Dynamic data correlation analysis of underwater hull structures and intelligent prediction method of acoustic radiation

Abstract

The present application relates to a method for dynamic data association analysis and intelligent prediction of acoustic radiation of underwater shell structures. The method comprises: constructing a Toeplitz matrix; performing singular value decomposition on the Toeplitz matrix, extracting a state matrix from an observation matrix, performing eigenvalue decomposition, and calculating the modal frequency of the underwater shell structure; constructing a vibration characteristic matrix based on the modal frequency of the underwater shell structure; performing dimension expansion and normalization on the vibration characteristic matrix to construct an acoustic pressure characteristic matrix; constructing an acoustic-vibration transfer network model; determining the number of nodes in the input layer based on the dimension of the normalized vibration characteristic matrix; determining the number of nodes in the output layer based on the dimension of the acoustic pressure characteristic matrix; training the acoustic-vibration transfer network model based on a pre-designed loss function and collected data, and outputting an intelligent prediction result of acoustic radiation using the trained acoustic-vibration transfer network model. This method can improve the accuracy of intelligent prediction of acoustic radiation.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method for dynamic data correlation analysis and intelligent prediction of acoustic radiation of underwater shell structures. Background Art

[0002] The radiated noise of underwater vehicles is the most important indicator of their safety and stealth, and is directly related to their survivability on the battlefield. During navigation, the radiated noise of underwater vehicles primarily originates from the structural vibrations generated by their internal mechanical equipment. Its spectral characteristics often exhibit low-frequency lines, a characteristic signature for ship target identification. Furthermore, low-frequency sound waves have low attenuation in the marine environment and travel long distances, making the vehicle more susceptible to detection and attack. Therefore, in order to accurately predict the low-frequency radiated noise level of a vehicle and monitor the acoustic radiation levels generated by its vibrations in real time, underwater vehicle acoustic radiation prediction technology is particularly important.

[0003] In engineering, acoustic field prediction is often achieved by real-time monitoring of aircraft structural vibration information, combined with accurate and efficient transfer prediction models. However, this method fails to account for the complex and variable nature of multi-source excitation combinations during actual flight, making it prone to mismatches between simulated acoustic radiation levels and actual operating conditions, resulting in reduced acoustic radiation prediction accuracy. Summary of the Invention

[0004] Based on this, it is necessary to provide a method for dynamic data correlation analysis of underwater shell structures and intelligent prediction of acoustic radiation to address the above technical problems, which can improve the accuracy of intelligent prediction of acoustic radiation.

[0005] A method for dynamic data correlation analysis and intelligent prediction of acoustic radiation of underwater hull structures, comprising:

[0006] Perform dynamic characteristic tests on underwater hull structures, collecting vibration response data under hammer impact, vibration data of the hull structure under operational excitation, and sound pressure data at specific locations. The vibration response data under hammer impact includes the vibration response time domain sequence.

[0007] The covariance matrix is ​​calculated based on the vibration response time domain sequence and the Toeplitz matrix is ​​constructed using the covariance matrix. The Toeplitz matrix is ​​subjected to singular value decomposition to obtain the observation matrix. The state matrix is ​​extracted from the observation matrix and then subjected to eigenvalue decomposition. The modal frequencies of the underwater shell structure are calculated using the obtained eigenvalues.

[0008] A vibration characteristic matrix is ​​constructed based on the vibration data of the shell structure under operational excitation and the modal frequency of the underwater shell structure; the vibration characteristic matrix is ​​dimensionally expanded and normalized to obtain a normalized vibration characteristic matrix; and an acoustic pressure characteristic matrix is ​​constructed using the acoustic pressure data;

[0009] Construct an acoustic-vibration transmission network model; the acoustic-vibration transmission network model includes an input layer, a two-layer hidden layer, and an output layer; the number of nodes in the input layer is determined according to the dimension of the normalized vibration characteristic matrix; the number of nodes in the output layer is determined using the dimension of the sound pressure characteristic matrix;

[0010] The acoustic-vibration transfer network model is trained based on the pre-designed loss function and the collected vibration response data, vibration data, and sound pressure data. The trained acoustic-vibration transfer network model is used to predict the input structural surface vibration data and output the intelligent sound radiation prediction results.

