Impact load time history reconstruction and positioning method, equipment, medium and product

Abstract

The invention discloses an impact load time history reconstruction and positioning method and device, a medium and a product, and relates to the field of impact load recognition, and the method comprises the steps: obtaining an acceleration signal matrix and a dynamic graph structure of an impact load acting object; according to the acceleration signal matrix and the dynamic graph structure, utilizing a time history reconstruction model to obtain a reconstructed impact load time history of the impact load acting object; the time history reconstruction model is obtained by training a graph neural network model by using a first training data set and a dynamic graph structure of an impact load acting object for training; according to the coding distance matrix of the acceleration signal matrix, utilizing an impact positioning model to determine the impact position of the impact load acting object; the impact positioning model is obtained by training the full-connection neural network by using the second training data set. According to the invention, the impact load time history and the impact position when the load action position is unknown can be accurately identified.

Description

Technical Field

[0001] The present application relates to the field of impact load identification, and in particular to a method, device, medium and product for reconstructing and locating the time history of an impact load. Background Art

[0002] Dynamic load identification is a major concern in aerospace structural design and has received extensive attention in recent years. There is an increasing demand for quantification and visualization of dynamic loads on aerospace structures for structural health monitoring, structural vibration control, and vibration transfer path analysis. Dynamic loads are reconstructed based on the dynamic characteristics of the structure and the measured responses (such as acceleration, velocity, displacement, and strain), which is a challenging inverse problem. There has been extensive research on the reconstruction of the time history of the loads when the impact load locations are assumed to be known. Many scholars have accurately dealt with the severe ill-posedness of the inverse problem through a series of frequency domain and time domain methods. However, in engineering practice and other practical problems, load identification is more practical but also more challenging when the impact location is unknown. It has two characteristics that make it difficult to solve. First, it is an inverse problem. Second, the transfer relationship between the input (external impact force) and the output (structural response) is also unknown, which makes the problem more complicated. In this case, many studies set some assumptions and use prior knowledge to constrain the problem so that it can be solved numerically. By assuming candidate impact locations, the time history of the impact load at each candidate location is regarded as a partial content of all loads acting at all candidate locations. However, when the impact load does not act completely on the candidate position, the transfer function between the impact position and the response position is completely uncertain, which challenges the effectiveness of the above methods.

[0003] With the rapid development of artificial intelligence in recent years, artificial neural networks (ANNs) have become an effective method to solve complex mapping relationships. As a data-driven method, ANNs do not rely on mechanical models and only require data for supervised and / or unsupervised learning. Therefore, ANNs have great potential for the application of load identification, especially when the above-mentioned model-driven methods show limitations. However, since the purely data-driven method operates like a black box, ANNs require a large amount of data to learn the response of the target structure. Although current studies on simple structures have shown that ANNs can achieve good accuracy in load time history reconstruction only through the time information of the structural dynamic characteristics. However, if additional spatial information can be provided, it will definitely help to reduce the sample size, and at the same time achieve higher accuracy and reliability on complex structures such as real aircraft structures.

[0004] Recently, graph neural networks (GNNs) have attracted increasing attention due to the powerful expressive power of graphs. A graph is a data structure that models a set of objects (nodes) and their relationships (edges). These properties also make GNNs less dependent on data-driven methods and more dependent on model-driven methods, where the model is the graph. GNNs can be used to represent a large number of systems across various fields, including social sciences (such as social networks), natural sciences (such as physical systems, structural health monitoring), and knowledge graphs, as well as many other research areas. In general, GNN models learn from graph edge and hidden layer representations that encode local graph structures and node features. Compared with ANNs, the unique advantage of GNNs is that the connection relationship between nodes is considered, which provides additional topological information to the neural network to help achieve the learning goal. However, due to the generality of GNNs, this topological information is constructed in an abstract way, resulting in the inability to accurately identify the impact load time history and impact location when the load action location is unknown. Summary of the invention

[0005] The purpose of this application is to provide a method, device, medium and product for reconstructing and locating an impact load time history, so as to accurately identify the impact load time history and impact position when the load action position is unknown.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides a method for reconstructing and locating an impact load time history, comprising:

[0008] Acquire the acceleration signal matrix and dynamic graph structure of the impact load object; the dynamic graph structure includes a dynamic weighted measurement matrix and a dynamic admittance matrix; the dynamic admittance matrix is ​​determined according to the dynamic weighted adjacency matrix and the dynamic weighted measurement matrix; the dynamic weighted adjacency matrix is ​​determined according to the adjacency matrix and the dynamic weight matrix; the adjacency matrix is ​​used to represent the undirected graph structure of the impact load object; the undirected graph structure is composed of N nodes and N×(N-1) edges; N acceleration sensors are set on the impact load object; one node corresponds to one acceleration sensor; the dynamic weight matrix is ​​determined according to the Euclidean distance between the characteristics of each node; the node feature is the acceleration signal collected by the acceleration sensor; the dynamic weighted measurement matrix is ​​determined according to the dynamic weighted adjacency matrix;

