Real-time intelligent ship underwater radiation noise forecasting method based on virtual and real data fusion

Abstract

The invention discloses a real-time intelligent ship underwater radiation noise forecasting method based on virtual and real data fusion. The method comprises the following steps: establishing an acoustic elasticity calculation model of a real ship; defining an excitation point, and randomly applying a group of normal frequency sweeping forces with different amplitudes; solving and calculating vibration sound radiation; respectively reading a vibration acceleration amplitude spectrum of a measuring point corresponding to each excitation point; reading a radiation noise result of the whole ship, and outputting the radiation noise result to a file sound source level-working condition; calculating results under different working conditions until the calculated working condition number is equal to the set working condition number M; forming a model training set by the output acceleration amplitude spectrums and the radiation noise sound source level results of the M working conditions; a sample number, a side point number and a frequency point number for BP neural network training are defined, a result of a training set is read, data are processed, and input and output matrixes for BP neural network training are generated; parameters of BP neural network training are defined, and real-time intelligent forecasting of the underwater radiation noise of the ship based on virtual and real data fusion is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of ship underwater radiation noise prediction, and in particular to a real-time intelligent prediction method for ship underwater radiation noise based on virtual and real data fusion. Background Art

[0002] As a core indicator of ship stealth performance and marine environmental assessment, the accuracy of ship underwater radiated noise prediction directly impacts ship design optimization and noise control strategies. Traditional prediction methods rely primarily on numerical models or semi-empirical formulas. The former employs numerical methods such as finite element or finite element-boundary element methods to analyze the structure-flow coupling mechanism. However, due to the large computational complexity of actual ship numerical models, real-time prediction is difficult to achieve. The latter, while simplifying calculations based on statistical laws, is limited by the assumption of structural similarity and lacks generalization capabilities.

[0003] With the development of online ship vibration monitoring technology, high-precision vibration data has provided a new approach for real-time noise prediction. However, existing research is plagued by a widespread disconnect between virtual and real data. Theoretical models struggle to capture complex operating conditions, while purely data-driven approaches are prone to overfitting due to insufficient measured data. Hybrid modeling, integrating physical mechanism models with real-time monitoring, has become a key approach to addressing this bottleneck. Summary of the Invention

[0004] In response to the shortcomings of the above-mentioned existing production technology, the applicant provides a real-time intelligent prediction method for ship underwater radiated noise based on virtual and real data fusion. Based on the ship three-dimensional acoustic elasticity calculation and analysis software Thafts-acoustic, a numerical calculation model of the hull's acoustic elasticity is established. The acceleration spectrum response of the measuring point on the base is used as input, and the hull radiated noise is used as output. Through the BP neural network algorithm, an intelligent relationship between the input and output is established, thereby realizing real-time intelligent prediction of ship underwater radiated noise based on virtual and real data fusion.

[0005] The technical solutions adopted in the present invention are as follows:

[0006] A real-time intelligent prediction method for underwater ship radiated noise based on virtual and real data fusion includes the following steps:

[0007] S1: Establish an acoustic-elastic calculation model of a real ship based on the ship's three-dimensional acoustic-elastic calculation software Thafts-acoustic;

[0008] S2: Define the excitation points, the set of all excitation points, and name and number each excitation point;

[0009] S3: Randomly apply a set of normal white noise sweep excitation forces with different amplitudes;

[0010] S4: Calculate the vibration and acoustic radiation under this set of excitation forces based on the ship's three-dimensional acoustic elasticity calculation software Thafts-acoustic;

[0011] S5: Read the vibration acceleration amplitude spectrum of each excitation point corresponding to the measuring point respectively;

[0012] S6: Read the radiation noise results of the entire ship and output them to the file sound source level-operating condition;

[0013] S7: Repeat the operations of steps S3-S6 to calculate the results under different working conditions until the number of calculated working conditions is equal to the set number of working conditions M;

[0014] S8: The output acceleration amplitude spectra and radiated noise source level results of M working conditions are combined into a model training set;

[0015] S9: Define the number of samples, side points and frequency points used for BP neural network training, read the results of the training set, and process the data to generate the input and output matrices for BP neural network training;

[0016] S10: Define the parameters for BP neural network training and carry out training of real-time intelligent forecasting model;

[0017] S11: Based on the trained real-time intelligent prediction model and the online monitored vibration acceleration amplitude spectrum data, the real-time intelligent prediction of ship underwater radiated noise based on virtual and real data fusion is realized.

