Aviation structure impact load non-negative sparse coding algorithm expansion identification method and system
By constructing a non-negative sparse coding algorithm to develop an identification system, the problem of monitoring impact loads on aerospace structures was solved, achieving high-precision and rapid impact load localization and reconstruction, and improving the stability and noise resistance of the identification system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XI AN JIAOTONG UNIV
- Filing Date
- 2024-10-28
- Publication Date
- 2026-04-28
AI Technical Summary
In existing technologies, it is difficult to effectively monitor and identify impact damage to aerospace structures under impact loads. Sensor measurements are limited and costly, and there is a lack of effective methods for force identification applications using algorithm-based network deployment.
A non-negative sparse coding algorithm is used to construct a deep learning network based on a dynamic model of impact load and non-negative sparse prior. The network is expanded through an iterative convergence threshold algorithm to identify impact loads. The network model is trained using a dataset to locate and reconstruct impact loads.
It improves the accuracy and stability of impact load positioning and reconstruction, achieves faster calculation speed and better noise resistance, and has better interpretability and accurate identification under uncertain sensor layout.
Smart Images
Figure CN121935596A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of structural health monitoring, and in particular to a method and system for identifying impact loads on aerospace structures using a non-negative sparse coding algorithm. Background Technology
[0002] Aerospace structures inevitably experience impact loads during service, such as those caused by bird strikes or falling maintenance tools. Under these loads, imperceptible damage may occur within the structure, affecting its mechanical properties and reliability, and potentially leading to failure or destruction. Therefore, monitoring impact loads on structures to promptly detect and assess impact damage is crucial for ensuring safe and reliable operation. However, directly measuring impact loads on a structure using force sensors has limitations, including limited measurement locations and high costs.
[0003] Impact loads exhibit sparsity in the time domain, which can serve as prior information to improve the accuracy of impact load localization. However, current research on algorithmic unfolding networks largely focuses on areas such as image processing and mechanical fault diagnosis, with a lack of algorithmic unfolding networks specifically for force recognition. Summary of the Invention
[0004] To address the aforementioned technical problems, the objective of this invention is to provide a method and system for identifying non-negative sparse coding algorithms for impact loads on aerospace structures.
[0005] To achieve the above objectives, the present invention adopts the following approach.
[0006] A method and system for identifying non-negative sparse coding algorithms for impact loads on aerospace structures, the method comprising the following steps:
[0007] S1. Collect and create a dataset containing the impact load force signal and the corresponding response signal data for the monitored aerospace structure;
[0008] S2. Construct a non-negative sparse coding model based on a dynamic model of impact load and a non-negative sparse prior.
[0009] S3. Construct an iterative convergence threshold algorithm for solving the non-negative sparse coding model;
[0010] S4. Construct a deep algorithm unfolding network based on the iterative convergence threshold algorithm;
[0011] S5. Use the dataset to train the deep algorithm unfolded network to obtain a network model with optimal parameters;
[0012] S6. Use the network model with optimal parameters to perform impact load identification test and output the impact load to be identified.
[0013] Optionally, in step S1, a dataset is created by conducting a hammer impact modal test on the monitored aerospace structure to obtain the transfer function S between the impact monitoring point and the response measurement point and the vibration response signal x of the measurement point.
[0014] Optionally, in step S2, the non-negative sparse coding model is:
[0015] .
[0016] Optionally, step S3 derives an iterative convergence threshold algorithm based on convex optimization and proximal gradient descent methods to solve the non-negative sparse coding model.
[0017] Optionally, step S4 includes the following steps:
[0018] S4.1: Update the parameters in the iterative convergence threshold algorithm through end-to-end training, that is:
[0019]
[0020] in, Represents the learnable weight parameters of each layer, and the adaptive coefficients. Represents learnable coefficients, bias vector Equal to the corresponding threshold, The activation function is consistent with the non-negativity constraint of the force vector, and the transfer matrix is... It is embedded into each layer as a fixed weight parameter;
[0021] S4.2: Set the number of expansions K;
[0022] S4.3: Reconstruct the impact force vector The initial value is set to Weight matrix of each layer Initialize to ; will be reconstructed The mean square error between the true force vector f and the actual force vector is used as the loss function, i.e.:
[0023]
[0024] in, The number of training samples. Force vector Size.