[0011] In one embodiment, calculating a covariance matrix based on a vibration response time domain sequence and constructing a Toeplitz matrix using the covariance matrix includes:

[0012] The covariance matrix is ​​calculated based on the vibration response time domain series:

[0013]

[0014] The Toeplitz matrix is ​​constructed using the covariance matrix as

[0015]

[0016] Among them, j is the delay point, y i is the time domain sequence of the vibration response at time i, n represents the number of points in the time domain sequence, and the superscript T represents the transpose operation.

[0017] In one embodiment, singular value decomposition is performed on the Toeplitz matrix to obtain a measurement matrix, including:

[0018] Perform singular value decomposition on the Toeplitz matrix and obtain the observation matrix as

[0019] T=USV T

[0020] P=US 1 / 2

[0021] Among them, S is a diagonal matrix composed of the main singular values, U and V are the left and right vector matrices corresponding to each singular value in the S matrix, and T is the Toeplitz matrix.

[0022] In one embodiment, the state matrix is ​​extracted from the observation matrix as

[0023]

[0024] Among them, P1 and P2 are matrices consisting of the first r rows and the last r rows in the observation matrix P, respectively. Represents the pseudoinverse operation of a matrix.

[0025] In one embodiment, calculating the modal frequency of the underwater shell structure using the obtained eigenvalues ​​includes:

[0026] The modal frequency of the underwater shell structure is calculated using the obtained eigenvalues:

[0027]

[0028] Among them, λ k is the kth order eigenvalue of the discrete time system, Δt is the discrete time interval, u k Represents the eigenvalues ​​of even-numbered items in the state matrix.

[0029] In one embodiment, a vibration characteristic matrix is ​​constructed based on vibration data of the hull structure under operational excitation and the modal frequency of the underwater hull structure, including:

[0030] Extract the amplitudes A1~A1 of the rotor 1~p times the frequency from the m groups of vibration response data under running excitation. p , and the first q modal frequencies f1,f2,...,f q The corresponding vibration energy E1~E q , construct the vibration characteristic matrix as

[0031]

[0032] In one embodiment, the vibration feature matrix is ​​dimensionally expanded and normalized to obtain a feature matrix, further comprising:

[0033] The vibration feature matrix is ​​expanded and normalized to obtain the feature matrix:

[0034]

[0035] Among them, F ij represents the eigenvalue of row i and column j, F jmin represents the minimum value of the jth column of the matrix, F jmax Represents the maximum value of the j-th column of the matrix.

[0036] In one embodiment, constructing a sound pressure feature matrix using sound pressure data includes:

[0037] Calculate the effective sound pressure values ​​v1~v at l positions from m groups of sound pressure data under running excitation N , the sound pressure characteristic matrix is

[0038]

[0039] In one embodiment, the formula for determining the number of hidden layer nodes is:

[0040]

[0041] Where d represents the number of input layer nodes, l represents the number of output layer nodes, and a is a constant.

[0042] In one embodiment, the pre-designed loss function is

[0043]

[0044] Among them, m is the number of data, v i is the true value of sound pressure, is the predicted sound pressure value.