[0018] As a further improvement of the above technical solution:

[0019] In S2, the connection point between the vibration device and the base is preferably used as the excitation point, and each excitation point is named A1 to AN, where 1 to N corresponds to the measurement point number.

[0020] In S5, the vibration acceleration amplitude spectrum of the corresponding measuring point of each excitation point A1~AN is read respectively and output to the A1_working condition J.txt~AN_working condition J.txt files respectively, where J is the working condition number and the initial number is 1.

[0021] In S6, output to the file Sound Source Level-Working Condition J.

[0022] A forward excitation-response correlation method is used: a swept-frequency excitation force is applied at the selected measuring point, and the radiated noise under the corresponding working condition is calculated using the acoustoelasticity software Thafts-acoustic. The vibration acceleration response at the measuring point is simultaneously collected, and an intelligent mapping relationship between acceleration and radiated noise is constructed based on this. This avoids the inverse solution of the frequency response function and achieves higher calculation accuracy.

[0023] By randomly adjusting the excitation force amplitude at the measuring point in the acoustic-elasticity software to generate different sample data sets, there is no need to repeat the modal analysis, fluid-solid contact boundary condition calculation, wet surface source strength and generalized hydrodynamic coefficient calculation in the acoustic-elasticity model calculation process. This greatly improves the sample generation efficiency and the convergence speed of intelligent model training.

[0024] The beneficial effects of the present invention are as follows:

[0025] (1) The present invention addresses the difficulty in engineering implementation caused by the need for densely arranged measuring points on the sound radiation surface in traditional methods. It proposes a mechanism for optimizing key measuring points where the vibration equipment is connected to the base, constructs an intelligent prediction model of measuring point acceleration-radiation noise based on a BP neural network, and directly establishes an intelligent association between the vibration signals of limited measuring points and underwater radiation noise, thus breaking through the bottleneck of missing sound radiation energy due to incomplete coverage of measuring points in traditional methods.

[0026] (2) To address the ill-posed issues in the traditional frequency response function matrix inversion process, this paper innovatively proposes a forward excitation-response correlation method: a swept frequency excitation force is applied at a selected measuring point, the radiated noise under the corresponding working condition is calculated using the acoustic elasticity software Thafts-acoustic, and the vibration acceleration response at the measuring point is simultaneously collected. Based on this, an intelligent mapping relationship between acceleration and radiated noise is constructed, avoiding the inverse solution of the frequency response function and achieving higher calculation accuracy.

[0027] (3) The present invention proposes a design parameter perturbation enhanced learning strategy: the excitation force amplitude of the measuring point can be randomly adjusted in the acoustic elasticity software to generate different sample data sets. There is no need to repeat the modal analysis, fluid-solid contact boundary condition calculation, wet surface source strength and generalized hydrodynamic coefficient calculation in the acoustic elasticity model calculation process. The sample generation efficiency and the convergence speed of the intelligent model training are greatly improved.

[0028] (4) The present invention focuses on the dynamic correlation characteristics between ship structure vibration and underwater radiation noise. By integrating the radiation noise numerical prediction model based on the three-dimensional acoustic elasticity theory of ships with the vibration data of online monitoring of the ship's navigation status, a data-physics dual-driven noise prediction framework is constructed to solve the problem of insufficient noise prediction accuracy of traditional single models or data-driven methods under complex ship working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a calculation flow chart of the present invention.

[0030] Figure 2 This is the acoustic-elastic calculation model diagram of the entire ship (the finite element calculation model of the entire ship) according to the present invention.

[0031] Figure 3 This is the acoustic elastic calculation model diagram of the whole ship (external flow domain wet surface element model) of the present invention.

