Noise Source Separation And Quantification Based On Partial Least Squares Regression

Submarine is an important underwater force of navy,acoustic stealth performance is one of its most basic performance indicators.However,the submarine has the potential to be exposed by submarine noise under the enemy’s firepower.Therefore,the identification,control and separation of submarine noise source are especially important,it is not only to optimize the structure of the submarine,but also to reduce the submarine noise effectively and enhance the hidden nature and combat capability of submarine.In view of the serious coupling among vibration and noise sources of the submarine,the common method is hard to segregate and quantify effectively.Partial least squares(PLS)regression is a multivariate statistical analysis method of ripe theory and wide application.Both PLS regression and PCA need to extract the main constituent which represents the characteristic information of the data variation,but PCA only extract information from independent variable,and PLS regression also extract information from dependent variable at the same time.On condition that multiple correlations among independent variable,PLS regression could better sort out to deal with this type of regression problems.Therefore,PLS regression fit for research on vibration and noise source separation and quantification of submarine cabin.In this paper,by means of the establishment of multi-input single-output and multi-input multi-output model,the physical processes of the internal electromechanical equipment vibration exciting the shell vibration to arise the formation of radiated noise is simulated.Respectively,the independent signal and coupling signal are simulated,firstly,analyze the power spectrum of the input signal and output signal,and extract energy of characteristic frequency spectrum from the input signal and output signal in each layer,then take the extracted energy as data observation sample matrix of PLS-regression model.According to PLS-regression,the PLS-regression equation of the output signal on the input signal is obtained,and then finally the leading noise source,the contribution proportion of the input signal and the result of separation and quantification are obtained.It provides the basis and reference for separation and quantification of vibration and noise source of submarine cabin model test.In this paper,the test of vibration and noise source separation and quantification of submarine cabin model is carried out.By processing the test data,the contribution proportion of vibration and noise sources is obtained,and it provides the reference for controlling the noise effectively.By means of power spectrum analysis,time frequency analysis and coherence analysis of the excitation equipment,pressure shell and radiation field measured signal,the spectrum characteristics and energy characteristics,as well as the coherence characteristics can be obtained.By means of the establishment of multi-input single-output and multi-input multi-output system of submarine cabin model,the separation and quantification of vibration and noise source is finished by taking advantage of PLS regression Hierarchical partial least square(Hi-PLS)regression.The result of separation and quantification is verified by Operation by parts.Submarine cabin model test indicates that PLS-regression can better separate and quantify the space coupling vibration and noise source.PLS-regression model is more advanced,the calculation results are more reliable,and its model interpretability is also stronger in the actual system.When there are lots of variables,Hi-PLS regression are more reasonable compared with PLS-regression.