Real-time Prediction Method Of Far-Field Acoustic Radiation For Complex Structure Based On Operational Data

Far-field acoustic radiation prediction is a hot issue in the acoustic fields.Because of noise source moving,inconvenience of placing far-field sensors and other reasons,it is inconvenient to directly measure far-field sound pressure.Thus,far-field sound pressure can only be indirectly predicted through near-field sensors.Due to practical issues of the complicated structure,boundary condition and sensor deployment,existing methods have some limitations for acoustic radiation prediction in complex structures.This paper brings the method of operational transfer path analysis into the field of acoustic radiation prediction,and focuses on the research and improvement of the acoustic radiation prediction model based on operational data.First,an acoustic radiation prediction model based on operational data is established.The pseudoinverse transfer function between near-field sensor and far-field sensor is derived by using the linear transfer function of the near-field sensor,the far-field sensor and the excitation source.Thus,the spectrum of the far-field sensor can be estimated with the near-field sensor data and the pseudoinverse transfer function.Theoretical analysis shows that prediction accuracy of the model is influenced by operational interference,matrix condition number and signal to noise ratio.On this basis,a simulation platform for acoustic radiation prediction based on Matlab GUI is constructed.Then,due to prediction error caused by operational interference,a model optimization method based on operational recognition is proposed.The main idea of this method is to achieve feature optimization and operational recognition by combining PCA and K-means methods,and replace the pseudoinverse transfer function of all operational conditions with that of each operational condition.Experimental result shows that this method can effectively reduce the acoustic radiation prediction error.Finally,due to the redundancy of the near-field sensors,a model optimization method based on sensor location is proposed.The main idea of this method is to find sensor combination with the weaker correlation using the correlation analysis method,and select sensor combination with higher classification weight using random forest algorithm.Experimental result shows that this method can reduce the number of near-field sensors without reducing the accuracy of acoustic radiation prediction at the same time.The research of this paper have a certain guiding significance for the problem of far-field acoustic radiation prediction in complex structures.