[0009] According to the acceleration signal matrix and the dynamic graph structure, a time history reconstruction model is used to obtain a reconstructed impact load time history of the impact load object; wherein the time history reconstruction model is obtained by training a graph neural network model using a first training data set and a dynamic graph structure of the training impact load object; the first training data set includes the acceleration signal matrix of the training impact load object and the corresponding real impact load time history; the graph neural network model includes a graph interaction layer, a reconstruction layer and a graph fusion layer connected in sequence;

[0010] According to the encoding distance matrix of the acceleration signal matrix, the impact position of the object acted by the impact load is determined using the impact positioning model; the encoding distance matrix is ​​determined based on the reconstructed impact load time history of each node and the dynamic weighted measurement matrix; the reconstructed impact load time history of each node is determined using the reconstruction layer of the time history reconstruction model; wherein the impact positioning model is obtained by training a fully connected neural network using a second training data set; the second training data set includes the encoding distance matrix of the acceleration signal matrix of the object acted by the impact load for training and the corresponding real impact position; the fully connected neural network includes a first fully connected layer and a second fully connected layer connected in sequence.

[0011] Optionally, training the graph neural network model using the first training data set and the dynamic graph structure of the training impact load object specifically includes:

[0012] Acquire the training data set, the verification data set and the dynamic graph structure of the object acted by the verification impact load; the verification data set includes the acceleration signal matrix of the object acted by the verification impact load and the corresponding real impact load time history;

[0013] Inputting the acceleration signal matrix and the dynamic graph structure of the training impact load object into the current graph neural network model to obtain the reconstructed impact load time history of the training impact load object;

[0014] Determine the loss function value of the training data set using a mean square error loss function according to the reconstructed impact load time history of the training impact load subject and the corresponding real impact load time history;

[0015] Adjust the model parameters of the current graph neural network model by back-propagation according to the loss function value of the training data set to obtain a trained graph neural network model;

[0016] Inputting the acceleration signal matrix and the dynamic graph structure of the object subjected to the verification impact load into the currently trained graph neural network model to obtain the reconstructed impact load time history of the object subjected to the verification impact load;

[0017] Determine the loss function value of the verification data set using a mean square error loss function according to the reconstructed impact load time history of the verification impact load object and the corresponding true impact load time history;

[0018] Determine whether the maximum number of iterations has been reached;

[0019] If yes, the training is terminated and the currently trained graph neural network model is used as the time history reconstruction model;

[0020] If not, adjust the current learning rate according to the loss function value of the training data set and the loss function value of the verification data set, use the trained graph neural network model as the current graph neural network model, and return "input the acceleration signal matrix and dynamic graph structure of the object acted by the training impact load into the current graph neural network model to obtain the reconstructed impact load time history of the object acted by the training impact load".

[0021] Optionally, the acceleration signal matrix and the dynamic graph structure of the training impact load object are input into the current graph neural network model to obtain the reconstructed impact load time history of the training impact load object, specifically including:

[0022] Determine the encoding of each node using the graph interaction layer according to the acceleration signal matrix and the dynamic graph structure of the object subjected to the training impact load;

[0023] Determining a reconstructed impact load time history of each node using the reconstruction layer according to the encoding of each node;

[0024] According to the reconstructed impact load time history of each node, the reconstructed impact load time history of the training impact load action object is determined by using the graph fusion layer.

[0025] Optionally, according to the reconstructed impact load time history of the training impact load subject and the corresponding real impact load time history, a loss function value of the training data set is determined using a mean square error loss function, specifically including:

[0026] Using the formula Determine the loss function value of the training data set; where, is the reconstructed impact load time history of the training impact load object; F is the real impact load time history; pr is the Pearson correlation coefficient; the Pearson correlation coefficient is determined according to the encoding distance of the acceleration signal matrix of the training impact load object and the physical distance between the impact position and the acceleration sensor.

[0027] Optionally, adjusting the current learning rate according to the training data set loss function value and the validation data set loss function value specifically includes:

[0028] Determine whether the loss function value of the validation data set is greater than a set multiple of the loss function value of the training data set;

[0029] If yes, adjust the current learning rate.

[0030] Optionally, determining the encoding distance matrix according to the reconstructed impact load time history of each node and the dynamic weighted metric matrix specifically includes:

[0031] Using the formula Determine the encoding distance corresponding to each node; where δ is the encoding distance; FG is the reconstructed impact load time history of the node; is the dynamic weighted metric matrix; is the Hadamard product; Mean(·) is the average function of each row of the matrix; Diag(·) is the transformation function that transforms a diagonal matrix into a column vector;

[0032] The encoding distances corresponding to N nodes constitute the encoding distance matrix.

[0033] Optionally, determining the impact position of the impact load object according to the encoding distance matrix of the acceleration signal matrix and using an impact positioning model specifically includes:

[0034] Using the formula Determine the impact position of the impact load object; wherein, is the impact position of the impact load object; σ(·) is the activation function; δ is the encoding distance; T is the transposition; is the weight matrix of the first fully connected layer; is the weight matrix of the second fully connected layer; Hid is the number of neurons in the hidden layer.

