The Study On The Separation Of Engine Noise Signals Based On Blind Source Seperation

The NVH of the vehicle directly affects the comfort of the vehicle,and with the continuous improvement of the performance requirements of automotive NVH,the vibration and noise of the engine has gradually become a hot issue and an important direction for the research of domestic engine disciplines.How to identify the main noise source of the engine accurately and quickly is an important prerequisite for controlling engine noise and is one of the important research contents to improve the comfort of the vehicle.The engine noise signal is a mixed signal generated by multiple excitation sources,and traditional mathematical tools are difficult to achieve accurate analysis of the specified excitation source.Blind source separation algorithm has a great research space in identifying mechanical vibration,fault diagnosis and noise processing,and provides a richer means for engine noise signal separation research.In this paper,based on the characteristics of engine noise signals,modern testing techniques and signal analysis methods such as noise testing technology,wavelet analysis theory and blind source separation algorithm are used to separate and identify the main noise source of the engine from complex noise signals.The main research contents are as follows:Three different sound signals are selected as the original signals,and the original signals are separately mixed on Matlab software.Four blind source separation algorithms,FastICA,SOBI,JADE and CuBICA,are analyzed and compared.The blind source separation simulation analysis of mixed signals is used to verify whether the four blind source separation algorithms have good separation performance.Three different sound signals are also selected for convolution mixing,and the simulation analysis is used to verify the feasibility of the above four algorithms.The convolutional mixed signal is separated by the instantaneous blind source separation algorithm,and compared with the effect of the convolution algorithm,the limitation of the instantaneous blind source separation algorithm in separating the real engine noise signal is verified.The separation performance of the four algorithms is compared with theseparation evaluation index,and the optimal algorithm is selected to analyze the real engine noise signal.Taking a four-cylinder four-stroke engine as the research object,the CuBICA algorithm based on convolution mixing is used to separate the three engine noise signals collected to obtain a series of independent components.In order to further identify the corresponding relationship between the separated independent components and the engine noise source,the Fourier transform and wavelet transform techniques are used to analyze the characteristics of each component.Combined with the results of time-frequency analysis and the mechanism of engine radiation noise,the excitation sources corresponding to the independent components are determined.