[0045] The above-mentioned underwater shell structure dynamic data correlation analysis and sound radiation intelligent prediction method, this application applies the deep learning model to the sound radiation prediction of a certain type of underwater equipment, starting from the data itself to explore the transfer relationship between vibration and sound field, and realize high-precision transfer prediction from structural vibration to sound field. It uses the structural surface vibration data as input and the sound field data as output, and can realize high-precision prediction from structural vibration to sound field under multiple excitation conditions. This application focuses on the rational processing of data, starting from the data itself, avoiding the analysis of the complexity of the underwater structure. By training a large number of measured data samples, the prediction model is obtained and the prediction of the sound field can be realized. It has certain engineering applicability. Under the experimental environment, the error between the prediction value and the test result does not exceed 3dB. Compared with traditional sound field prediction technology, this application has greater application value in the prediction of underwater vehicle sound radiation. First, this application ignores considerations of physical characteristics such as complex structures, multi-source excitation, and nonlinear vibration noise, focusing instead on solving the transformation relationship between data to achieve sound field prediction. Second, a large amount of data has been accumulated from underwater vehicle calibration, noise index assessment, and noise source level testing between two ships in shallow and near-shore waters. This application efficiently utilizes existing test data and explores the transmission relationship between vibration and radiated noise, which is an inevitable requirement for guiding real-time monitoring and prediction of vehicle noise. It is expected to reduce the significant difference between the predicted values ​​of traditional methods and the actual flight noise level, which is of great significance to improving the overall design level of underwater vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 1. A schematic flow chart of a method for dynamic data correlation analysis and intelligent acoustic radiation prediction of an underwater shell structure in one embodiment;

[0047] Figure 2 FIG. 1 is a schematic diagram of a BP neural network structure containing two hidden layers in one embodiment. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0049] In one embodiment, Figure 1 As shown, a method for dynamic data association analysis and intelligent prediction of acoustic radiation of underwater shell structures is provided, comprising the following steps:

[0050] Step 102 , performing a dynamic characteristic test on the underwater shell structure, collecting vibration response data under a force hammer, vibration data of the shell structure under operating excitation, and sound pressure data at a specific position; the vibration response data under the force hammer includes a vibration response time domain sequence.

[0051] Dynamic testing of underwater hull structures was conducted, collecting vibration response data under impact with a force hammer, as well as vibration data and sound pressure data at specific locations under operational excitation. The number of measurement points was s and l, respectively. By exploring the transfer relationship between vibration and the acoustic field from the data itself, high-precision predictions of the transfer from structural vibration to the acoustic field were achieved. This approach, while ignoring considerations of physical characteristics such as the complex structure of the vehicle, multi-source excitation, and nonlinear vibration and noise, focused on solving the transformation relationship between the data to achieve acoustic field predictions. By focusing on the data itself, analysis of the complexity of the underwater structure was avoided, resulting in lower computational complexity and higher prediction accuracy.

[0052] Step 104 , calculates the covariance matrix based on the vibration response time domain sequence and constructs a Toeplitz matrix using the covariance matrix; performs singular value decomposition on the Toeplitz matrix to obtain a measurement matrix; extracts the state matrix from the measurement matrix and performs eigenvalue decomposition, and calculates the modal frequency of the underwater shell structure using the obtained eigenvalues.

[0053] According to the time domain series of vibration response under hammer excitation, a covariance matrix R is first constructed: j ,

[0054]

[0055] Where j is the number of delay points. i is the time domain vibration response at time i, and n represents the number of time domain sequence points.

[0056] Using formula (1) to construct the Toeplitz matrix T is

[0057]

[0058] Perform singular value decomposition on formula (2) and obtain the observation matrix P as follows

[0059]

[0060] Among them, S is a diagonal matrix composed of main singular values, and U and V are the left and right vector matrices corresponding to each singular value in the S matrix.

[0061] Extract the first r rows and the last r rows of the observation matrix P and calculate the state matrix A as

[0062]

[0063] Among them, P1 and P2 are matrices consisting of the first r rows and the last r rows in the observation matrix P, respectively. Represents the pseudoinverse operation of a matrix.

[0064] The eigenvalue decomposition of formula (4) is as follows

[0065] A=ΨΛΨ -1 (5)

[0066] Where, Λ=diag(u k ), u k represents the eigenvalue of the even-numbered items in the state matrix, and Ψ is the eigenvector matrix of the system.

[0067] According to the eigenvalues ​​and vector matrix of the discrete system, the modal frequency f under the continuous time structural system can be further calculated. k as follows.