[0035] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described methods for reconstructing and locating the time history of impact loads.

[0036] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for reconstructing and locating the time history of impact loads.

[0037] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned impact load time history reconstruction and positioning methods.

[0038] According to the specific embodiments provided in this application, this application has the following technical effects:

[0039] The present application provides a method, device, medium and product for reconstructing and locating the time history of an impact load, and obtains the acceleration signal matrix and dynamic graph structure of an impact load object; the dynamic graph structure includes a dynamic weighted metric matrix and a dynamic admittance matrix; the dynamic admittance matrix is ​​determined based on the dynamic weighted adjacency matrix and the dynamic weighted metric matrix; the dynamic weighted adjacency matrix is ​​determined based on the adjacency matrix and the dynamic weight matrix; the adjacency matrix is ​​used to represent the undirected graph structure of the impact load object; the undirected graph structure is composed of nodes and edges; an acceleration sensor is set on the impact load object; one node corresponds to one acceleration sensor; the dynamic weight matrix is ​​determined based on the Euclidean distance between the features of each node; the dynamic weighted metric matrix is ​​determined based on the dynamic weighted adjacency matrix ; According to the acceleration signal matrix and the dynamic graph structure, the time history reconstruction model is used to obtain the reconstructed impact load time history of the impact load object; wherein, the time history reconstruction model is obtained by training the graph neural network model using the first training data set and the dynamic graph structure of the training impact load object; according to the encoding distance matrix of the acceleration signal matrix, the impact position of the impact load object is determined using the impact positioning model; the encoding distance matrix is ​​determined based on the reconstructed impact load time history and the dynamic weighted measurement matrix of each node; the reconstructed impact load time history of each node is determined using the reconstruction layer of the time history reconstruction model; wherein, the impact positioning model is obtained by training the fully connected neural network using the second training data set. The present application can accurately identify the impact load time history and impact position when the load position is unknown. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A schematic diagram of a flow chart of a method for reconstructing and locating a time history of an impact load provided in an embodiment of the present application;

[0042] Figure 2It is a flow chart of a method for reconstructing the time history of impact load and locating impact of a plate structure under unknown load action position in the present application;

[0043] Figure 3 It is a schematic diagram of the graph structure adopted in this application;

[0044] Figure 4 is a diagram of the helicopter belly structure used in this application;

[0045] Figure 5 It is the layout diagram of the sensor acquisition system used in this application;

[0046] Figure 6 It is a time history reconstruction comparison diagram of the identified load and the real applied load of the present application and a schematic diagram of the error evaluation index;

[0047] Figure 7 It is a comparison chart of the impact location results of the identified load and the actual applied load of the present application;

[0048] Figure 8 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0051] This application provides a graph neural network model that can identify the time history and impact position of impact loads when the load action is unknown by learning the time and space information of the structural dynamic response, and is applied to real aircraft structures. This method has stronger applicability than traditional numerical methods and existing neural network methods.

[0052] In an exemplary embodiment, Figure 1 As shown, a method for reconstructing and locating the time history of an impact load is provided, comprising the following steps:

[0053] S1: Obtain the acceleration signal matrix and dynamic graph structure of the object under impact load; the dynamic graph structure includes a dynamic weighted measurement matrix and a dynamic admittance matrix; the dynamic admittance matrix is ​​determined based on the dynamic weighted adjacency matrix and the dynamic weighted measurement matrix; the dynamic weighted adjacency matrix is ​​determined based on the adjacency matrix and the dynamic weight matrix; the adjacency matrix is ​​used to represent the undirected graph structure of the object under impact load; the undirected graph structure consists of N nodes and N×(N-1) edges; N acceleration sensors are set on the object under impact load; one node corresponds to one acceleration sensor; the dynamic weight matrix is ​​determined based on the Euclidean distance between the features of each node; the node feature is the acceleration signal collected by the acceleration sensor; the dynamic weighted measurement matrix is ​​determined based on the dynamic weighted adjacency matrix.

[0054] S2: According to the acceleration signal matrix and the dynamic graph structure, a time history reconstruction model is used to obtain a reconstructed impact load time history of the object acted by the impact load; wherein the time history reconstruction model is obtained by training a graph neural network model using a first training data set and a dynamic graph structure of the object acted by the training impact load; the first training data set includes the acceleration signal matrix of the object acted by the training impact load and the corresponding real impact load time history; the graph neural network model includes a graph interaction layer, a reconstruction layer and a graph fusion layer connected in sequence.

[0055] S3: According to the encoding distance matrix of the acceleration signal matrix, the impact position of the impact load object is determined using the impact positioning model; the encoding distance matrix is ​​determined based on the reconstructed impact load time history of each node and the dynamic weighted measurement matrix; the reconstructed impact load time history of each node is determined using the reconstruction layer of the time history reconstruction model; wherein the impact positioning model is obtained by training a fully connected neural network using a second training data set; the second training data set includes the encoding distance matrix of the acceleration signal matrix of the training impact load object and the corresponding real impact position; the fully connected neural network includes a first fully connected layer and a second fully connected layer connected in sequence.