[0025] Optionally, in step S5, an early stopping strategy is adopted, whereby the network stops stopping when the loss function value on the validation set is continuously... If no decrease occurs in any batch, the network training process is terminated to obtain a network model with optimal parameters.
[0026] Optionally, in step S6, the response signal of the monitored object under the action of the impact load to be identified is collected as the input of the network, the impact load is identified using the network model with the best parameters, and the impact load to be identified is output.
[0027] A non-negative sparse coding algorithm expansion and recognition system for impact loads on aerospace structures, the system comprising:
[0028] The first unit is used to collect and create a dataset containing impact load force signals and response signals from the monitored aerospace structures.
[0029] The second unit is used to construct a non-negative sparse coding model based on a dynamic model of impact load and a non-negative sparse prior.
[0030] The third unit is used to construct an iterative convergence threshold algorithm to solve the non-negative sparse coding model.
[0031] The fourth unit is used to construct a deep algorithm unfolding network based on the iterative convergence threshold algorithm;
[0032] The fifth unit involves training the deep learning algorithm network using the dataset to obtain a network model with optimal parameters.
[0033] The sixth unit uses the network model with optimal parameters to perform impact load identification tests and outputs the impact load to be identified.
[0034] A computer-readable storage medium for storing a computer program configured to implement the method when invoked by a processor.
[0035] An electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor; wherein the processor implements the method when executing the program.
[0036] Compared with the prior art, the present invention has the following beneficial technical effects:
[0037] 1. Compared with iterative convergence threshold algorithm and learning iterative convergence threshold algorithm, it can improve the accuracy and stability of impact load localization and reconstruction;
[0038] 2. The neural network provided by this invention achieves more accurate impact load identification with far fewer iterations (K = 6) than the iterative convergence threshold algorithm, demonstrating superior computational speed;
[0039] 3. The neural network provided by this invention has better interpretability in its structural design, can accurately locate and reconstruct the impact force when the sensor layout is uncertain, and exhibits good noise resistance.
[0040] This invention is applicable to identifying impact loads on aerospace structures. It can simultaneously locate the impact load's position and reconstruct its time history. Compared with iterative convergence threshold algorithms and learning iterative convergence threshold algorithm networks, it can obtain solutions with higher accuracy. It has the advantages of fast calculation speed, no need to manually set parameters, and good noise resistance. Compared with traditional artificial neural networks, the neural network provided by this invention has better interpretability in structural design. Attached Figure Description
[0041] The accompanying drawings illustrate exemplary embodiments of the invention and, together with the description thereof, serve to explain the principles of the invention. These drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification.
[0042] Figure 1 This is a flowchart of a non-negative sparse coding algorithm expansion and identification method for impact loads on aerospace structures, according to one embodiment of the present invention.
[0043] Figure 2 This is an overall structural diagram of the network identification method based on the iterative convergence threshold algorithm of the non-negative sparse coding model in one embodiment of the present invention;
[0044] Figure 3 This is a schematic diagram of impact load identification of a cartridge structure in one embodiment of the present invention;
[0045] Figures 4(a) and 4(b) show the impact event applied to the cartridge and the corresponding strain response signals of the cartridge structure at four response measurement points in one embodiment of the present invention. In this figure, 4(a) shows the applied impact load and 4(b) shows the strain response signals at measurement points S1-S4.
[0046] Figures 5(a) and 5(b) show the impact load identification results of the cartridge structure in one embodiment of the present invention. Figure 5(a) shows the network time history reconstruction result based on the iterative algorithm, and Figure 5(b) shows the network impact localization result based on the iterative algorithm. Detailed Implementation
[0047] The following is in conjunction with the appendix Figures 1 to 5(b) The present invention will be further described in detail below with reference to the embodiments. It is to be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present invention are shown in the accompanying drawings.