[0068]

[0069] λ k is the kth order eigenvalue of the discrete time system, and Δt is the discrete time interval.

[0070] Step 106: construct a vibration characteristic matrix based on the vibration data of the shell structure under operating excitation and the modal frequency of the underwater shell structure; perform dimension expansion and normalization on the vibration characteristic matrix to obtain a normalized vibration characteristic matrix; and construct a sound pressure characteristic matrix using the sound pressure data.

[0071] Extract the amplitudes A1~A1 of the rotor 1~p times the frequency from the m groups of vibration response data under running excitation. p , and the vibration energy E1~E corresponding to the first q-order modal frequencies q , construct the vibration characteristic matrix F as follows.

[0072]

[0073] Combined with s vibration channels, the dimension of the feature matrix F will be expanded to m×s(p+q). According to formula (8), the feature matrix is ​​normalized and the output is the feature matrix G, which is simplified to m×(p+q).

[0074]

[0075] Calculate the effective sound pressure values ​​v1~v at l positions from m groups of sound pressure data under running excitation N , the dimension of the sound pressure feature matrix V is m×l,

[0076]

[0077] Step 108, constructing an acoustic-vibration transfer network model; the acoustic-vibration transfer network model includes an input layer, a double hidden layer, and an output layer; the number of nodes in the input layer is determined according to the dimension of the normalized vibration characteristic matrix; the number of nodes in the output layer is determined using the dimension of the sound pressure characteristic matrix.

[0078] Construct an acoustic-vibration transmission network model such as Figure 2 As shown, the network has four layers, including an input layer, two hidden layers, and an output layer. The number of input layer nodes is the dimension of G, i.e., the d-class feature, d = p + q. The number of output layer nodes is the dimension of the sound pressure feature V, i.e., the sound pressure values ​​at N locations. The number of hidden layer nodes can usually be calculated using formula (10). The hidden layer transfer function uses the logsig function, and the output layer uses a linear function.

[0079]

[0080] Step 110: Train the acoustic-vibration transfer network model based on the pre-designed loss function and the collected vibration response data, vibration data, and sound pressure data, use the trained acoustic-vibration transfer network model to predict the input structural surface vibration data, and output the sound radiation intelligent prediction result.

[0081] The network training function uses a gradient descent algorithm, and parameters such as the learning rate, network training times, and target error are set according to specific needs. A subset of collected vibration response data, vibration data, and sound pressure data is selected as the training set and a pre-designed loss function is used to train the network model. The loss function value is used as a metric to reflect model accuracy; larger loss function values ​​indicate poorer prediction results.

[0082] The above-mentioned underwater shell structure dynamic data correlation analysis and sound radiation intelligent prediction method, this application applies the deep learning model to the sound radiation prediction of a certain type of underwater equipment, starting from the data itself to explore the transfer relationship between vibration and sound field, and realize high-precision transfer prediction from structural vibration to sound field. It uses the structural surface vibration data as input and the sound field data as output, and can realize high-precision prediction from structural vibration to sound field under multiple excitation conditions. This application focuses on the rational processing of data, starting from the data itself, avoiding the analysis of the complexity of the underwater structure. By training a large number of measured data samples, the prediction model is obtained and the prediction of the sound field can be realized. It has certain engineering applicability. Under the experimental environment, the error between the prediction value and the test result does not exceed 3dB. Compared with traditional sound field prediction technology, this application has greater application value in the prediction of underwater vehicle sound radiation. First, this application ignores considerations of physical characteristics such as complex structures, multi-source excitation, and nonlinear vibration noise, focusing instead on solving the transformation relationship between data to achieve sound field prediction. Second, a large amount of data has been accumulated from underwater vehicle calibration, noise index assessment, and noise source level testing between two ships in shallow and near-shore waters. This application efficiently utilizes existing test data and explores the transmission relationship between vibration and radiated noise, which is an inevitable requirement for guiding real-time monitoring and prediction of vehicle noise. It is expected to reduce the significant difference between the predicted values ​​of traditional methods and the actual flight noise level, which is of great significance to improving the overall design level of underwater vehicles.

