Research On Low Frequency Acoustic Radiation Prediction Of Underwater Complex Elastic Structure

Real-time monitoring and prediction of vibro-acoustic radiation levels of underwater vehicles and their equipment has always been a concern.At present,in order to grasp the realtime acoustic state of the underwater vehicles,especially to correctly predict the low-frequency radiation noise level that is difficult to control,real-time monitoring of the vibration of the surface structure of the underwater vehicles is usually adopted in engineering.Then,the vibration data is combined with an accurate and efficient transfer prediction model for sound field prediction.Therefore,how to obtain a more accurate and efficient transfer model based on vibration to sound field is particularly important.This paper firstly optimizes and improves the vibro-acoustic transfer function method based on structural vibration velocity commonly used in engineering,proposes a modeling method for subdivision,and analyzes the influence of complex excitation force on the prediction of acoustic radiation.Finally,the acoustic radiation prediction of underwater large scale complex elastic structures is realized.Firstly,based on the structural vibration and the sound field showing low-frequency global distribution and high-frequency local distribution characteristics,an acoustic radiation prediction method combining the whole and subdivision of large-scale elastic structures is proposed,and it is simulated and verified.Research shows that this method can improve the modeling efficiency on the premise of ensuring accuracy.Then,for the simplified point,line,surface excitation,rigid installation and elastic installation mass load excitation sources,and mass load excitation sources with different mass sizes,the influence of these different excitation source forms on the acoustic radiation prediction is simulated and analyzed.The research shows that at lower frequencies,due to the large scale of the structural model and the close excitation positions of different excitation sources,different excitation sources excite similar vibration modes.Therefore,the structural acoustic radiation efficiency corresponding to different working conditions is basically the same.However,when different forms of excitation force are applied to the elastic structure at higher frequencies,the vibration mode distribution of the excited structure is different,which makes the acoustic radiation efficiency of the structure obtained by simulation different.Therefore,when carrying out the acoustic radiation prediction based on the vibro-acoustic transfer function method,it is necessary to model as accurately and finely as possible according to the excitation form of the actual structure,so as to improve the prediction accuracy.At the same time,the study also shows that in the same action area of the same cabin,when the position of the mass load excitation source changes,the excited vibration modes are relatively similar,so that the influence of the excitation source position distribution on the acoustic radiation efficiency is small.Finally,it is also found that when the vibro-acoustic transfer function method is used to predict the acoustic radiation of the multi-source excitation conditions,due to the complexity and variability of the multi-source excitation combination in the actual conditions,it is easy to have a mismatch between the simulated acoustic radiation efficiency and the actual working conditions,resulting in a decrease in the accuracy of acoustic radiation prediction.At present,in order to solve this problem and realize high-precision prediction of multiexcitation source conditions,a vibration and noise prediction method of underwater structures based on GA-LM-BP neural network is also proposed in this paper.This method takes the vibration data of the underwater structure surface as the network input,and takes the sound pressure or radiated acoustic power of the sound field assessment point as the network output.And this method realizes high-precision transfer prediction from structural vibration to sound field.Firstly,the GA-LM-BP neural network is constructed,and based on the algorithm,the two vibration and noise prediction models of vibration-to-sound field sound pressure and vibration-to-radiated sound power under the condition of multiple excitation sources are numerically simulated.The research shows that the method has high prediction accuracy when dealing with the sound field prediction problem,which verifies the feasibility of the method.The influence of GA algorithm,LM algorithm and transfer function on network prediction results is analyzed.The GA algorithm and the LM algorithm are respectively used to optimize the initialization of the network weight threshold and the dynamic adjustment of the network weight threshold during the training process,meanwhile,the purelin and square functions are used to optimize the network training transfer function.Studies have shown that the optimized network requires less sample size for training,the network converges faster,and the prediction accuracy is higher.The adaptability of neural network to sample data with different signal-tonoise ratios is analyzed.The research shows that even if the sample’s signal-to-noise ratio is only 14 d B,the qualified rate of model prediction has reached more than 90%,indicating that the method has good adaptability.Finally,this paper analyzes the similarities and differences of the two forecasting methods from the theoretical aspect,and summarizes the applicable conditions of the methods.That is,the vibro-acoustic transfer function method needs to master the parameters of the entity structure and the information of the actual working conditions,and the key lies in the accuracy of the finite element modeling and simulation.And the neural network method requires sufficient representative and highly reliable data samples,and the key lies in the accurate processing of the data samples.From the comparison of prediction accuracy under various working conditions,it is shown that the two methods can achieve high-precision acoustic radiation prediction for single excitation source working conditions after satisfying their respective applicable conditions.At the same time,it also shows that in the face of multiple excitation source conditions,compared with the vibro-acoustic transfer function method,the neural network method can obtain higher prediction accuracy and better performance after satisfying certain applicable conditions.