy function:

[0145]

[0146] (2) The predicted interface local binding energy value With the pre-set binding energy threshold A direct comparison is performed to determine the preliminary abnormal region candidate judgment conditions as follows:

[0147] If the predicted binding energy value of a local unit satisfies:

[0148]

[0149] Then the unit is preliminarily marked as an abnormal region candidate;

[0150] According to the physical relationship mapping error, the latent space features are iteratively corrected by optimizing the regularized loss function to obtain the latent variable features that meet the interface binding energy anomaly constraint;

[0151] The specific optimization process of this step includes:

[0152] (1) Define the total optimization loss function , which integrates the VAE reconstruction error ( ) and physical relationship mapping error ( ):

[0153]

[0154] in: It is the comprehensive loss of VAE network reconstruction error and KL divergence; Define the error between the predicted binding energy and the threshold, for example:

[0155]

[0156] Where M represents the number of candidate units in the abnormal area; λ is the regularization weight factor, which ranges from 0.1 to 1.0;

[0157] It should be noted that in order to ensure the balanced accuracy of prediction for abnormal and normal regions, a bidirectional penalty form is selected to improve the ability of latent space features to distinguish the two regions.

[0158] (2) Taking the latent space feature vector z as the optimization variable, use the gradient descent method (such as Adam optimizer) to optimize, and the optimization update rule is:

[0159]

[0160] in: is the learning rate, ranging from 0.001 to 0.005; the iteration termination condition is that the loss function converges stably or the maximum number of iterations (such as 50 to 100) is reached.

[0161] The local response data field of the interface is reversely reconstructed using the latent variable characteristics, the reconstruction error distribution is calculated, and cross-validation is performed through the difference with the initial response data field to correct the latent variable characteristics;

[0162] Specifically, the specific process of this step is as follows:

[0163] (1) The optimized latent variable features Input into the decoder network of VAE and reversely reconstruct the response data field:

[0164]

[0165] (2) Calculate the reconstruction error distribution and define the reconstruction error as the unit-by-unit Euclidean distance between the original response data field and the reconstructed data field:

[0166]

[0167] (3) Use cross-validation (e.g., 5-fold cross-validation) to check whether the reconstruction error is within the allowable range (typically, the error threshold is set to within 0.05). If the error exceeds the threshold, return to the above optimization steps for further optimization.

[0168] The modified latent variable features are used to form the final interface binding energy abnormal latent variable feature space; the specific implementation method is:

[0169] The latent variable features after optimization and cross-validation are Integrate into a complete latent variable feature space :

[0170]

[0171] Among them, each It is the final latent variable feature representation of the local unit of the interface, which is used for the next step of topological correlation analysis and accurate identification of abnormal areas.

[0172] S103: Extracting the topological correlation structure between the local latent variable features in the latent variable feature space through a graph convolutional network to obtain the initial candidate region of local binding energy anomaly;

[0173] In a specific implementation, the extraction of the topological correlation structure between local latent variable features in the latent variable feature space through the graph convolutional network includes:

[0174] According to the latent variable characteristics of each local interface in the latent variable feature space, the Euclidean distance and cosine similarity between nodes are determined to construct the initial association graph;

[0175] The specific implementation is as follows:

[0176] First, define the local latent variable feature space of the interface as: , where the latent variable characteristics of each local unit (node) of the interface are is the latent variable vector obtained in step S102, with a dimension of d=20;

[0177] The latent variable feature vector of any two nodes and , calculate their Euclidean distance respectively Similarity with cosine :

[0178] Among them, the Euclidean distance calculation formula is:

[0179]

[0180] Where: The latent variable feature vector of node i is a vector of dimension d=20, expressed as: ;in, represents the component of the kth dimension of the latent variable feature vector of node i;

[0181] Among them, the cosine similarity calculation formula is:

[0182]

[0183] The calculation result range is [-1, 1]. The closer the value is to 1, the more similar the directions of the feature vectors between nodes are.