[0048] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other. The technical solution of this invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0049] Unless otherwise stated, the exemplary embodiments / exemplifications shown are to be understood as providing exemplary features of various details that provide ways in which the technical concept of the invention can be implemented in practice. Therefore, unless otherwise stated, the features of the various embodiments / exemplifications may be additionally combined, separated, interchanged and / or rearranged without departing from the technical concept of the invention.
[0050] The use of crosshairs and / or shading in the accompanying drawings is generally used to clarify the boundaries between adjacent components. Thus, unless otherwise stated, the presence or absence of crosshairs or shading does not convey or indicate any preference or requirement for the specific material, material properties, dimensions, proportions, commonalities between the illustrated components, or any other characteristics, properties, etc., of the components. Furthermore, in the accompanying drawings, the dimensions and relative dimensions of components may be exaggerated for clarity and / or descriptive purposes. When exemplary embodiments can be implemented differently, a specific process sequence may be performed in a different order than that described. For example, two consecutively described processes may be performed substantially simultaneously or in the reverse order of their description. Furthermore, the same reference numerals denote the same components.
[0051] When a component is referred to as being "on" or "above" another component, "connected to," or "joined to" another component, the component may be directly on, directly connected to, or directly joined to the other component, or there may be intermediate components. However, when a component is referred to as being "directly on" another component, "directly connected to," or "directly joined to" another component, there are no intermediate components. Therefore, the term "connection" can refer to a physical connection, an electrical connection, etc., and may or may not have intermediate components.
[0052] For descriptive purposes, the present invention may use spatial relative terms such as “below,” “under,” “below,” “down,” “above,” “above,” “higher,” and “side (e.g., in a “sidewall”)” to describe the relationship between one component and another component as shown in the accompanying drawings. In addition to the orientations depicted in the drawings, the spatial relative terms are also intended to encompass different orientations of the device during use, operation, and / or manufacture. For example, if the device in the drawings is flipped, a component described as “below” or “under” another component or feature would subsequently be positioned “above” said other component or feature. Thus, the exemplary term “below” can encompass both “above” and “below” orientations. Furthermore, the device may be otherwise positioned (e.g., rotated 90 degrees or in other orientations), thus interpreting the spatial relative descriptive terms used herein accordingly.
[0053] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms “a” and “the” are intended to include the plural forms as well. Furthermore, when the terms “comprising” and / or “including” and variations thereof are used in this specification, it indicates the presence of the stated features, integrals, steps, operations, parts, components, and / or groups thereof, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, parts, components, and / or groups thereof. It should also be noted that, as used herein, the terms “substantially,” “about,” and other similar terms are used as approximate terms rather than as terms of degree, thus explaining the inherent biases in measurements, calculated values, and / or provided values that would be recognized by one of ordinary skill in the art.
[0054] In one embodiment, the present invention provides a method and system for identifying non-negative sparse coding algorithms for impact loads on aerospace structures, the method comprising the following steps:
[0055] S1. Collect and create a dataset containing the impact load force signal and the corresponding response signal data for the monitored aerospace structure;
[0056] S2. Construct a non-negative sparse coding model based on a dynamic model of impact load and a non-negative sparse prior.
[0057] S3. Construct an iterative convergence threshold algorithm for solving the non-negative sparse coding model;
[0058] S4. Construct a deep algorithm unfolding network based on the iterative convergence threshold algorithm;
[0059] S5. Use the dataset to train the deep algorithm unfolded network to obtain a network model with optimal parameters;
[0060] S6. Use the network model with optimal parameters to perform impact load identification test and output the impact load to be identified.
[0061] Optionally, in step S1, a dataset is created by conducting a hammer impact modal test on the monitored aerospace structure to obtain the transfer function S between the impact monitoring point and the response measurement point and the vibration response signal x of the measurement point.
[0062] Optionally, in step S2, the non-negative sparse coding model is:
[0063] .
[0064] Optionally, step S3 derives an iterative convergence threshold algorithm based on convex optimization and proximal gradient descent methods to solve the non-negative sparse coding model.