[0083] In one embodiment, calculating a covariance matrix based on a vibration response time domain sequence and constructing a Toeplitz matrix using the covariance matrix includes:

[0084] The covariance matrix is ​​calculated based on the vibration response time domain series:

[0085]

[0086] The Toeplitz matrix is ​​constructed using the covariance matrix as

[0087]

[0088] Among them, j is the delay point, y i is the time domain sequence of the vibration response at time i, n represents the number of points in the time domain sequence, and the superscript T represents the transpose operation.

[0089] In one embodiment, singular value decomposition is performed on the Toeplitz matrix to obtain a measurement matrix, including:

[0090] Perform singular value decomposition on the Toeplitz matrix and obtain the observation matrix as

[0091] T=USV T

[0092] P=US1 / 2

[0093] Among them, S is a diagonal matrix composed of the main singular values, U and V are the left and right vector matrices corresponding to each singular value in the S matrix, and T is the Toeplitz matrix.

[0094] In one embodiment, the state matrix is ​​extracted from the observation matrix as

[0095]

[0096] Among them, P1 and P2 are matrices consisting of the first r rows and the last r rows in the observation matrix P, respectively. Represents the pseudoinverse operation of a matrix.

[0097] In one embodiment, calculating the modal frequency of the underwater shell structure using the obtained eigenvalues ​​includes:

[0098] The modal frequency of the underwater shell structure is calculated using the obtained eigenvalues:

[0099]

[0100] Among them, λ k is the kth order eigenvalue of the discrete time system, Δt is the discrete time interval, u k Represents the eigenvalues ​​of even-numbered items in the state matrix.

[0101] In one embodiment, a vibration characteristic matrix is ​​constructed based on vibration data of the hull structure under operational excitation and the modal frequency of the underwater hull structure, including:

[0102] Extract the amplitudes A1~A1 of the rotor 1~p times the frequency from the m groups of vibration response data under running excitation. p , and the first q modal frequencies f1,f2,...,f q The corresponding vibration energy E1~E q , construct the vibration characteristic matrix as

[0103]

[0104] In one embodiment, the vibration feature matrix is ​​dimensionally expanded and normalized to obtain a feature matrix, further comprising:

[0105] The vibration feature matrix is ​​expanded and normalized to obtain the feature matrix:

[0106]

[0107] Among them, F ij represents the eigenvalue of row i and column j, F jmin represents the minimum value of the jth column of the matrix, Fjmax Represents the maximum value of the j-th column of the matrix.

[0108] In one embodiment, constructing a sound pressure feature matrix using sound pressure data includes:

[0109] Calculate the effective sound pressure values ​​v1~v at l positions from m groups of sound pressure data under running excitation N , the sound pressure characteristic matrix is

[0110]

[0111] In one embodiment, the formula for determining the number of hidden layer nodes is:

[0112]

[0113] Where d represents the number of input layer nodes, l represents the number of output layer nodes, and a is a constant.

[0114] In one embodiment, the pre-designed loss function is

[0115]

[0116] Among them, m is the number of data, v i is the true value of sound pressure, is the predicted sound pressure value.

[0117] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0118] The technical features of the above embodiments can 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.