[0184] To construct the initial association graph, the criteria for determining an association edge are defined as satisfying the following two threshold conditions simultaneously:

[0185] Euclidean distance Less than the preset threshold (For example );

[0186] Cosine similarity Greater than the preset threshold (For example );

[0187] When the characteristics of two nodes meet the above conditions at the same time, an associated edge is established between the nodes; otherwise, no edge is established; thus, the initial associated graph is obtained. , where the node set For all interface local nodes, edge sets is the set of associated edges determined according to the above judgment conditions.

[0188] Input the initial association graph into the graph convolutional network to extract the node feature topology structure;

[0189] It should be noted that the graph convolutional network (GCN) used in this embodiment has a clear network structure to fully extract the topological information of the association graph. The specific network structure is defined as follows:

[0190] (1) Graph Convolutional Network Structure:

[0191] The graph convolution network used in this embodiment is a three-layer graph convolution structure, and the propagation formula of each layer of graph convolution is:

[0192]

[0193] in: Indicates the Layer node feature matrix, layer 0 That is the initial node feature matrix, which is composed of the node's latent variable features constitute; is the network weight matrix of the lth layer, which needs to be trained; , is the adjacency matrix of the graph Add the identity matrix The results are based on the node's own characteristics; is the degree matrix, the diagonal elements ; As the activation function, this embodiment adopts the conventional ReLU function.

[0194] The specific three-layer GCN network structure is:

[0195] Input layer: input initial feature matrix , the dimension is the number of nodes N×d, where d=20;

[0196] First hidden layer: output feature dimension is 64;

[0197] Second hidden layer: output feature dimension is 32;

[0198] Output layer: The output feature dimension is 16, and the topological feature representation matrix of the node is finally extracted , dimension is N×16;

[0199] The input of the graph convolutional network after training is the adjacency matrix A of the initial association graph and the node features , the output is the node features enhanced by topological structure features .

[0200] Specifically, the method for obtaining the graph convolutional network includes:

[0201] Collect a large number of titanium-steel composite plate sample data with known interface bonding states, and obtain the latent variable characteristics of the sample interface and the corresponding true distribution of bonding energy;

[0202] Specifically, a plurality of titanium-steel composite plate samples with known interface bonding states are collected. The interface latent variable characteristics and true bonding energy state distribution of each sample are extracted using the method disclosed in steps S101-S102. The interface local regions are clearly labeled to obtain a training data set:

[0203] Sample latent variable feature matrix: ;

[0204] Abnormal label of the sample's true binding energy: ;

[0205] Where the label is defined as:

[0206] ;

[0207] Construct a supervised loss function for binding energy anomaly judgment, using the true binding energy anomaly distribution of the sample as the supervised label;

[0208] Specifically, the training goal is to enable the node features output by the network to accurately determine whether the local binding energy of the interface is abnormal or not. The loss function uses the conventional binary cross entropy loss:

[0209]

[0210] in: Output features based on GCN The result of binary classification prediction (after Sigmoid activation).

[0211] Train the graph convolutional network until the judgment error is less than the predetermined loss threshold, and perform two-way cross-validation using training data and simulation data alternately;

[0212] In the implementation, training data and simulation data are alternately used for two-way cross-validation, including:

[0213] The training dataset is randomly divided into k mutually exclusive subsets (in this example, k = 5). In each cross-validation iteration, one of the subsets is selected as the validation set, and the remaining k-1 subsets are used to train the model. This process is repeated k times to ensure that each subset is used as a validation set at least once.

[0214] The goal of the cross-validation process is to ensure that the model's prediction accuracy is higher than a preset threshold (for example, accuracy > 95%) and the judgment error (loss function) is lower than a predetermined loss threshold (for example, loss < 0.05), ultimately obtaining a graph convolutional network model with stable performance.

[0215] Dynamically adjust the training data ratio based on the cross-validation results, and iteratively optimize the network parameters to ultimately form a graph convolutional network with generalization capabilities;

[0216] In order to further improve the generalization performance of the network model, the ratio of positive and negative samples used for training is dynamically adjusted according to the results of each cross-validation. The specific method is as follows:

[0217] (1) If the cross-validation results show that the model's accuracy in determining abnormal regions (positive samples) is lower than that in determining normal regions (negative samples), indicating insufficient positive sample training, the next round of training should increase the proportion of abnormal region samples (e.g., adjust the positive-to-negative sample ratio from the initial 1:1 to 1.5:1 or 2:1) and retrain;

[0218] (2) If the judgment accuracy is relatively balanced but the overall accuracy does not reach the preset target, the total number of training samples should be appropriately increased or the learning rate should be adjusted (e.g., from 0.005 to 0.001) until the performance meets the predetermined requirements.