[0056] As an optional implementation, determining the encoding distance matrix according to the reconstructed impact load time history of each node and the dynamic weighted metric matrix specifically includes:

[0057] Using the formula Determine the encoding distance corresponding to each node; where δ is the encoding distance; FG is the reconstructed impact load time history of the node; is the dynamic weighted metric matrix; is the Hadamard product; Mean(·) is the average function of each row of the matrix; Diag(·) is the transformation function that transforms a diagonal matrix into a column vector.

[0058] The encoding distances corresponding to N nodes constitute the encoding distance matrix.

[0059] As an optional implementation, S3 specifically includes:

[0060] Using the formula Determine the impact position of the impact load object; wherein, is the impact position of the impact load object; σ(·) is the activation function; δ is the encoding distance; T is the transposition; is the weight matrix of the first fully connected layer; is the weight matrix of the second fully connected layer; Hid is the number of neurons in the hidden layer.

[0061] In practical applications, inspired by GNN, this application develops a new graph neural network model. This model is driven by graph structure. Under the constraints of real physical information, the model can simultaneously identify the time history and impact position of impact loads by learning from the time and space information of the structural dynamic response. Finally, the application verification of impact load reconstruction and positioning based on graph neural network is carried out in combination with a typical aircraft structure such as the helicopter belly structure.

[0062] For the impact load identification system of helicopter belly structure, N acceleration sensors S i (i=1, 2, ..., N) arranged on the helicopter web structure Position, used to measure the time domain acceleration signal. i The measured data is in vector form S i Indicates that the vector S i The length of is L, named as sensor signal vector. The load is a single-point impact force, and its impact position is P F (x, y) is random in the identified area. The reconstructed impact load time history is represented by the vector The measured data of the applied load is expressed in vector form F, named as the real impact load time history, as a comparison sample.

[0063] For the helicopter web structure, the load identification control equation in the time domain is as follows:

[0064]

[0065] Where S(t) is the time history of the sensor signal, P F (x, y) is the impact position, P s(x, y) is the sensor position, F(t) is the real impact load time history, To reconstruct the impact load time history, ε is the acceptable error between the two. Ω s,F It is the transfer function between the sensor signal and the impact load signal.

[0066] The acceleration sensor is abstracted as a response node, and all response nodes are connected with undirected edges, so that each acceleration sensor is connected with other acceleration sensors. The time domain acceleration signal S collected by each acceleration sensor i As each node feature. The undirected graph structure has a total of N nodes and N×(N-1) edges. The undirected graph structure is represented by the adjacency matrix It is expressed as:

[0067]

[0068] Among them, the adjacency matrix is a square matrix with all diagonal elements being 0 and all other elements being 1. It is worth noting that the adjacency matrix The element values ​​in only represent the abstract topological connection relationship between nodes. However, the feature differences between nodes, that is, the time domain feature differences between acceleration sensors, are closely related to the edges between nodes. These differences are caused by the different arrangements of acceleration sensors, so similarity judgment can be used to quantify such differences.

[0069] Therefore, calculate each pair of node features S i and S j The Euclidean distance between i，j , as shown below:

[0070] M i，j =||S i -S j ||2 (4)

[0071] Among them, M i，j is the Euclidean distance between the i-th node feature and the j-th node feature, i, j = 1, 2, ..., N, and the Euclidean distance M i，j The larger it is, the worse the similarity between node features is.

[0072] Therefore, the following Euclidean distance matrix can be obtained

[0073]

[0074] in, is a symmetric matrix with zero diagonal elements. The difference values ​​of its elements can be considered as the information validity weights generated by the different positions of the acceleration sensors. The matrix It is called the dynamic weight matrix.

[0075] By transforming the adjacency matrix and the dynamic weight matrix By multiplying the elements, the information validity difference between node features can be used as a specific representation of the edge, replacing the abstract topological relationship of the matrix. A dynamic weighted adjacency matrix can be obtained As shown below:

[0076]

[0077] in, is the Hadamard product. Then, we can get a dynamic weighted metric matrix As shown below:

[0078]

[0079] Among them, the dynamic weighted measurement matrix The elements of represent the sum of the information validity weights between the node and other nodes in the graph.

[0080] With a dynamic weighted adjacency matrix and the dynamic weighted metric matrix A dynamic admittance matrix can be calculated The matrix is ​​as follows:

[0081]

[0082] Therefore, based on the above theory, the designed three-layer graph neural network model is as follows. First, the node features of each node are combined into an input matrix As shown below:

[0083]

[0084] The first layer is the graph interaction layer, in which the input matrix and dynamic graph structures (including dynamic weighted metric matrices and the dynamic admittance matrix ) as input. Each node feature exchanges and propagates information through the graph structure. The first layer can be expressed as follows:

[0085]

[0086] Among them, σ(·) is the activation function, which is generally the ReLU function. and is a learnable weight matrix, where Hid is the number of hidden neurons, is the output signal of the graph interaction layer, that is, the encoding of each node. Effective information extraction of each node feature, the second term Δ·S·W 12 Represents the information exchange and propagation of the entire graph structure.