[0119] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for dynamic data correlation analysis and intelligent prediction of acoustic radiation of underwater shell structures, characterized in that: The method comprises: Performing a dynamic characteristic test on the underwater hull structure, collecting vibration response data under a force hammer, vibration data of the hull structure under operating excitation, and sound pressure data at a specific location; the vibration response data under the force hammer includes a vibration response time domain sequence; Calculating a covariance matrix based on the vibration response time domain sequence and constructing a Toeplitz matrix using the covariance matrix; performing singular value decomposition on the Toeplitz matrix to obtain a measurement matrix; extracting a state matrix from the measurement matrix and performing eigenvalue decomposition thereon, and calculating the modal frequency of the underwater shell structure using the obtained eigenvalues; Constructing a vibration characteristic matrix based on the vibration data of the shell structure under the operating excitation and the modal frequency of the underwater shell structure; performing dimension expansion and normalization processing on the vibration characteristic matrix to obtain a normalized vibration characteristic matrix; and constructing a sound pressure characteristic matrix using the sound pressure data; Constructing an acoustic-vibration transfer network model; the acoustic-vibration transfer network model includes an input layer, a double hidden layer, and an output layer; determining the number of nodes in the input layer according to the dimension of the normalized vibration characteristic matrix; and determining the number of nodes in the output layer according to the dimension of the sound pressure characteristic matrix; The acoustic-vibration transfer network model is trained according to a pre-designed loss function and the collected vibration response data, vibration data and sound pressure data. The trained acoustic-vibration transfer network model is used to predict the input structural surface vibration data and output the sound radiation intelligent prediction result.

2. The method according to claim 1, characterized in that Calculating a covariance matrix according to the vibration response time domain sequence and constructing a Toeplitz matrix using the covariance matrix includes: The covariance matrix is ​​calculated based on the vibration response time domain sequence: The Toeplitz matrix is ​​constructed using the covariance matrix as Among them, j is the delay point, y i is the time domain sequence of the vibration response at time i, n represents the number of points in the time domain sequence, and the superscript T represents the transpose operation.

3. The method according to claim 1, characterized in that Perform singular value decomposition on the Toeplitz matrix to obtain a measurement matrix, including: Perform singular value decomposition on the Toeplitz matrix and obtain the observation matrix: T=USV T P=US 1 / 2 Among them, S is a diagonal matrix composed of the main singular values, U and V are the left and right vector matrices corresponding to each singular value in the S matrix, and T is the Toeplitz matrix.

4. The method according to claim 1, wherein The state matrix is ​​extracted from the observation matrix as Among them, P1 and P2 are matrices consisting of the first r rows and the last r rows in the observation matrix P, respectively. Represents the pseudoinverse operation of a matrix.

5. The method according to claim 1, characterized in that The modal frequencies of the underwater shell structure are calculated using the obtained eigenvalues, including: The modal frequency of the underwater shell structure is calculated using the obtained eigenvalues: Among them, λ k is the kth order eigenvalue of the discrete time system, Δt is the discrete time interval, u k Represents the eigenvalues ​​of even-numbered items in the state matrix.

6. The method according to claim 1, wherein A vibration characteristic matrix is ​​constructed according to the vibration data of the shell structure under the operational excitation and the modal frequency of the underwater shell structure, including: Extract the amplitudes A1~A1 of the rotor 1~p times the frequency from the m groups of vibration response data under running excitation. p , and the first q modal frequencies f1,f2,...,f q The corresponding vibration energy E1~E q , construct the vibration characteristic matrix as 7. The method according to claim 1, characterized in that Performing dimension expansion and normalization processing on the vibration feature matrix to obtain a feature matrix, further comprising: The vibration characteristic matrix is ​​dimensionally expanded and normalized to obtain the characteristic matrix: Among them, F ij represents the eigenvalue of row i and column j, F jmin represents the minimum value of the jth column of the matrix, F jmax Represents the maximum value of the j-th column of the matrix.

8. The method according to claim 1, characterized in that Constructing a sound pressure feature matrix using the sound pressure data includes: Calculate the effective sound pressure values ​​v1~v at l positions from m groups of sound pressure data under running excitation N , the sound pressure characteristic matrix is 9. The method according to claim 1, characterized in that The formula for determining the number of hidden layer nodes is: Where d represents the number of input layer nodes, l represents the number of output layer nodes, and a is a constant.

10. The method according to claim 1, characterized in that The pre-designed loss function is Among them, m is the number of data, v i is the true value of sound pressure, is the predicted sound pressure value.

Images