[0219] Reversely map the node feature topology structure to the latent variable feature space and calculate the constraint strength of the topological feature on the latent variable space;

[0220] In the implementation, the constraint strength of the topological features on the latent variable space is calculated, including:

[0221] Output the feature matrix of the node from the trained graph convolutional network In , extract the topological feature vector of each node (dimension 16);

[0222] The constraint strength of each node on other nodes in the latent variable feature space is calculated based on the topological feature vector. The specific constraint strength is defined as the cosine similarity between topological features:

[0223]

[0224] Where: It represents the topological feature constraint strength of node i on node j. By calculating the constraint strength, the degree of topological association between nodes in the latent variable feature space can be quantitatively expressed.

[0225] Based on the constraint strength difference feedback, the initial correlation graph is modified to form a final correlation graph with a topological feedback structure;

[0226] It should be noted that the goal of this step is to calculate the topological constraint strength matrix The initial correlation graph is iteratively modified to further optimize the graph structure so that abnormal interface binding energy regions can be identified more accurately in the future. The specific implementation method is as follows:

[0227] 1) Define topology feedback correction rules:

[0228] First, set the threshold of topological constraint strength (determined by experiment), the initial correlation graph Check each connected edge in:

[0229] If the topological constraint strength between any two nodes i and j Less than threshold , then remove the edge connecting nodes i and j from the association graph; otherwise, if the topological constraint strength is greater than or equal to the threshold, then retain the edge connection;

[0230] (2) Using the above correction rules, the initial correlation graph Perform a complete edge correction to obtain the intermediate correlation graph ;

[0231] (3) Further feedback iteration:

[0232] The corrected intermediate correlation graph Re-enter the graph convolutional network and perform the topological feature extraction and constraint strength calculation process again to generate a new constraint strength matrix ;

[0233] (4) Compare the difference between the new and old constraint strength matrices (for example, calculate the difference by the norm of the matrix difference). If the difference between the new and old constraint strength matrices is less than a preset threshold (for example, the difference norm is less than 0.01), the graph structure is considered to be stable. Otherwise, repeat the above steps with the new constraint strength matrix until the association graph is stable, and obtain a stable topological feedback association graph. .

[0234] According to the topological feedback characteristics of the final correlation graph, the initial candidate region of local binding energy anomaly is determined;

[0235] It should be noted that the goal of this step is to feedback the association graph based on the final topology formed Accurately identify areas where abnormal bonding energy may exist (such as cold solder joints or weak bonds). The specific implementation method is as follows:

[0236] (1) Based on the final association graph , extract connected subgraphs or connected regions in the graph;

[0237] (2) Calculate the mean topological constraint strength of the nodes in each connected area :

[0238]

[0239] in, Represents the total number of connecting edges between nodes in the region;

[0240] (3) Setting the mean threshold of topological constraint strength (For example, a value of 0.75) is used to further screen the initial candidate regions with potential binding energy anomalies:

[0241] If the mean topological constraint strength of a region is Less than threshold , indicating that the constraint relationship between nodes in this area is weak and the topological structure is abnormal. It is preliminarily determined that there may be abnormal binding energy problems in this area;

[0242] On the contrary, if the mean topological constraint strength of the region is greater than or equal to the threshold , then the topological structure of the region is considered to be stable and there is no obvious binding energy anomaly;

[0243] (4) All connected regions that meet the initial judgment criteria for binding energy anomaly are defined as initial candidate regions for local binding energy anomaly. , expressed as: .