[0087] The second layer is the reconstruction layer. It encodes each node output by the first layer Reconstructed into a load feature. Similar to the first layer, the second layer can be expressed as follows:

[0088]

[0089] in, and is a learnable weight matrix. It is worth noting that the output of the second layer should be the reconstructed impact load time history of each node. and the dynamic weighted metric matrix A hidden variable can be calculated, through which a direct linear relationship can be established with the different distances of the sensor at the moment of load impact. This hidden variable is named the encoding distance δ N×1 , which is expressed as follows:

[0090]

[0091] Among them, Mean(·) is the average function of each row of the matrix, and Diag(·) is the transformation of a diagonal matrix into a column vector. In order to give the encoding distance matrix δ N×1 Physically, constraints are applied to guide the model to extract the physical distance matrix ρ between the impact location and the sensor location N×1 The Pearson correlation coefficient is used to measure δ N×1 and ρ N×1 The linear correlation between them is expressed as:

[0092]

[0093] Among them, pr is Covariance with ρ and their standard deviation and The result is always between -1 and 1.

[0094] The third layer is the graph fusion layer, which uses the output of the second layer As input, the reconstructed impact load time history of each node is fused. Finally, the time domain impact load characteristics are output The third layer can be expressed as follows:

[0095]

[0096] Among them, the trainable weight matrix Node features for fusing and decoding graphs.

[0097] The graph neural network model uses the modified mean square error loss function as shown below:

[0098]

[0099] Among them, the term |pr| is added as the denominator to the reconstructed impact load time history In the mean squared error loss function between the sample F and the control sample F, the encoding distance that is poorly correlated with the physical distance is penalized. Finally, the loss function LossGNN is back-propagated to update the weights of the graph neural network model.

[0100] After the graph neural network training is completed, a two-layer fully connected neural network is set up as shown below. First, the encoding distance δ between the training dataset samples and the validation dataset samples is extracted N×1 , as the input of the training set samples and the validation set samples of the fully connected network, the final output is the location of the impact load As shown below:

[0101]

[0102] in, is the impact position of the output, and They are the weight matrices of the first and second layers of the two-layer fully connected neural network model respectively.

[0103] The two-layer fully connected neural network uses the mean square error loss function as shown below:

[0104]

[0105] Finally, the loss function LossBP is back-propagated to update the weights of the fully connected neural network model.

[0106] As an optional implementation, the graph neural network model is trained using the first training data set and the dynamic graph structure of the training impact load object, specifically including:

[0107] Step 1: Obtain the training data set, the verification data set and the dynamic graph structure of the object to which the verification impact load is applied; the verification data set includes the acceleration signal matrix of the object to which the verification impact load is applied and the corresponding real impact load time history.

[0108] Step 2: Input the acceleration signal matrix and dynamic graph structure of the object subjected to the training impact load into the current graph neural network model to obtain the reconstructed impact load time history of the object subjected to the training impact load.

[0109] As an optional implementation, step 2 specifically includes:

[0110] Step 21: According to the acceleration signal matrix and dynamic graph structure of the object subjected to the training impact load, the coding of each node is determined by using the graph interaction layer.

[0111] Step 22: According to the encoding of each node, using the reconstruction layer, determine the reconstructed impact load time history of each node.

[0112] Step 23: According to the reconstructed impact load time history of each node, the graph fusion layer is used to determine the reconstructed impact load time history of the training impact load action object.

[0113] Step 3: According to the reconstructed impact load time history of the training impact load object and the corresponding real impact load time history, the loss function value of the training data set is determined using the mean square error loss function.

[0114] As an optional implementation, step 3 specifically includes:

[0115] Using the formula Determine the loss function value of the training data set; where, is the reconstructed impact load time history of the training impact load object; F is the real impact load time history; pr is the Pearson correlation coefficient; the Pearson correlation coefficient is determined according to the encoding distance of the acceleration signal matrix of the training impact load object and the physical distance between the impact position and the acceleration sensor.

[0116] Step 4: Adjust the model parameters of the current graph neural network model by back-propagation according to the loss function value of the training data set to obtain the trained graph neural network model.

[0117] Step 5: Input the acceleration signal matrix and dynamic graph structure of the object subjected to the verification impact load into the currently trained graph neural network model to obtain the reconstructed impact load time history of the object subjected to the verification impact load.

[0118] Step 6: According to the reconstructed impact load time history of the object subjected to the verification impact load and the corresponding real impact load time history, a loss function value of the verification data set is determined using a mean square error loss function.

[0119] Step 7: Determine whether the maximum number of iterations has been reached.

[0120] Step 8: If yes, end the training and use the currently trained graph neural network model as the time history reconstruction model.