[0065] Optionally, step S4 includes the following steps:
[0066] S4.1: Update the parameters in the iterative convergence threshold algorithm through end-to-end training, that is:
[0067]
[0068] in, Represents the learnable weight parameters of each layer, and the adaptive coefficients. Represents learnable coefficients, bias vector Equal to the corresponding threshold, The activation function is consistent with the non-negativity constraint of the force vector, and the transfer matrix is... It is embedded into each layer as a fixed weight parameter;
[0069] S4.2: Set the number of expansions K;
[0070] S4.3: Reconstruct the impact force vector The initial value is set to Weight matrix of each layer Initialize to ; will be reconstructed The mean square error between the true force vector f and the actual force vector is used as the loss function, i.e.:
[0071]
[0072] in, The number of training samples. Force vector Size.
[0073] Optionally, in step S5, an early stopping strategy is adopted, whereby the network stops stopping when the loss function value on the validation set is continuously... If no decrease occurs in any batch, the network training process is terminated to obtain a network model with optimal parameters.
[0074] Optionally, in step S6, the response signal of the monitored object under the action of the impact load to be identified is collected as the input of the network, the impact load is identified using the network model with the best parameters, and the impact load to be identified is output.
[0075] In one embodiment, reference is made to Figure 1This invention provides a method for identifying impact loads on aerospace structures using a non-negative sparse coding algorithm. The method involves collecting and creating a dataset containing impact load force and response signals from the monitored object, establishing a non-negative sparse coding model based on the impact load dynamic model and non-negative sparse priors, constructing a deep learning network using an iterative convergence threshold algorithm, and then training and testing the model. The specific steps are as follows:
[0076] S1. Collect and create a dataset containing impact load force signals and response signals for the monitored aerospace structures;
[0077] In step S1, the transfer function S between the impact monitoring point and the response measurement point and the vibration response signal x of the measurement point are obtained by conducting a hammer impact modal test on the monitored aerospace structure, and a dataset is constructed.
[0078] S2. Considering the sparsity of impact loads in the time domain and the unidirectionality of the impact force direction, a non-negative sparse coding model based on the dynamic model of impact loads and non-negative sparse priors is constructed; the non-negative sparse coding model is expressed as:
[0079]
[0080] in, The vibration response signal at the measuring point. Represents the transfer matrix, for a given input We need to find an optimal solution. , Represents the residual term. Describes the L1 norm of a vector. Represents the regularization parameter;
[0081] S3. Construct an iterative convergence threshold algorithm for solving the non-negative sparse coding model;
[0082] Specifically, an iterative convergence threshold algorithm based on convex optimization and proximal gradient descent is derived to solve non-negative sparse coding models, including:
[0083] S3.1: Let , , It is a smooth convex function that is differentiable and satisfies the Lipshitz condition, i.e.:
[0084]
[0085] in, Describes the differential operator. Denotes the Lipschitz constant, which is greater than In the function The largest eigenvalue;
[0086] S3.2: Using the second-order Taylor expansion at point Approximate :
[0087]
[0088]
[0089] in, This represents the inner product operation. It is a with Irrelevant constants;
[0090] S3.3: The minimum value of the formula in S3.2 is obtained by the following formula:
[0091]
[0092] in, Indicates the step size;
[0093] S3.4: Consider Since it is continuous and not differentiable, the formula for S3.3 is corrected as follows:
[0094] ;
[0095] S3.5: First minimize using gradient descent Then, the solution is obtained using the proximal operator, i.e.:
[0096]
[0097] in, This represents the soft threshold operator, where the threshold is... ,Right now:
[0098]
[0099] in, represents the rectified linear unit, which is widely used as a non-linear activation function in deep learning;
[0100] S3.6: Correct the proximal operator in S3.5 only if all elements f are non-negative:
[0101]
[0102] in, Represents a non-negative soft thresholding function. This represents the threshold.