[0244] S104: performing active response mode enhancement excitation on the local area of ​​the initial candidate region to obtain a dynamic enhanced response of each local area, and determining the abnormal binding energy area according to the dynamic enhanced response change rate;

[0245] In a specific implementation, determining the abnormal binding energy region according to the dynamic enhancement response change rate includes:

[0246] Based on the differences in the characteristics of local latent variables in the candidate areas, the spatial application strategy of dynamic enhancement incentives is determined;

[0247] Specifically, first, the initial candidate region of abnormal binding energy obtained in step S103 is In the example, extract the latent variable feature vector corresponding to each local node (The dimension is 20), calculate the variance or dispersion of the latent variable feature vector in each candidate area to measure the degree of difference in the latent variable characteristics;

[0248] The latent variable feature difference measure of the mth candidate area is defined as the variance mean of the feature vector set:

[0249]

[0250] Where, is the number of local nodes in the mth candidate area, is the average value of the node latent variable feature vector in the mth candidate area.

[0251] Further, according to the feature difference metric , determine the priority of dynamic enhancement incentives for each candidate area in descending order; the specific strategy is as follows:

[0252] Regions with large feature differences indicate significant inhomogeneity in the local feature distribution and are more likely to have abnormal binding energy, so dynamic enhancement excitation is preferentially applied.

[0253] The spatial strategy for applying incentives is: the area with the greatest difference is stimulated with the highest intensity, the area with the second greatest difference is stimulated with a relatively lower intensity, and so on;

[0254] The incentive intensity is distributed in a ratio manner based on a preset threshold value. For example, the incentive intensity of the largest difference area is 100% of the baseline value, the second largest difference area is 80%, and so on.

[0255] Real-time acquisition of the local dynamic displacement response of the interface under active enhanced excitation;

[0256] In specific implementation: Based on the above-determined spatial application strategy, the candidate area of ​​the titanium-steel composite plate interface is subjected to enhanced excitation of the active response mode, for example, by adopting the ultrasonic stress wave dynamic excitation method. The excitation intensity and frequency can be selected according to the actual engineering implementation range, for example, the intensity range is 5MPa-10MPa, and the frequency range is 50kHz-200kHz;

[0257] Laser Doppler vibrometer (LDV) or digital image correlation (DIC) technology is used to collect the dynamic displacement response signals of each local node of the interface in real time. ;

[0258] Specific operations include:

[0259] precisely aligning the measuring device to the determined candidate area;

[0260] During the duration of the excitation, the dynamic displacement response waveform of each local node is recorded;

[0261] The sampling frequency is set to more than 5 times the dynamic excitation frequency to ensure the accuracy of signal capture;

[0262] The data obtained in the above process constitutes the dynamic enhanced response data set of each local node: .

[0263] The dynamic stress distribution field is derived from the dynamic displacement response of the interface, and the dynamic stress change rate is calculated;

[0264] In specific implementation, this step aims to further calculate the local dynamic stress distribution field of the interface based on the obtained dynamic displacement response data, specifically including:

[0265] (1) First, the dynamic displacement response collected in real time from the local interface is obtained by using the material mechanical equilibrium relationship and material constitutive relationship. The dynamic strain field is obtained by inversion:

[0266] It should be understood that: considering the possible noise influence in actual measurement, the dynamic displacement response of the acquisition should be Perform digital filtering, such as using a Butterworth low-pass filter or Savitzky-Golay smoothing filter, to obtain smoother and more accurate displacement gradient data, thereby ensuring the reliability of dynamic stress inversion calculations;

[0267] According to the linear elastic constitutive relation, the relationship between strain and displacement gradient is:

[0268] strain tensor is the symmetric part of the displacement gradient tensor, that is:

[0269]

[0270] (2) Then the dynamic stress field is obtained based on the Hooke's law material constitutive relation:

[0271] According to the conventional constitutive relation of linear elastic materials, the stress tensor and the strain tensor The relationship is:

[0272]

[0273] (3) The dynamic stress response field corresponding to each local node of the interface is obtained , specifically expressed as: ;

[0274] The dynamic stress change rate is further calculated, which is defined as the relative change of stress at each local node over time. Specifically, for any node i, the dynamic stress change rate is defined as: ;

[0275] Get the dynamic stress change rate field of all nodes: .