[0032] Figure 4 This is the acoustic elastic calculation model diagram of the whole ship (inner basin wet surface element model) of the present invention.

[0033] Figure 5 This is a measurement point diagram of the internal base excitation point of the present invention.

[0034] Figure 6 This is a comparison and verification chart of the intelligent forecast accuracy of the present invention.

[0035] Figure 7 This is a diagram of the intelligent prediction results of ship radiation noise based on online monitoring data in the present invention. DETAILED DESCRIPTION

[0036] The specific embodiments of the present invention will be described below with reference to the accompanying drawings.

[0037] like Figure 1-Figure 7 As shown, the real-time intelligent prediction method of underwater ship radiated noise based on virtual and real data fusion of this embodiment includes the following steps:

[0038] S1: Establish an acoustic-elastic calculation model of a real ship based on the ship's three-dimensional acoustic-elastic calculation software Thafts-acoustic;

[0039] S2: Define the excitation points, the set of all excitation points, and name and number each excitation point;

[0040] S3: Randomly apply a set of normal white noise sweep excitation forces with different amplitudes;

[0041] S4: Calculate the vibration and acoustic radiation under this set of excitation forces based on the ship's three-dimensional acoustic elasticity calculation software Thafts-acoustic;

[0042] S5: Read the vibration acceleration amplitude spectrum of each excitation point corresponding to the measuring point respectively;

[0043] S6: Read the radiation noise results of the entire ship and output them to the file sound source level-operating condition;

[0044] S7: Repeat the operations of steps S3-S6 to calculate the results under different working conditions until the number of calculated working conditions is equal to the set number of working conditions M;

[0045] S8: The output acceleration amplitude spectra and radiated noise source level results of M working conditions are combined into a model training set;

[0046] S9: Define the number of samples, side points and frequency points used for BP neural network training, read the results of the training set, and process the data to generate the input and output matrices for BP neural network training;

[0047] S10: Define the parameters for BP neural network training and carry out training of real-time intelligent forecasting model;

[0048] S11: Based on the trained real-time intelligent prediction model and the online monitored vibration acceleration amplitude spectrum data, the real-time intelligent prediction of ship underwater radiated noise based on virtual and real data fusion is realized.

[0049] This embodiment provides a real-time intelligent prediction method for underwater ship radiated noise based on virtual and real data fusion. The calculation process is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0050] 101: Establish an acoustic-elastic calculation model of a real ship based on the ship's three-dimensional acoustic-elastic calculation software Thafts-acoustic;

[0051] 102: The connection point between the vibration device and the base is preferably used as the excitation point. Each excitation point is named A1 to AN, and 1 to N corresponds to the measurement point number;

[0052] 103: Randomly apply a set of normal white noise sweep excitation forces with different amplitudes, such as Figure 5 As shown;

[0053] 104: Calculate the vibration and acoustic radiation under this set of excitation forces based on the ship's three-dimensional acoustic elasticity calculation software Thafts-acoustic;

[0054] 105: Read the vibration acceleration amplitude spectrum of the corresponding measuring points of the excitation points A1 to AN respectively, and output them to the A1_working condition J.txt to AN_working condition J.txt files respectively, where J is the working condition number and the initial number is 1;

[0055] 106: Read the radiation noise results of the entire ship and output them to the file sound source level-condition J;

[0056] 107: Repeat steps 103-106 to calculate results under different working conditions until the calculated working condition number J is equal to the set working condition number M;

[0057] 108: The output acceleration amplitude spectra and radiated noise source level results of M working conditions are used to form a model training set;

[0058] 109: Define the number of samples, side points and frequency points used for BP neural network training, read the results of the training set, and process the data to generate the input and output matrices for BP neural network training;

[0059] 110: Define the parameters for BP neural network training and carry out training of real-time intelligent forecasting model;

[0060] 111: Based on the trained real-time intelligent prediction model and the online monitored vibration acceleration amplitude spectrum data, the real-time intelligent prediction of ship underwater radiated noise based on virtual and real data fusion is realized, such as Figure 7 shown.