[0103] S4. Construct a deep algorithm unfolding network based on the iterative convergence threshold algorithm; refer to... Figure 2 ,include:
[0104] S4.1: Each layer in the deep algorithm unfolds the network, corresponding to an iterative step in the algorithm. The parameters in the iterative convergence threshold algorithm are updated through end-to-end training, i.e.:
[0105]
[0106] in, Represents the learnable weight parameters of each layer, and the adaptive coefficients. Represents learnable coefficients, bias vector Equal to the corresponding threshold, The activation function is consistent with the non-negativity constraint of the force vector, and the transfer matrix is... It is embedded into each layer as a fixed weight parameter;
[0107] S4.2: Set the expansion number K; set the expansion number K to 6;
[0108] Increasing the number of unfolds K can deepen the network and increase the number of parameters, thereby improving network performance; however, it can also increase the risk of overfitting when the dataset size is limited. Drawing on the design experience of classic networks, the number of unfolds K is set to 6.
[0109] S4.3: Reconstruct the impact force vector The initial value is set to Weight matrix of each layer Initialize to ; will be reconstructed The mean square error between the true force vector f and the actual force vector is used as the loss function, i.e.:
[0110]
[0111] in, The number of training samples. Force vector The size of the loss function is determined by using the Adam algorithm as the optimizer. Backpropagation is used to obtain parameter gradients, which are then used to update the weights in the network model, thereby reducing the loss function and improving model performance.
[0112] S5. Using the dataset described in step S1, train the deep learning algorithm network. To obtain the optimal network model, adopt an early stopping strategy. When the loss function value of the network on the validation set is continuously... If no decrease is observed in any batch, the network training process is terminated to prevent overfitting. The tolerance is set; the optimal network model has the best parameters for subsequent impact load identification.
[0113] S6. Using the response of the monitored object under impact load, the trained network model is tested for impact load identification. The input of the network is the response signal of the monitored object structure under the impact load to be identified, and the output is the impact load to be identified.
[0114] Compared with existing technologies, the method described in this invention utilizes an iterative convergence threshold algorithm to construct a deep learning network, which can simultaneously locate the impact load's position and reconstruct its time history even when the sensor layout is uncertain, thereby improving the impact load identification accuracy. It achieves accurate impact load identification with fewer iterations (K = 6), demonstrating superior computational speed, and also exhibits good noise resistance.
[0115] In one embodiment, the impact load is identified using a finite element model of a cartridge with a fixed bottom surface. Figure 3 As shown, the degree of freedom constraint on the bottom surface of the structure is zero. The material of the projectile is 2Cr13, the outer diameter of the cylinder is 253mm, and the height of the cylinder is 500mm. Taking the finite element model of this projectile as the research object, the process of identifying the impact load is as follows:
[0116] 1. Establish a structure on the cartridge case as follows Figure 3 The coordinate system shown represents 20 impact monitoring points (P1-P20) with a grid size of approximately 100 × 100 mm. Strain response signals were collected at points S1-S4 on the surface of the cartridge case. 108 different impact forces were applied to each monitoring point, generating a total of 2160 training samples. Each training sample consisted of the response signals (inputs) from four strain sensors and a combined force vector. Additionally, four impact forces were applied to each monitoring point for testing, generating a total of 80 test samples.
[0117] 2. Simulated impact loads were applied to P1-P20, and strain response signals were collected at points S1-S4 on the surface of the cartridge. Simultaneously, the time histories of the impact load and strain signals were recorded. To collect complete impact signals, based on the Nyquist sampling theorem, the sampling frequency was set to 10240 Hz, the sampling time to 25 ms, and the corresponding sampling length to 256. Figures 4(a) and 4(b) show the actual time histories of the impact load and the corresponding strain signals. The impact load location results and time histories reconstructions serve as comparison objects for the neural network recognition results of the interpretable algorithm of the non-negative sparse coding model.
[0118] 3. Solve the iterative convergence threshold algorithm for solving non-negative sparse coding models:
[0119] 3.1: Let , , It is a smooth convex function that is differentiable and satisfies the Lipshitz condition, i.e.:
[0120]
[0121] in, Describes the differential operator. Denotes the Lipschitz constant, which is greater than In the function The largest eigenvalue.
[0122] 3.2: Using the second-order Taylor expansion at the point Approximate :
[0123]
[0124]
[0125] in, This represents the inner product operation. It is a with Irrelevant constants.