[0276] According to the degree of deviation between the dynamic stress change rate and the latent variable characteristics, combined with the local topological structure characteristics of the candidate area, the actual binding energy abnormal area is determined;

[0277] In specific implementation, the goal of this step is to accurately identify areas with abnormal bonding energy (such as cold welds, weak bonds, etc.) at the titanium-steel interface by integrating dynamic stress change rate characteristics, latent variable characteristics, and topological structure characteristics. The details are disclosed as follows:

[0278] (1) First, clarify the correlation between the dynamic stress change rate and the latent variable characteristics:

[0279] Specifically, through multiple linear regression or canonical correlation analysis, the statistical relationship between the dynamic stress change rate characteristics of each node in the candidate area and the corresponding node latent variable characteristics is established, which can be specifically expressed as:

[0280] Define the prediction mapping function between dynamic stress change rate and latent variable characteristics , whose input is the node latent variable feature vector , the output is the predicted dynamic stress change ;

[0281] To implement the above mapping function, this embodiment specifically adopts a conventional multi-layer perceptron (MLP) model:

[0282] Input layer: latent variable feature vector of input node (20 dimensions).

[0283] Hidden layer: 2 hidden layers, each layer contains 64 neurons, and the activation function is ReLU.

[0284] Output layer: Output predicted dynamic stress change rate .

[0285] The training process of the model is as follows:

[0286] Training data: The actual value of the dynamic stress change rate of the candidate area node is obtained As the supervision label, the latent variable feature vector As input;

[0287] Training objective: Define the mean square error (MSE) loss function, specifically:

[0288]

[0289] Where N is the total number of training samples. The training process is implemented using the existing Adam optimization algorithm until the loss function converges;

[0290] (2) Calculate the correlation deviation degree of dynamic stress change rate:

[0291] According to the trained prediction model, the predicted dynamic stress change rate is calculated for each node , and then compared with the actual measured value Compare and define the degree of deviation from the association (prediction residuals) ;

[0292] The degree of correlation deviation quantifies the deviation of the latent variable characteristics from accurately representing the dynamic stress response, that is, it represents the degree to which the local node may be abnormal;

[0293] (3) Combine the topological structure characteristics of the candidate area to conduct fine screening of abnormal areas:

[0294] In this step, the obtained topological association structure is further utilized. The specific method is as follows:

[0295] First, calculate the average value of the association deviation of all nodes in each candidate area (connected area):

[0296]

[0297] Furthermore, considering the average value of topological constraint strength in the region Define comprehensive indicators for regional anomaly determination ,in, is the weight factor (e.g. 0.7);

[0298] Determine the actual binding energy anomaly region:

[0299] Setting comprehensive indicators Abnormal area judgment threshold (For example, the value is 0.3, which is determined through experiments); when the comprehensive index of the candidate area exceeds the threshold, that is:

[0300] like , then the region is clearly determined to be an abnormal region of actual binding energy;

[0301] like , then the area is considered to be a normal area or an area with insignificant abnormality;

[0302] Finally, the actual abnormal binding energy region is clearly expressed as: .

[0303] S105: Determine the specific location and range of the hidden cold weld zone at the titanium-steel composite plate interface based on the correlation topological structure difference between the local characteristics in the abnormal binding energy region and the normal region;

[0304] In a specific implementation, the specific position and range of the hidden cold weld zone in the titanium-steel composite plate interface are determined, including:

[0305] Extract the topological structure adjacency matrix of the abnormal binding energy region and the normal region;

[0306] Specifically, this step aims to clearly distinguish the topological differences between the local features of the abnormal binding energy region and the normal region (the rest of the region):

[0307] (1) According to the final correlation graph obtained , the node index sets belonging to the abnormal area and the normal area in the node set V are defined as:

[0308] Abnormal area node collection: ;

[0309] Normal area node set: ;

[0310] (2) Based on the above node set, the adjacency matrix of the abnormal area and normal area nodes is extracted respectively:

[0311] Taking the abnormal area as an example, the abnormal area adjacency matrix is ​​defined as:

[0312]

[0313] Similarly, the normal region adjacency matrix is ​​defined as:

[0314]

[0315] Among them, the adjacency matrix represents the topological connection relationship between the nodes in two regions.