[0121] Step 9: If not, adjust the current learning rate according to the loss function value of the training data set and the loss function value of the verification data set, and use the trained graph neural network model as the current graph neural network model, and return to "input the acceleration signal matrix and dynamic graph structure of the object subjected to the training impact load into the current graph neural network model to obtain the reconstructed impact load time history of the object subjected to the training impact load".

[0122] As an optional implementation manner, adjusting the current learning rate according to the loss function value of the training data set and the loss function value of the validation data set specifically includes:

[0123] Determine whether the loss function value of the validation data set is greater than a set multiple of the loss function value of the training data set.

[0124] If yes, adjust the current learning rate.

[0125] In practical applications, after building the graph neural network model and the fully connected neural network, such as Figure 2 As shown, the specific training steps are as follows:

[0126] Step 1: Arrange sensors and acquisition systems on the helicopter belly structure, where the sensors use N accelerometers, which are connected to an external laser vibrometer to collect the acceleration time history of the belly structure when it vibrates. The force sensor on the hammer is connected to an external laser vibrometer to collect the impact load time history of each strike. The entire belly structure is divided into Q sub-areas, and a hammer is used to strike Q1, Q2, and Q3 times at any position in each sub-area, which are used to form a training data set, a validation data set, and a test data set, respectively, where Q1:Q2:Q3 is approximately 15:1:1. At the same time, the actual distance ρ between each impact position and the N accelerometers is recorded. N×1 .

[0127] Step 2: Construct the sensor system on the web structure into a graph structure, such as Figure 3 As shown, the adjacency matrix is ​​calculated And the sensor acceleration signal is used as the node feature of the graph structure.

[0128] Step 3: Get the dynamic weight matrix by calculating the node features between each node Based on dynamic weight matrix and adjacency matrix Calculate the dynamic weighted adjacency matrix Dynamically weighted metric matrix and the dynamic admittance matrix

[0129] Step 4: After shuffling the training data set, form the acceleration signal matrix according to Eq. Input into the above graph neural network model for training, and the output is the reconstructed impact load time history The actual impact load time history collected from the training set samples For comparison, according to Eq. and the Pearson correlation coefficient pr corresponding to this group of samples, the loss function LossGNNTr of the graph neural network is calculated, and then back propagation is performed. After the training of the training data set is completed and the back propagation is completed, the same operation is performed on the verification data set and the loss function LossGNNVa is calculated. The verification data set does not participate in the back propagation of the graph neural network model. If LossGNNVa>1.1×LossGNNTr, the learning rate at this time is adjusted to 0.7 times the learning rate of the previous step to prevent overfitting of the graph neural network model. Complete one training of the graph neural network model as described above, train the graph neural network model about 200 times, observe the loss function in each training, and adjust the initial learning rate of the graph neural network model so that the loss function curve converges.

[0130] Step 5: After the graph neural network model training is completed, the hyperparameters and weights are fixed and will not change. The training dataset and the validation dataset are input into the trained graph neural network in the above form in sequence, and the encoding distance corresponding to the training dataset and the validation dataset is output.

[0131] Step 6: After the graph neural network model training is completed, the hyperparameters and weights are fixed and no longer changed. The test data set is input into the trained graph neural network in the above form, and the reconstructed test set impact load time history is output. and the test set encoding distance.

[0132] Step 7: Shuffle the encoding distances corresponding to the training data set and input them into the above fully connected neural network for training. The output load impact position The actual load impact position collected in the training data set For comparison, calculate the loss function LossBPTr of the fully connected neural network according to Eq., and then perform back propagation. After the encoding distance training corresponding to the training data set is completed and the back propagation is completed, the encoding distance corresponding to the verification data set is operated in the same way and the loss function LossBPVa is calculated. The encoding distance corresponding to the verification data set does not participate in the back propagation of the fully connected neural network. If LossBPVa>1.1×LossBPTr, the learning rate at this time is adjusted to 0.7 times the learning rate of the previous step to prevent overfitting of the fully connected neural network. Complete a training of the fully connected neural network as described above, train the fully connected neural network about 1000 times, observe the loss function in each training, and adjust the initial learning rate of the fully connected neural network so that the loss function curve converges.

[0133] Step 8: After the fully connected neural network model is trained, the hyperparameters and weights are fixed and no longer changed. The encoding distance corresponding to the test data set is input into the trained fully connected neural network, and the load impact position of the test set is output.

[0134] Step 9: In order to uniformly evaluate the generalization and reliability of the graph neural network model, two evaluation criteria are proposed to evaluate the accuracy of the impact load time history reconstruction and the accuracy of the impact location: For the reconstruction of the impact load time history, since the impact load has the characteristics of high energy and concentration, the peak error standard PE commonly used in the field of structural health monitoring is used to evaluate the true impact load vector F and the reconstructed impact load time history The peak difference is For load impact positioning, use the real load impact position (x, y) and the reconstructed load impact position The distance difference DE between them is evaluated, where Each set of sample data in the test data set is identified and the error is calculated. The PE and DE of all sample data are averaged. When the averaged identification error PE < 10% and DE < 10cm, the identification is considered successful.