[0126] The minimum value of the formula in 3.3:3.2 can be obtained from the following formula:
[0127]
[0128] in, Indicates the step size.
[0129] 3.4: Consideration Since it is continuous and not differentiable, the formula in 3.3 is corrected as follows:
[0130]
[0131] 3.5: We can first minimize it using gradient descent. Then, through the proximal operator, i.e.:
[0132]
[0133] in, This represents the soft threshold operator, where the threshold is... ,Right now:
[0134]
[0135] in, Represents a rectified linear unit, which is widely used as a non-linear activation function in deep learning.
[0136] 3.6: If all elements of f are non-negative, the proximal operator in 3.5 can be corrected as follows:
[0137]
[0138] in, Represents a non-negative soft thresholding function. This represents the threshold.
[0139] 4. Constructing deep learning networks based on iterative algorithms:
[0140] 4.1: Each layer in the network corresponds to an iterative step in the algorithm. The parameters in the algorithm are updated through end-to-end training, i.e.:
[0141]
[0142] in, Represents the learnable weight parameters of each layer, and the adaptive coefficients. Represents learnable coefficients, bias vector Equal to the corresponding threshold, The activation function is consistent with the non-negativity constraint of the force vector, and the transfer matrix is... It is embedded into each layer as a fixed weight parameter.
[0143] 4.2: Setting the number of unfolds K: Increasing the number of unfolds K can deepen the network and increase the number of parameters, thereby improving network performance; however, it also increases the risk of overfitting when the dataset size is limited. Based on experimental results and drawing on the design experience of classic networks, the number of unfolds K is set to 6.
[0144] 4.3: Reconstructing the impact force vector The initial value is set to Weight matrix of each layer Initialize to Utilizing the reconstructed The mean square error between the true force vector f and the actual force vector is used as the loss function, i.e.:
[0145]
[0146] in, The number of training samples. Force vector The size of the network is determined. The Adam algorithm is used as the optimizer, and the parameters in the network are updated through backpropagation. An early stopping strategy is used to prevent overfitting, and the tolerance is set to 50.
[0147] 5. Train the deep learning algorithm network using the existing dataset. To obtain the optimal network model, an early stopping strategy is adopted. The network stops when the loss function value on the validation set is continuously decreasing. If no decrease is observed in any batch, the network training process is terminated to prevent overfitting. The tolerance is set; the optimal network model has the best parameters for subsequent impact load identification.
[0148] 6. Using the response of the monitored object under impact load, the trained network model is tested for impact load identification. The input of the network is the response signal of the monitored object structure under the impact load to be identified, and the output is the impact load to be identified.
[0149] 7. To quantitatively analyze the accuracy of impact force identification results, three evaluation indicators were introduced: relative error (RE), peak relative error (PRE), and positioning accuracy (LA).
[0150]
[0151]
[0152]
[0153] Among them, and These are the actual impact force and the impact location, respectively. The impact force identified at the point. Of these three indicators... This reflects errors in the reconstruction of time history. Used to assess the accuracy of identifying peak force. Used to display the concentration of force recognition results. When When it equals 100%, it means that the impact force is identified only at the actual impact location, and the identification results for non-impact locations are all zero.
[0154] Figures 5(a) and 5(b) show the impact load identification results of the cartridge structure. Figures 5(a) and 5(b) show the network time history reconstruction results and the impact localization results based on the iterative algorithm, respectively. In the impact events shown in Figures 5(a) and 5(b), the relative error and peak relative error of the present invention are 6.532% and 0.13901%, respectively, accurately identifying the time history of the impact load and accurately locating the node where the impact load was applied.
[0155] In one embodiment, the present invention provides a non-negative sparse coding algorithm expansion and recognition system for impact loads on aerospace structures, the system comprising:
[0156] The first unit is used to collect and create a dataset containing impact load force signals and response signals from the monitored aerospace structures.
[0157] The second unit is used to construct a non-negative sparse coding model based on a dynamic model of impact load and a non-negative sparse prior.
[0158] The third unit is used to construct an iterative convergence threshold algorithm to solve the non-negative sparse coding model.