[0316] Calculate the difference tensor of the topological feature matrices of the two regions;

[0317] Specifically, this step uses a graph convolutional network (GCN) to extract topological features from the adjacency matrix and clarify the difference between the topological features of the two regions:

[0318] (1) Adjacency matrix of abnormal regions With the normal region adjacency matrix Perform graph convolution operations to extract topological features of nodes in the corresponding area:

[0319] Specifically, the graph convolutional network structure that has been clearly disclosed in step S103 is used for feature extraction to obtain:

[0320] Abnormal area topological feature matrix: ;

[0321] Normal region topological feature matrix: ;

[0322] Among them, the output feature dimension ;

[0323] (2) Calculate the difference tensor:

[0324] Defining the difference tensor The statistical difference in node feature distribution between abnormal and normal areas is constructed using the mean difference in topological features between the two areas:

[0325]

[0326] in:

[0327] Mean value of topological characteristics of abnormal area: ;

[0328] Mean value of topological characteristics of normal area: .

[0329] Based on the cross-validation results of the difference tensor spatial distribution and the dynamic enhancement response, the boundary between the abnormal and normal areas of the interface binding energy is determined;

[0330] Specifically, this step aims to define the precise boundary between the abnormal area and the normal area. The specific implementation method is as follows:

[0331] (1) Using the obtained difference tensor , perform topological feature difference mapping on all nodes of the interface, that is, perform inner product calculation on the topological feature vector of each node and the difference tensor to obtain the topological difference mapping value of the node:

[0332] For any node feature :

[0333]

[0334] Among them, the node topology difference mapping value The larger the absolute value of , the more the topological structure of the node deviates from the normal area;

[0335] (2) Correlation deviation between the node topology difference mapping value and the dynamic stress change rate obtained in step S104 Perform cross validation:

[0336] Specific implementation includes:

[0337] Define cross-validation metrics for each node , comprehensively considering topological difference mapping and dynamic enhancement response differences:

[0338]

[0339] The weight factor 𝛽 typically ranges from 0.5 to 0.7, and in this embodiment, the typical value is 0.6;

[0340] Determine the topological difference cross-validation threshold for the interface region For example, the value is 0.2 (the specific value can be determined through experiments);

[0341] According to the above indicators Clearly define the interface boundary between abnormal area and normal area:

[0342] If the node index , the node clearly belongs to the abnormal area;

[0343] If the node index , the node belongs to the normal area;

[0344] Clearly define the node set between the abnormal and normal areas as the boundary node set: .

[0345] Based on the boundary continuity constraint rule, the specific location and range of the continuous hidden weld zone are precisely determined through spatial cluster analysis.

[0346] Specifically, the goal of this step is to clarify the spatial position and range of the continuous virtual welding strip. The specific implementation method is:

[0347] (1) Clarify the boundary continuity constraint rules:

[0348] If the nodes are adjacent in spatial position (e.g., the grid cells are adjacent or the distance is less than a set spatial threshold, such as 1 mm), then these nodes are considered to have spatial continuity;

[0349] (2) Use spatial clustering analysis algorithm to analyze the boundary node set Perform cluster analysis, such as DBSCAN or K-Means clustering algorithm:

[0350] Taking DBSCAN as an example: the neighborhood radius ϵ is typically set to 1 mm; the minimum number of nodes MinPts is typically set to 3-5; cluster analysis is performed on the boundary node set to obtain several clustering results;

[0351] (3) Clearly define the specific location and scope of the continuous virtual welding strip:

[0352] For each cluster subset, specify the node coordinate set ;

[0353] According to the spatial coordinates of the nodes in the cluster subset, the spatial position, length and width range of the virtual weld band are clearly determined;

[0354] Finally, an accurate description of the hidden weld zone is obtained, such as:

[0355]

[0356] Thus, the specific spatial position and continuous range of the hidden cold solder joint are clearly given.