[0135] like Figure 4 and Figure 5As shown, 7 sensors and acquisition systems are arranged on the helicopter belly structure. The acceleration sensor is connected to an external laser vibrometer to collect the acceleration time history of the belly structure when it vibrates. The force sensor on the force hammer is connected to an external laser vibrometer to collect the dynamic load time history of each strike. The entire belly structure is divided into 84 sub-areas. A force hammer is used to strike 15 times, 1 time, and 1 time at any position in each sub-area, respectively, to form training set samples, validation set samples, and test set samples. At the same time, the distance between the impact position of each sample and the sensor is recorded. The sensor system on the belly structure is constructed as a graph structure, and the adjacency matrix is ​​calculated. The sensor acceleration signal is used as the node feature of the graph structure. The dynamic weight matrix is ​​obtained by calculating the node features between each node. and the adjacency matrix Calculate together to get the dynamic weighted adjacency matrix Dynamically weighted metric matrix and the dynamic admittance matrix After shuffling the training data set samples, the acceleration signal matrix is ​​formed according to Eq. Input into the above graph neural model for training, adjust the hyperparameters of the graph neural network until the loss function reaches convergence, and use the validation data set samples to participate in the training to prevent overfitting. After the graph neural network model training is completed, output the encoding distances corresponding to the training data set samples and the validation data set samples. After shuffling the training set encoding distances, input them into the above fully connected neural model for training in turn, adjust the hyperparameters of the fully connected neural network until the loss function reaches convergence, and use the validation set encoding distances to participate in the training to prevent overfitting. Input the test set samples into the trained graph neural network, and output the identified test set dynamic load time history and test set encoding distance. Input the test set encoding distance into the trained fully connected neural network, and output the identified test set dynamic load impact position. The average error of all recognition results of the test set is PE=9.12% and DE=6.73cm, which meets the error standard. The impact load time history recognition result closest to the average value is taken as follows Figure 6 As shown in Figure 2, the impact location results of the helicopter web structure are as follows: Figure 7 shown.

[0136] It can be seen from the results of the above embodiments that the present application is able to identify the dynamic load time history and the impact load action position at different unknown positions on the helicopter belly structure, and is basically consistent with the applied load in terms of magnitude and regularity.

[0137] This application uses the structural acceleration response collected by the sensor and the real distance between the sensor and the impact position as the input of the graph neural network, and uses the load time history of the hammer excitation as the output of the graph neural network. With the help of physical distance, the graph neural network obtains the complex mapping relationship between the impact load and the sensor at different positions through graph structure learning, and finally realizes the reconstruction of the impact load time history when the impact load position is unknown, which is impossible to achieve with traditional model-driven methods.

[0138] This application introduces the influence of impact loads at different positions on sensor response into the graph neural network by calibrating and sampling the loads at different positions, constructs a latent variable encoding distance, and constructs a linear relationship between the encoding distance and the true distance, thereby achieving high-precision positioning of the impact load through a simple fully connected neural network. Compared with the ANNs method, the number of sensors required is less and it is more interpretable.

[0139] This application has been tested on a real helicopter belly structure, occupies less training hardware, and can ensure that the recognition accuracy meets the requirements, making it suitable for promotion and use in real aerospace structures.

[0140] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above-mentioned impact load time history reconstruction and positioning method when executing the computer program.

[0141] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the above-mentioned impact load time history reconstruction and positioning method.

[0142] In an exemplary embodiment, a computer program product is provided, comprising a computer program, which implements the above-mentioned impact load time history reconstruction and positioning method when executed by a processor.

[0143] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for reconstructing and locating the time history of an impact load is implemented.

[0144] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0145] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0146] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0147] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0148] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0149] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for reconstructing and locating the time history of an impact load, characterized in that: include: Acquire the acceleration signal matrix and dynamic graph structure of the object subjected to the impact load; the dynamic graph structure includes a dynamic weighted measurement matrix and a dynamic admittance matrix; The dynamic admittance matrix is ​​determined according to the dynamic weighted adjacency matrix and the dynamic weighted metric matrix; the dynamic weighted adjacency matrix is ​​determined according to the adjacency matrix and the dynamic weight matrix; the adjacency matrix is ​​used to represent the undirected graph structure of the impact load object; the undirected graph structure is composed of N nodes and N×(N-1) edges; N acceleration sensors are set on the impact load object; one node corresponds to one acceleration sensor; the dynamic weight matrix is ​​determined according to the Euclidean distance between the characteristics of each node; the node feature is the acceleration signal collected by the acceleration sensor; the dynamic weighted metric matrix is ​​determined according to the dynamic weighted adjacency matrix; According to the acceleration signal matrix and the dynamic graph structure, a time history reconstruction model is used to obtain a reconstructed impact load time history of the impact load object; wherein the time history reconstruction model is obtained by training a graph neural network model using a first training data set and a dynamic graph structure of the training impact load object; the first training data set includes the acceleration signal matrix of the training impact load object and the corresponding real impact load time history; the graph neural network model includes a graph interaction layer, a reconstruction layer and a graph fusion layer connected in sequence; According to the encoding distance matrix of the acceleration signal matrix, the impact position of the object acted by the impact load is determined using the impact positioning model; the encoding distance matrix is ​​determined based on the reconstructed impact load time history of each node and the dynamic weighted measurement matrix; the reconstructed impact load time history of each node is determined using the reconstruction layer of the time history reconstruction model; wherein the impact positioning model is obtained by training a fully connected neural network using a second training data set; the second training data set includes the encoding distance matrix of the acceleration signal matrix of the object acted by the impact load for training and the corresponding real impact position; the fully connected neural network includes a first fully connected layer and a second fully connected layer connected in sequence.