[0061] This embodiment has three major innovations: measurement point optimization configuration, forward excitation response modeling, and design parameter disturbance enhancement. It overcomes the difficulties of traditional methods such as measurement point redundancy, instability in the inverse solution of excitation force, and poor adaptability to working conditions. It can achieve high-precision real-time intelligent prediction of ship underwater radiated noise based on virtual and real data fusion with a small number of samples (such as Figure 6 As shown in the figure, the model verification results of 10 samples are given, with an average error of 0.86dB and a maximum error of 6.3dB), providing core technical support for ship stealth performance evaluation and low-noise navigation control.

[0062] The above description is an explanation of the present invention, not a limitation of the present invention. The scope of the present invention is defined in the claims. Any modifications may be made within the scope of protection of the present invention.

Claims

1. A real-time intelligent prediction method for underwater ship radiated noise based on virtual and real data fusion, characterized by: The steps are as follows: S1: Establish an acoustic-elastic calculation model of a real ship based on the ship's three-dimensional acoustic-elastic calculation software Thafts-acoustic; S2: Define the excitation points, the set of all excitation points, and name and number each excitation point; S3: Randomly apply a set of normal white noise sweep excitation forces with different amplitudes; S4: Calculate the vibration and acoustic radiation under this set of excitation forces based on the ship's three-dimensional acoustic elasticity calculation software Thafts-acoustic; S5: Read the vibration acceleration amplitude spectrum of each excitation point corresponding to the measuring point respectively; S6: Read the radiation noise results of the entire ship and output them to the file sound source level-operating condition; S7: Repeat the operations of steps S3-S6 to calculate the results under different working conditions until the number of calculated working conditions is equal to the set number of working conditions M; S8: The output acceleration amplitude spectra and radiated noise source level results of M working conditions are combined into a model training set; S9: Define the number of samples, side points and frequency points used for BP neural network training, read the results of the training set, and process the data to generate the input and output matrices for BP neural network training; S10: Define the parameters for BP neural network training and carry out training of real-time intelligent forecasting model; S11: Based on the trained real-time intelligent prediction model and the online monitored vibration acceleration amplitude spectrum data, the real-time intelligent prediction of ship underwater radiated noise based on virtual and real data fusion is realized.

2. The method for real-time intelligent prediction of underwater ship radiated noise based on virtual and real data fusion according to claim 1 is characterized in that: In S2, the connection point between the vibration device and the base is preferably used as the excitation point, and each excitation point is named A1 to AN, where 1 to N corresponds to the measurement point number.

3. The method for real-time intelligent prediction of underwater ship radiated noise based on virtual and real data fusion according to claim 1 is characterized in that: In S5, the vibration acceleration amplitude spectrum of the corresponding measuring point of each excitation point A1~AN is read respectively and output to the A1_working condition J.txt~AN_working condition J.txt files respectively, where J is the working condition number and the initial number is 1.

4. The method for real-time intelligent prediction of underwater ship radiated noise based on virtual and real data fusion according to claim 1 is characterized in that: In S6, output to the file Sound Source Level-Working Condition J.

5. The method for real-time intelligent prediction of underwater ship radiated noise based on virtual and real data fusion according to claim 1 is characterized in that: A forward excitation-response correlation method is used: a swept-frequency excitation force is applied at the selected measuring point, and the radiated noise under the corresponding working condition is calculated using the acoustoelasticity software Thafts-acoustic. The vibration acceleration response at the measuring point is simultaneously collected, and an intelligent mapping relationship between acceleration and radiated noise is constructed based on this. This avoids the inverse solution of the frequency response function and achieves higher calculation accuracy.

6. The method for real-time intelligent prediction of underwater ship radiated noise based on virtual and real data fusion according to claim 1 is characterized in that: By randomly adjusting the excitation force amplitude at the measuring point in the acoustic-elasticity software to generate different sample data sets, there is no need to repeat the modal analysis, fluid-solid contact boundary condition calculation, wet surface source strength and generalized hydrodynamic coefficient calculation in the acoustic-elasticity model calculation process. This greatly improves the sample generation efficiency and the convergence speed of intelligent model training.

Images