[0357] Example 2 like Figure 2 As shown, the parts not described in detail in this embodiment are as shown in Example 1. This embodiment discloses a quality assessment system for composite panels, including:

[0358] The excitation acquisition module 201 is used to apply a preset excitation field to the interface of the titanium-steel composite plate to obtain the initial response data field of the local interface;

[0359] A feature construction module 202 is configured to construct a latent variable feature space representing anomalies in interface binding energy based on the initial response data field and the physical constraint relationship between the local response characteristics of the interface and the binding energy distribution;

[0360] A topology extraction module 203 is used to extract the topological correlation structure between the local latent variable features in the latent variable feature space through a graph convolutional network to obtain an initial candidate region of local binding energy anomaly;

[0361] The dynamic determination module 204 is configured to perform active response mode enhancement excitation on the local area of ​​the initial candidate region to obtain a dynamic enhanced response of each local area, and determine the abnormal binding energy area based on the change rate of the dynamic enhanced response;

[0362] The cold weld positioning module 205 is used to determine the specific position and range of the hidden cold weld zone in the titanium-steel composite plate interface based on the difference in the associated topological structure between the local characteristics in the abnormal binding energy area and the normal area.

[0363] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters, weights and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0364] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. 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 comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are fully or partially performed. 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 transferred from one computer-readable storage medium to another. For example, the computer instructions can be transferred 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 accessible 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 magnetic media (e.g., floppy disks, hard disks, or magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0365] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A quality assessment method for a composite board, characterized in that: include: S101: applying a preset excitation field to the interface of the titanium-steel composite plate to obtain an initial response data field of the local interface; S102: Based on the initial response data field, constructing a latent variable feature space characterizing the interface binding energy anomaly through the physical constraint relationship between the interface local response characteristics and the binding energy distribution; S103: Extracting the topological correlation structure between the local latent variable features in the latent variable feature space through a graph convolutional network to obtain the initial candidate region of local binding energy anomaly; S104: performing active response mode enhancement excitation on the local area of ​​the initial candidate region to obtain a dynamic enhanced response of each local area, and determining the abnormal binding energy area according to the dynamic enhanced response change rate; S105: Determine the specific location and range of the hidden cold weld zone in the titanium-steel composite plate interface based on the difference in the associated topological structure between the local characteristics in the abnormal binding energy region and the normal region.

2. The quality assessment method of the composite board according to claim 1, characterized in that: The obtaining of the initial response data field of the local interface area includes: Establishing an initial finite element simulation model of the titanium-steel composite plate according to the elastic modulus, density, and acoustic impedance distribution data of the titanium-steel composite plate; Thermal field, ultrasonic stress wave, and electromagnetic field excitations were applied to the initial finite element simulation model. The most sensitive excitation type was determined by comparing the differences in displacement response distribution after the three excitations. According to the most sensitive excitation type determined, the initial displacement response of the local interface under this excitation is extracted, and the initial displacement response field is established; Inverting the initial stress response field using the material mechanics equilibrium equation according to the initial displacement response field; The initial displacement response field and the initial stress response field are fused to form an initial response data field that characterizes the local area of ​​the interface.

3. The quality assessment method of the composite board according to claim 2, characterized in that: The determination of the most sensitive stimulus type includes: Taking the interface binding energy threshold of the normal bonding area and the virtual welding zone area as the constraint condition, the displacement sensitivity matrix of each local area under different excitation types is calculated respectively. With the goal of maximizing the entropy of the local binding energy difference of titanium-steel composite plates, a genetic algorithm is used to optimize the displacement sensitivity matrix and determine the most sensitive excitation type. The sensitivity matrix corresponding to the determined most sensitive excitation type is cross-fed back to the initial finite element simulation model, and the boundary conditions of the simulation model are re-corrected to update and determine the most sensitive excitation type.

4. The quality assessment method of the composite plate according to claim 3, characterized in that: The construction of the latent variable feature space characterizing the abnormal interface binding energy includes: Taking the initial response data field of the local interface as input, the latent space features are extracted based on the variational autoencoder model; By utilizing the physical constraint relationship of the interface binding energy anomaly, the latent space features are regularized with physical constraints to generate latent variable features constrained by the interface binding energy distribution. The local response data field of the interface is reversely reconstructed using the latent variable characteristics, the reconstruction error distribution is calculated, and cross-validation is performed through the difference with the initial response data field to correct the latent variable characteristics; The corrected latent variable features are used to form the final interface binding energy abnormal latent variable feature space.