2. The impact load time history reconstruction and positioning method according to claim 1, characterized in that: The graph neural network model is trained using the first training data set and the dynamic graph structure of the training impact load object, specifically including: Acquire the training data set, the verification data set and the dynamic graph structure of the object acted by the verification impact load; the verification data set includes the acceleration signal matrix of the object acted by the verification impact load and the corresponding real impact load time history; Inputting the acceleration signal matrix and the dynamic graph structure of the training impact load object into the current graph neural network model to obtain the reconstructed impact load time history of the training impact load object; Determine the loss function value of the training data set using a mean square error loss function according to the reconstructed impact load time history of the training impact load subject and the corresponding real impact load time history; Adjust the model parameters of the current graph neural network model by back-propagation according to the loss function value of the training data set to obtain a trained graph neural network model; Inputting the acceleration signal matrix and the dynamic graph structure of the object subjected to the verification impact load into the currently trained graph neural network model to obtain the reconstructed impact load time history of the object subjected to the verification impact load; Determine the loss function value of the verification data set using a mean square error loss function according to the reconstructed impact load time history of the verification impact load object and the corresponding true impact load time history; Determine whether the maximum number of iterations has been reached; If yes, the training is terminated and the currently trained graph neural network model is used as the time history reconstruction model; If not, adjust the current learning rate according to the loss function value of the training data set and the loss function value of the verification data set, use the trained graph neural network model as the current graph neural network model, and return "input the acceleration signal matrix and dynamic graph structure of the object acted by the training impact load into the current graph neural network model to obtain the reconstructed impact load time history of the object acted by the training impact load".

3. The impact load time history reconstruction and positioning method according to claim 2, characterized in that: Inputting the acceleration signal matrix and the dynamic graph structure of the training impact load object into the current graph neural network model to obtain the reconstructed impact load time history of the training impact load object specifically includes: Determine the encoding of each node using the graph interaction layer according to the acceleration signal matrix and the dynamic graph structure of the object subjected to the training impact load; Determining a reconstructed impact load time history of each node using the reconstruction layer according to the encoding of each node; According to the reconstructed impact load time history of each node, the reconstructed impact load time history of the training impact load action object is determined by using the graph fusion layer.

4. The impact load time history reconstruction and positioning method according to claim 2, characterized in that: According to the reconstructed impact load time history of the training impact load object and the corresponding real impact load time history, the loss function value of the training data set is determined by using the mean square error loss function, specifically including: Using the formula Determine the loss function value of the training data set; where, is the reconstructed impact load time history of the training impact load object; F is the real impact load time history; pr is the Pearson correlation coefficient; the Pearson correlation coefficient is determined according to the encoding distance of the acceleration signal matrix of the training impact load object and the physical distance between the impact position and the acceleration sensor.

5. The impact load time history reconstruction and positioning method according to claim 2, characterized in that: According to the loss function value of the training data set and the loss function value of the validation data set, the current learning rate is adjusted, specifically including: Determine whether the loss function value of the validation data set is greater than a set multiple of the loss function value of the training data set; If yes, adjust the current learning rate.

6. The impact load time history reconstruction and positioning method according to claim 1, characterized in that: Determining the encoding distance matrix according to the reconstructed impact load time history of each node and the dynamic weighted metric matrix specifically includes: Using the formula Determine the encoding distance corresponding to each node; where δ is the encoding distance; FG is the reconstructed impact load time history of the node; is a dynamic weighted metric matrix; ο is a Hadamard product; Mean(·) is the average function of each row of the matrix; Diag(·) is a transformation function that transforms a diagonal matrix into a column vector; The encoding distances corresponding to N nodes constitute the encoding distance matrix.

7. The impact load time history reconstruction and positioning method according to claim 1, characterized in that: Determining the impact position of the impact load object according to the encoding distance matrix of the acceleration signal matrix and using the impact positioning model specifically includes: Using the formula Determine the impact position of the impact load object; wherein, is the impact position of the impact load object; σ(·) is the activation function; δ is the encoding distance; T is the transposition; is the weight matrix of the first fully connected layer; is the weight matrix of the second fully connected layer; Hid is the number of neurons in the hidden layer.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the impact load time history reconstruction and positioning method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the impact load time history reconstruction and positioning method described in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the impact load time history reconstruction and positioning method described in any one of claims 1 to 7 is implemented.

Images