[0159] The fourth unit is used to construct a deep algorithm unfolding network based on the iterative convergence threshold algorithm;
[0160] The fifth unit involves training the deep learning algorithm network using the dataset to obtain a network model with optimal parameters.
[0161] The sixth unit uses the network model with optimal parameters to perform impact load identification tests and outputs the impact load to be identified.
[0162] In one embodiment, the present invention provides a computer-readable storage medium for storing a computer program configured to implement the method when invoked by a processor.
[0163] In one embodiment, the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor; wherein the processor implements the method when executing the program.
[0164] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment / mode or example is included in at least one embodiment / mode or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.
[0165] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0166] Those skilled in the art should understand that the above embodiments are merely for illustrating the present invention and are not intended to limit the scope of the invention. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present invention.
Claims
1. A method and system for identifying non-negative sparse coding algorithms for impact loads on aerospace structures, characterized in that, The method includes the following steps: S1. Collect and create a dataset containing the impact load force signal and the corresponding response signal data for the monitored aerospace structure; S2. Construct a non-negative sparse coding model based on a dynamic model of impact load and a non-negative sparse prior. S3. Construct an iterative convergence threshold algorithm for solving the non-negative sparse coding model; S4. Construct a deep algorithm unfolding network based on the iterative convergence threshold algorithm; S5. Use the dataset to train the deep algorithm unfolded network to obtain a network model with optimal parameters; S6. Use the network model with optimal parameters to perform impact load identification test and output the impact load to be identified.
2. The method according to claim 1, characterized in that, Preferably, in step S1, a dataset is created by conducting a hammer impact modal test on the monitored aerospace structure to obtain the transfer function S between the impact monitoring point and the response measurement point and the vibration response signal x of the measurement point.
3. The method according to claim 1, characterized in that, In step S2, the non-negative sparse coding model is: .
4. The method according to claim 1, characterized in that, Step S3 derives an iterative convergence threshold algorithm based on convex optimization and proximal gradient descent methods, which is used to solve the non-negative sparse coding model.
5. The method according to claim 4, characterized in that, Step S4 includes the following steps: S4.1: Update the parameters in the iterative convergence threshold algorithm through end-to-end training, that is: ， in, Represents the learnable weight parameters of each layer, and the adaptive coefficients. Represents learnable coefficients, bias vector Equal to the corresponding threshold, The activation function is consistent with the non-negativity constraint of the force vector, and the transfer matrix is... It is embedded into each layer as a fixed weight parameter; S4.2: Set the number of expansions K; S4.3: Reconstruct the impact force vector The initial value is set to Weight matrix of each layer Initialize to ; will be reconstructed The mean square error between the true force vector f and the actual force vector is used as the loss function, i.e.: ， in, The number of training samples. Force vector Size.
6. The method according to claim 5, characterized in that, In step S5, an early stopping strategy is adopted, whereby the network stops stopping when the loss function value on the validation set is continuously... If no decrease occurs in any batch, the network training process is terminated to obtain a network model with optimal parameters.
7. The method according to claim 6, characterized in that, In step S6, the response signal of the monitored object under the action of the impact load to be identified is collected as the input of the network. The network model with the optimal parameters is used to identify the impact load and output the impact load to be identified.
8. A non-negative sparse coding algorithm expansion and recognition system for impact loads on aerospace structures, characterized in that, The system includes: The first unit is used to collect and create a dataset containing impact load force signals and response signals from the monitored aerospace structures. The second unit is used to construct a non-negative sparse coding model based on a dynamic model of impact load and a non-negative sparse prior. The third unit is used to construct an iterative convergence threshold algorithm to solve the non-negative sparse coding model. The fourth unit is used to construct a deep algorithm unfolding network based on the iterative convergence threshold algorithm; The fifth unit involves training the deep learning algorithm network using the dataset to obtain a network model with optimal parameters. The sixth unit uses the network model with optimal parameters to perform impact load identification tests and outputs the impact load to be identified.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program configured to implement the method of any one of claims 1-7 when invoked by a processor.
10. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor; wherein, when the processor executes the program, it implements the method of any one of claims 1-7.