5. The quality assessment method of the composite board according to claim 4, characterized in that: The physical constraint regularization of latent space features includes: According to the intrinsic physical relationship between the material interface binding energy and the strain energy density, a physical energy function of the binding energy is established; Specific implementation method: Taking latent space features as input, the strain energy density distribution of each local area is calculated through the physical energy function; Based on the binding energy threshold condition, the physical relationship between the strain energy density distribution and the initial response field of the interface is mapped; According to the physical relationship mapping error, the latent space features are iteratively corrected by optimizing the regularization loss function to obtain latent variable features that meet the interface binding energy anomaly constraints.

6. The quality assessment method of the composite board according to claim 5, characterized in that: The method of extracting the topological correlation structure between local latent variable features in the latent variable feature space through the graph convolutional network includes: According to the latent variable characteristics of each local interface in the latent variable feature space, the Euclidean distance and cosine similarity between nodes are determined to construct the initial association graph; Input the initial association graph into the graph convolutional network to extract the node feature topology structure; Reversely map the node feature topology structure to the latent variable feature space and calculate the constraint strength of the topological feature on the latent variable space; Based on the constraint strength difference feedback, the initial correlation graph is modified to form a final correlation graph with a topological feedback structure; Based on the topological feedback characteristics of the final correlation graph, the initial candidate regions of local binding energy anomaly are determined.

7. The quality assessment method of the composite plate according to claim 6, characterized in that: The method for obtaining the graph convolutional network includes: Collect a large number of titanium-steel composite plate sample data with known interface bonding states, and obtain the latent variable characteristics of the sample interface and the corresponding true distribution of bonding energy; Construct a supervised loss function for binding energy anomaly judgment, using the true binding energy anomaly distribution of the sample as the supervised label; Train the graph convolutional network until the judgment error is less than the predetermined loss threshold, and perform two-way cross-validation using training data and simulation data alternately; The training data ratio is dynamically adjusted according to the cross-validation results, and the network parameters are re-iteratively optimized to eventually form a graph convolutional network with generalization capabilities.

8. The quality assessment method of the composite board according to claim 7, characterized in that: Determining the abnormal binding energy region according to the dynamic enhancement response change rate includes: Based on the differences in the characteristics of local latent variables in the candidate areas, the spatial application strategy of dynamic enhancement incentives is determined; Real-time acquisition of the local dynamic displacement response of the interface under active enhanced excitation; The dynamic stress distribution field is derived from the dynamic displacement response of the interface, and the dynamic stress change rate is calculated; According to the degree of deviation between the correlation between the dynamic stress change rate and the latent variable characteristics, combined with the local topological structure characteristics of the candidate area, the actual binding energy abnormal area is determined.

9. The quality assessment method of the composite plate according to claim 8, characterized in that: Determining the specific location and range of the hidden cold weld zone in the titanium-steel composite plate interface includes: Extract the topological structure adjacency matrix of the abnormal binding energy region and the normal region; Calculate the difference tensor of the topological feature matrices of the two regions; Based on the cross-validation results of the difference tensor spatial distribution and the dynamic enhancement response, the boundary between the abnormal and normal areas of the interface binding energy is determined; Based on the boundary continuity constraint rule, the specific location and range of the continuous hidden weld zone are precisely determined through spatial cluster analysis.

10. A composite board quality assessment system, implemented based on the composite board quality assessment method according to any one of claims 1 to 9, characterized in that: include: The excitation acquisition module is used to apply a preset excitation field to the interface of the titanium-steel composite plate to obtain the initial response data field of the local interface; A feature construction module is used to construct a latent variable feature space characterizing the interface binding energy anomaly based on the initial response data field and through the physical constraint relationship between the interface local response characteristics and the binding energy distribution; The topology extraction module is used to extract the topological correlation structure between the local latent variable features in the latent variable feature space through the graph convolutional network, and obtain the initial candidate area of ​​local binding energy anomaly; A dynamic determination module is used to perform active response mode enhancement excitation on the local area of ​​the initial candidate region to obtain a dynamic enhanced response of each local area, and determine the abnormal binding energy area according to the change rate of the dynamic enhanced response; The cold weld positioning module is used to determine the specific location and range of the hidden cold weld zone in the titanium-steel composite plate interface based on the difference in the associated topological structure between the local characteristics in the abnormal binding energy area and the normal area.

Images